Output device, learning method, and program

Abstract

To provide a technique which can alleviate a burden for design in designing a building.SOLUTION: According to one aspect of the present invention, an output device includes a contribution degree estimation unit which executes a learned model obtained by a method of machine learning using a first learning data set which includes a plurality of shape parameters representing shapes of buildings as learning data and nature parameters representing natures of the buildings as teacher data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an output device, a learning method, and a program.

Background Art

[0002] For the construction and repair of buildings such as roofs, there are cases where a designer who considers the shape of the building and a designer who designs a building that satisfies physical laws work together on the design work. Since the proposals of the designer often cover a wide range, when the designer and the designer work together, the designer and the designer proceed with the design work while confirming the design results.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, the designs proposed by the designer often cover a wide range. Therefore, when confirming the design results in accordance with the proposals of the designer, it takes time and places a heavy burden on the designer to perform analysis by making all of them targets of numerical calculation. Further, this is not limited to the case where the designer and the designer jointly design a building, but is the same even when the designer is solely responsible for design and design.

[0005] In view of the above circumstances, an object of the present invention is to provide a technique capable of reducing the burden required for design in the design of a building.

Means for Solving the Problems

[0006] One aspect of the present invention is an output device comprising a contribution estimation unit that executes a trained model obtained by a machine learning method using a first training dataset which includes a plurality of shape parameters indicating the shape of a building as training data and property parameters indicating the properties of the building as training data. [Effects of the Invention]

[0007] This invention makes it possible to provide a technology that can reduce the burden required for designing buildings. [Brief explanation of the drawing]

[0008] [Figure 1] An explanatory diagram illustrating the outline of the design support system 100 of the first embodiment. [Figure 2] A diagram showing an example of the hardware configuration of the relational information acquisition device 1 in the first embodiment. [Figure 3] A diagram showing an example of the functional configuration of the control unit 11 in the first embodiment. [Figure 4] A flowchart showing an example of the processing flow executed by the relational information acquisition device 1 in the first embodiment. [Figure 5] A diagram showing an example of the hardware configuration of the support device 2 in the first embodiment. [Figure 6] A diagram showing an example of the functional configuration of the control unit 21 in the first embodiment. [Figure 7] A flowchart showing an example of the processing flow performed by the support device 2 in the first embodiment. [Figure 8] Figure 1 shows an example of the output result of the output unit 20 in the first embodiment. [Figure 9] An explanatory diagram illustrating the outline of the design support system 100a of the second embodiment. [Figure 10] This figure shows an example of the hardware configuration of the relational information acquisition device 1a in the second embodiment. [Figure 11] A diagram showing an example of the functional configuration of the control unit 11a in the second embodiment. [Figure 12]A flowchart showing an example of the process flow executed by the relationship information acquisition device 1a in the second embodiment. [Figure 13] A diagram showing an example of the hardware configuration of the support device 2a in the second embodiment. [Figure 14] A diagram showing an example of the functional configuration of the control unit 21a in the second embodiment. [Figure 15] A flowchart showing an example of the process flow executed by the support device 2a in the second embodiment. [Figure 16] An explanatory diagram explaining the outline of the design support system 100b in the third embodiment. [Figure 17] A diagram showing an example of the hardware configuration of the relationship information acquisition device 1b in the third embodiment. [Figure 18] A diagram showing an example of the functional configuration of the control unit 11b in the third embodiment. [Figure 19] A flowchart showing an example of the process flow executed by the relationship information acquisition device 1b in the third embodiment. [Figure 20] A diagram showing an example of the hardware configuration of the support device 2b in the third embodiment. [Figure 21] A diagram showing an example of the functional configuration of the control unit 21b in the third embodiment. [Figure 22] A flowchart showing an example of the process flow executed by the support device 2b in the third embodiment. [Figure 23] The first flowchart showing the first example of the input assistance process flow executed by the output control unit 150 in the modification example. [Figure 24] The second flowchart showing the first example of the input assistance process flow executed by the output control unit 150 in the modification example. [Figure 25] A flowchart showing the second example of the input assistance process flow executed by the output control unit 150 in the modification example. [Figure 26] A diagram showing an example of the truss structure in the modification example. [Figure 27] A diagram showing an example of the beam string structure in the modification example. [Figure 28]An explanatory diagram illustrating the upper chord, lower chord, diagonal members, bracing members, and nodes in a modified example. [Figure 29] Figure 1 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 30] Figure 2 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 31] Figure 3 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 32] Figure 4 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 33] Figure 5 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 34] Figure 6 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 35] Figure 7 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 36] Figure 8 shows an example of the results of a performance evaluation experiment in a modified example. [Figure 37] Figure 9 shows an example of the results of a performance evaluation experiment in a modified example. [Modes for carrying out the invention]

[0009] (First Embodiment) Figure 1 is an explanatory diagram illustrating the outline of the design support system 100 of the first embodiment. The design support system 100 assists the user in designing the design target. The design target is a building. The user is, for example, a designer who designs the design target or a designer who designs the shape of the design target. For the sake of simplicity, the related information acquisition device 1 will be described below using the case where the design target is a roof as an example. In particular, the related information acquisition device 1 will be described using the case where the design target includes a spatial structure as an example. A spatial structure is a structure that has columns only on the outer perimeter, with no columns in the middle, and covers a large space. An example of a design target that includes a spatial structure is a roof with an arch structure, truss structure, or tensioned beam structure.

[0010] Specifically, the design support system 100 accepts input of values ​​(hereinafter referred to as "shape parameter values") of variables (hereinafter referred to as "shape parameters") that indicate the shape of a building. More specifically, values ​​for multiple shape parameters are input to the related information acquisition device 1.

[0011] Shape parameters include, for example, the depth of a building. The definition of depth is the maximum distance between the upper chord and the lower chord. Shape parameters include, for example, the rise of a building. The definition of rise is the distance between the column base position and the lower chord or upper chord. Shape parameters include, for example, the number of main beams in a building. Shape parameters may also include the number of upper chords or the number of lower chords in a building. Shape parameters include, for example, the number of bracing members in a building. Shape parameters may also include, for example, the number of diagonal members in a building. Shape parameters may also include, for example, the number of tensioned beams in a building. Shape parameters may also include, for example, the number of trusses in a building. Multiple shape parameters input to the relational information acquisition device 1 include at least one of the following for a building: depth, rise, number of main beams, number of upper chords, number of lower chords, number of bracing members, number of diagonal members, number of tensioned beams, and number of trusses.

[0012] The design support system 100 estimates the parameter contribution of a shape parameter based on the input values ​​of one or more shape parameters relating to the design object. Hereinafter, the set of input values ​​of one or more shape parameters relating to the design object is referred to as the first analysis target data. Parameter contribution is information that indicates the degree to which changes in the values ​​of shape parameters have an influence on changes in the properties of the building. More specifically, parameter contribution is information that indicates the degree to which changes in shape parameter values ​​have an influence on changes in the values ​​of variables that indicate the properties of the building (hereinafter referred to as "property parameters").

[0013] The properties of a building are properties related to the building, such as reaction force. The properties of a building are properties related to the building, such as deformation. The properties of a building are properties related to the building, such as stress. The properties of a building are properties related to the building, such as strain energy. The properties of a building are properties related to the building, such as weight. The related information acquisition device 1 acquires the parameter contribution for at least one of the following related to the building: reaction force, deformation, stress, strain energy, and weight. Therefore, the property parameters indicate at least one of the following related to the building: reaction force, deformation, stress, strain energy, and weight.

[0014] The design support system 100 comprises a related information acquisition device 1 and a support device 2.

[0015] The relational information acquisition device 1 acquires information showing the relationship between shape parameters and property parameters (hereinafter referred to as "first relational information"). The relational information acquisition device 1 acquires the first relational information by, for example, a machine learning method. That is, the relational information acquisition device 1 acquires a mathematical model showing the relationship between shape parameters and property parameters by a machine learning method. The design support system 100 will be described below using the case in which the relational information acquisition device 1 acquires the first relational information by a machine learning method as an example.

[0016] When the relational information acquisition device 1 acquires first relational information using a machine learning method, the relational information acquisition device 1 updates a learning model (hereinafter referred to as the "first learning model") that shows the relationship between shape parameters and property parameters through learning. The relational information acquisition device 1 continues learning until a predetermined termination condition (hereinafter referred to as the "learning termination condition") is met. The first learning model at the time the learning termination condition is met is used for the analysis of the design target. That is, the learned first learning model is used for the analysis of the design target. The learning termination condition may be, for example, a condition that a predetermined number of learning sessions have been performed, or it may be, for example, a condition that the change in the learning model due to learning is less than a predetermined change.

[0017] The relational information acquisition device 1 performs training on the first training dataset when training the first learning model. The first training dataset is a collection of one or more first unit training sets. Each first unit training set is a pair of shape parameters and property parameters. More specifically, when training the first learning model, the relational information acquisition device 1 performs training using shape parameters as training data and property parameters as training data. That is, when training the first learning model, the relational information acquisition device 1 performs training using shape parameters as explanatory variables and property parameters as the target variable.

[0018] The training data may be measurement results of already constructed buildings, or it may be values ​​obtained based on training data according to a predetermined rule for obtaining training data based on the values ​​of the training data. The predetermined rule for obtaining training data based on the values ​​of the training data (hereinafter referred to as the "training data acquisition rule") may include, for example, structural analysis. The training data acquisition rule may include, for example, cross-sectional analysis. The training data acquisition rule may also include, for example, static analysis. The training data acquisition rule may also include, for example, dynamic analysis.

[0019] The training data corresponding to the learning data may be input into the relational information acquisition device 1 by an external device or user, or it may be calculated by the relational information acquisition device 1 based on the input learning data in accordance with the training data acquisition rules.

[0020] For the sake of simplicity, the design support system 100 will be described below using the example of a case where the relational information acquisition device 1 acquires corresponding teacher data based on the input learning data. Hereinafter, the process by which the relational information acquisition device 1 acquires teacher data based on the learning data in accordance with the teacher data acquisition rules will be referred to as the first teacher data acquisition process.

[0021] Furthermore, if training data is input to the relational information acquisition device 1 by an external device or user, the relational information acquisition device 1 only needs to use the input training data for learning and does not need to acquire the training data by executing the first training data acquisition process. In other words, if training data is input to the relational information acquisition device 1 by an external device or user, the relational information acquisition device 1 only needs to use the input training data for learning and does not need to execute the first training data acquisition process.

[0022] Furthermore, training a learning model means updating the learning model. Updating a learning model means updating the parameter values ​​of the circuit that represents the learning model. A learning model can be represented, for example, by a neural network. A neural network is a circuit such as an electronic circuit, electrical circuit, optical circuit, or integrated circuit, and is an example of a circuit that represents a learning model. When a learning model is updated through learning, it means that the parameter values ​​of the circuit that represents the learning model are updated. If the circuit that represents the learning model is a neural network, the parameters of the neural network are suitably adjusted based on the value of the objective function (i.e., the loss).

[0023] Support device 2 accepts input of multiple shape parameter values ​​(i.e., first analysis target data) for the design object. Based on the input first analysis target data, support device 2 uses first relational information, such as a learned first learning model acquired by relational information acquisition device 1, to estimate the parameter contribution for each shape parameter for some or all of the shape parameters included in the first analysis target data with respect to the design object. Hereinafter, the process of estimating the parameter contribution using the first relational information based on the input shape parameter values ​​will be referred to as the first contribution estimation process.

[0024] This section explains a specific example of how parameter contributions are estimated in the first contribution estimation process, using the case where the first relational information is a pre-trained first learning model.

[0025] In the first contribution estimation process, the parameter contributions are estimated using a method that estimates the contribution of each explanatory variable to the estimation results of a trained model. An example of such a method is gradient boosting. When gradient boosting is used, the parameter contribution is specifically the importance in gradient boosting. Alternatively, SHAP (SHapley Additive exPlanations) using a regression analysis model may also be used. More specifically, SHAP is used when estimating parameter contributions from trained models of algorithms other than regression trees, such as gradient boosting, decision trees, and random forests. An example of a trained model of an algorithm other than a regression tree is a neural network.

[0026] Support device 2 includes an output control unit 200. The output control unit 200 controls the operation of a controlled object that performs output such as display and sound output, and causes the controlled object to output the parameter contribution obtained by executing the first contribution estimation process. Details of the output control unit 200 will be described later. Specifically, the controlled object of the output control unit 200 is the output unit 20, which will be described in detail later.

[0027] Figure 2 shows an example of the hardware configuration of the relational information acquisition device 1 in the first embodiment. The relational information acquisition device 1 includes a control unit 11 which has a processor 91 such as a CPU (Central Processing Unit) and memory 92 connected by a bus, and executes a program. The relational information acquisition device 1 functions as a device comprising the control unit 11, communication unit 12, input unit 13, storage unit 14, and output unit 15 through the execution of the program. The control unit 11 may also include a GPU (Graphics Processing Unit).

[0028] More specifically, the relational information acquisition device 1 has the processor 91 read a program stored in the storage unit 14 and store the read program in the memory 92. By having the processor 91 execute the program stored in the memory 92, the relational information acquisition device 1 functions as a device comprising a control unit 11, a communication unit 12, an input unit 13, a storage unit 14, and an output unit 15.

[0029] The control unit 11 controls the operation of various functional units of the relational information acquisition device 1. For example, the control unit 11 performs the first training data acquisition process. For example, the control unit 11 acquires first relational information. The control unit 11 performs training of the first learning model. For example, the control unit 11 records various information generated by the training of the first learning model in the storage unit 14.

[0030] The communication unit 12 is configured to include a communication interface for connecting the relational information acquisition device 1 to an external device. The communication unit 12 communicates with the external device via wired or wireless connection. The external device is, for example, a support device 2. The communication unit 12 transmits the learned first learning model to the support device 2, for example, through communication with the support device 2. The external device may also be, for example, a device that transmits the learning data. The communication unit 12 may acquire the learning data by communicating with the device that transmits the learning data.

[0031] The input unit 13 includes input devices such as a mouse, keyboard, and touch panel. The input unit 13 may also be configured as an interface for connecting these input devices to the relational information acquisition device 1. The input unit 13 receives various types of information input to the relational information acquisition device 1. For example, the input unit 13 receives an instruction to start learning. For example, the input unit 13 receives learning data.

[0032] The storage unit 14 is configured using a computer-readable storage medium such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 14 stores various information related to the relational information acquisition device 1. The storage unit 14 stores information input via, for example, the communication unit 12 or the input unit 13. The storage unit 14 stores various information generated by, for example, the execution of processing by the control unit 11. If the first relational information is a first learning model, the storage unit 14 stores the first learning model in advance before the start of the learning process.

[0033] Furthermore, the learning data does not necessarily have to be input only to the communication unit 12, nor does it have to be input only to the input unit 13. Part of the learning data may be input to the communication unit 12 and the other part to the input unit 13. Also, the learning data does not necessarily have to be obtained from the communication unit 12 or the input unit 13; at least some of the shape parameter values ​​that may be used for learning may be stored in the storage unit 14 in advance as learning data.

[0034] The output unit 15 outputs various types of information. The output unit 15 is comprised of a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display. The output unit 15 may be configured as an interface for connecting these display devices to the related information acquisition device 1. The output unit 15 displays, for example, the information input to the input unit 13. The output unit 15 may also display, for example, the result of processing performed by the control unit 11.

[0035] The output unit 15 may include, for example, a device that outputs sound, such as a speaker. The output unit 15 may be configured as an interface for connecting these sound output devices to the related information acquisition device 1. The output unit 15 may, for example, emit sound indicating the information input to the input unit 13. The output unit 15 may also emit sound indicating the result of processing performed by the control unit 11.

[0036] Figure 3 shows an example of the functional configuration of the control unit 11 in the first embodiment. The control unit 11 includes a learning dataset acquisition unit 110, a relational information acquisition unit 120, a storage control unit 130, an input control unit 140, an output control unit 150, and a communication control unit 160.

[0037] The training dataset acquisition unit 110 acquires the first training dataset. The training dataset acquisition unit 110 acquires the training data input to the communication unit 12, for example, when training data is input to the communication unit 12. The training dataset acquisition unit 110 acquires the training data input to the input unit 13, for example, when training data is input to the input unit 13.

[0038] The learning dataset acquisition unit 110 reads out information from the storage unit 14 if, for example, an external device or user of the relational information acquisition device 1 instructs that a portion of the learning data already stored in the storage unit 14 be used for learning. Hereinafter, the instruction to use a portion of the learning data stored in the storage unit 14 for learning will be referred to as the first read instruction. The first read instruction is input to the communication unit 12 or the input unit 13.

[0039] The training dataset acquisition unit 110 acquires training data for each training data set by executing a first training data acquisition process based on the acquired training data. More specifically, the training dataset acquisition unit 110 acquires training data corresponding to each training data set by executing a first training data acquisition process for each acquired training data set. In this way, the training dataset acquisition unit 110 acquires a set of pairs of training data and corresponding training data (i.e., a first unit training set) (i.e., a first training dataset).

[0040] The relational information acquisition unit 120 acquires first relational information using the first training dataset acquired by the training dataset acquisition unit 110. For example, the relational information acquisition unit 120 uses the first training dataset acquired by the training dataset acquisition unit 110 to train the first training model until the training completion condition is met. In this way, the relational information acquisition unit 120 acquires the trained first training model. The trained first training model is an example of first relational information.

[0041] The memory control unit 130 records various information in the storage unit 14. The memory control unit 130 records various information generated by the operation of the control unit 11, for example, in the storage unit 14. The information generated by the operation of the control unit 11 is, for example, the updated first learning model updated by learning. The memory control unit 130 records, for example, the learned first learning model in the storage unit 14.

[0042] The input control unit 140 controls the operation of the input unit 13. The output control unit 150 controls the operation of the output unit 15. The communication control unit 160 controls the operation of the communication unit 12.

[0043] Figure 4 is a flowchart showing an example of the processing flow executed by the relational information acquisition device 1 in the first embodiment. The learning dataset acquisition unit 110 acquires the first learning dataset (step S101). The learning data of the first learning dataset acquired by the learning dataset acquisition unit 110 is, for example, the learning data input to the communication unit 12. The learning data of the first learning dataset acquired by the learning dataset acquisition unit 110 may also be, for example, the learning data input to the input unit 13. The learning data of the first learning dataset acquired by the learning dataset acquisition unit 110 may also be, for example, the learning data read from the storage unit 14 in accordance with the first read instruction.

[0044] Next, the relational information acquisition unit 120 uses one of the first unit learning sets included in the first learning dataset acquired in step S101, which has not yet been used to train the first learning model, to train the first learning model (step S102). Next, the relational information acquisition unit 120 determines whether the learning termination condition has been met (step S103). If the learning termination condition is met (step S103: YES), the memory control unit 130 records the trained first learning model in the memory unit 14 (step S104). Note that the process in step S104 may be replaced by the communication control unit 160 transmitting the trained first learning model to the support device 2 instead of the memory control unit 130 recording the trained first learning model in the memory unit 14. On the other hand, if the learning termination condition is not met (step S103: NO), the process returns to step S102.

[0045] Figure 4 shows an example of the processing flow executed by the relational information acquisition device 1, using the case where the first relational information is a first learned model that has already been learned. Therefore, the processing in step S102 and step S103 is an example of the processing flow for acquiring the first relational information.

[0046] Figure 5 shows an example of the hardware configuration of the support device 2 in the first embodiment. The support device 2 includes a control unit 21 which has a processor 93 such as a CPU (Central Processing Unit) and memory 94 connected by a bus, and executes a program. The support device 2 functions as a device comprising the control unit 21, communication unit 22, input unit 23, storage unit 24 and output unit 20 by executing the program.

[0047] More specifically, the support device 2 reads a program stored in the storage unit 24 by the processor 93 and stores the read program in the memory 94. By the processor 93 executing the program stored in the memory 94, the support device 2 functions as a device comprising a control unit 21, a communication unit 22, an input unit 23, a storage unit 24, and an output unit 20.

[0048] The control unit 21 controls the operation of various functional units of the support device 2. For example, the control unit 21 performs a first contribution estimation process using first relational information such as a first learned model that has been trained. For example, the control unit 21 records various information generated by the execution of the first contribution estimation process in the storage unit 24.

[0049] The communication unit 22 is configured to include a communication interface for connecting the support device 2 to an external device. The communication unit 22 communicates with the external device via wired or wireless means. The external device is, for example, a relational information acquisition device 1. The communication unit 22 acquires a learned first learning model from the relational information acquisition device 1, for example, by communicating with the relational information acquisition device 1. The external device may also be, for example, a device that transmits the first data to be analyzed. The communication unit 22 may acquire the first data to be analyzed by communicating with the device that transmits the first data to be analyzed.

[0050] The input unit 23 includes input devices such as a mouse, keyboard, or touch panel. The input unit 23 may also be configured as an interface for connecting these input devices to the support device 2. The input unit 23 receives various types of information for input to the support device 2. For example, the input unit 23 receives an instruction to start the analysis. Specifically, the start of the analysis means the start of parameter contribution estimation. For example, the input unit 23 receives first analysis target data.

[0051] The storage unit 24 is configured using a computer-readable storage medium such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 24 stores various information related to the support device 2. The storage unit 24 stores information input via, for example, the communication unit 22 or the input unit 23. The storage unit 24 stores various information generated by, for example, the execution of processing by the control unit 21. The storage unit 14 stores the first relational information in advance before the execution of the first contribution estimation process begins. That is, if the first relational information is a learned first learning model, the storage unit 14 stores the learned first learning model in advance before the execution of the first contribution estimation process begins.

[0052] Furthermore, the first data to be analyzed does not necessarily have to be input only to the communication unit 22, nor does it have to be input only to the input unit 23. Part of the first data to be analyzed may be input to the communication unit 22 and the other part to the input unit 23. In addition, the first data to be analyzed does not necessarily have to be obtained from the communication unit 22 or the input unit 23, and at least a portion of the values ​​of the shape parameters that may be used in the first contribution estimation process as the first data to be analyzed may be stored in the storage unit 24 in advance.

[0053] The output unit 20 outputs various types of information. The output unit 20 is comprised of a display device such as a CRT display, liquid crystal display, or organic EL display. The output unit 20 may also be configured as an interface for connecting these display devices to the support device 2. The output unit 20 displays, for example, the information input to the input unit 23. The output unit 20 may also display, for example, the result of processing performed by the control unit 21.

[0054] The output unit 20 may include, for example, a device that outputs sound, such as a speaker. The output unit 20 may be configured as an interface for connecting these sound output devices to the support device 2. The output unit 20 may, for example, emit sound indicating the information input to the input unit 23. The output unit 20 may also emit sound indicating the result of processing performed by, for example, the control unit 21.

[0055] Figure 6 shows an example of the functional configuration of the control unit 21 in the first embodiment. The control unit 21 includes an analysis target data acquisition unit 210, a contribution estimation unit 220, a storage control unit 230, an input control unit 240, an output control unit 200, and a communication control unit 260.

[0056] The data acquisition unit 210 acquires the first data to be analyzed. If the first data to be analyzed is input to the communication unit 22, for example, the data acquisition unit 210 acquires the first data to be analyzed input to the communication unit 22. If the first data to be analyzed is input to the input unit 23, for example, the data acquisition unit 210 acquires the first data to be analyzed input to the input unit 23.

[0057] If an external device or user of the support device 2 instructs the data acquisition unit 210 to use a portion of the first data to be analyzed, which is already stored in the storage unit 24, for the first contribution estimation process, the unit reads information from the storage unit 24. Hereinafter, the instruction to use a portion of the first data to be analyzed, which is already stored in the storage unit 24, for the first contribution estimation process will be referred to as the second read instruction. The second read instruction is input to the communication unit 22 or the input unit 23.

[0058] The contribution estimation unit 220 performs a first contribution estimation process on the first data to be analyzed acquired by the data acquisition unit 210. Specifically, in the first contribution estimation process, the contribution estimation unit 220 first performs a data application process. The data application process is a process that estimates the values ​​of property parameters for some or all of the shape parameters included in the first data to be analyzed, based on the first data to be analyzed, using first relational information such as a trained first learning model.

[0059] For example, the process of applying data to be analyzed is the process of obtaining the values ​​of property parameters by executing the first trained model on the first data to be analyzed. Executing the first trained model on the first data to be analyzed means inputting the first data to be analyzed into the first trained model and then executing the first trained model.

[0060] The contribution estimation unit 220 then performs a contribution acquisition process in the first contribution estimation process. The contribution acquisition process is a process that acquires the parameter contribution for each of the multiple shape parameters contained in the first data to be analyzed, based on the results of the data to be analyzed application process. The contribution acquisition process is a process that acquires the parameter contribution using the gradient boosting method, for example, based on the results of the data to be analyzed application process. The feature importance in the gradient boosting method corresponds to the parameter contribution. Therefore, the parameter contribution is calculated based on the importance of the feature.

[0061] The contribution acquisition process may, for example, be a process that acquires parameter contributions using SHAP with a regression analysis model based on the results of the data to be analyzed. The feature importance in SHAP corresponds to the parameter contribution. Therefore, the parameter contribution is calculated based on the importance of the features.

[0062] Thus, in the contribution retrieval process, parameter contributions are obtained, for example, based on the importance of the features. In the contribution retrieval process, the importance of the features is obtained for each shape parameter, for some or all of the shape parameters included in the first analysis target data.

[0063] Thus, the first contribution estimation process includes the process of applying the data to be analyzed and the process of obtaining the contribution. Furthermore, in the first contribution estimation process, the contribution acquisition process is executed after the process of applying the data to be analyzed.

[0064] The memory control unit 230 records various information in the storage unit 24. The memory control unit 230 records various information generated by the operation of the control unit 21, for example, in the storage unit 24. The information generated by the operation of the control unit 21 is, for example, the information generated in the first contribution estimation process. The memory control unit 230 records the parameter contributions estimated by the execution of the first contribution estimation process, for example, in the storage unit 24.

[0065] The input control unit 240 controls the operation of the input unit 23. The output control unit 200 controls the operation of the output unit 20, which is the object of control. The communication control unit 260 controls the operation of the communication unit 22.

[0066] Figure 7 is a flowchart showing an example of the processing flow performed by the support device 2 in the first embodiment. The data acquisition unit 210 acquires the first data to be analyzed (step S201). The first data to be analyzed acquired by the data acquisition unit 210 is, for example, the first data to be analyzed that has been input to the communication unit 22. The first data to be analyzed acquired by the data acquisition unit 210 may also be, for example, the first data to be analyzed that has been input to the input unit 23. The first data to be analyzed acquired by the data acquisition unit 210 may also be, for example, the first data to be analyzed that has been read from the storage unit 24 in accordance with the second read instruction.

[0067] Next, the contribution estimation unit 220 performs the first contribution estimation process (step S202). The execution of the first contribution estimation process estimates the values ​​of property parameters for some or all of the shape parameters included in the first analysis target data acquired in step S201. Next, the output control unit 200 causes the output unit 20 to output the parameter contributions estimated in step S202 (step S203).

[0068] Figure 8 is the first figure showing an example of the output result of the output unit 20 in the first embodiment. In the example of Figure 8, the output unit 20 was a display device. The vertical axis in Figure 8 represents the shape parameters. "depth" indicates the depth. "divide" indicates the number of upper chord members in the case of a truss or tensioned beam, and the number of main beams in the case of an arch structure. In the following explanation, "divide" will refer to the upper chord member. "radius" represents the radius of the tension ring. "rise" indicates the rise. "divide2" indicates the number of divisions or bundle members of the adjacent upper chord member. "type" indicates whether it is a truss or a tensioned beam. Indicating whether it is a truss or a tensioned beam means indicating whether or not diagonal members are included.

[0069] The horizontal axis of Figure 8 represents the degree of influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in vertical deformation Z. The degree of influence that changes in each shape parameter have on the change in vertical deformation Z is a type of parameter contribution. The example in Figure 8 shows that changes in depth have the greatest influence on the change in vertical deformation Z. The example in Figure 8 also shows that the influence of changes in the number of main beams on the change in vertical deformation Z is the second greatest after depth.

[0070] Note that the diagram showing the parameter contributions in Figure 8 is just one example. The horizontal axis of the diagram showing the parameter contributions may represent the strength of the influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in the sum of strain energy, which is composed of axial strain energy, shear strain energy, and bending strain energy. The horizontal axis of the diagram showing the parameter contributions may also represent the strength of the influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in the magnitude of the horizontal deformation X.

[0071] The horizontal axis of a figure representing parameter contribution may represent the strength of the influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in the magnitude of the horizontal deformation Y. Alternatively, the horizontal axis of a figure representing parameter contribution may represent the strength of the influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in the magnitude of the reaction force at the base of the column. Note that the values ​​of each shape parameter on the vertical axis are just examples, and it is not necessary to show all of "depth", "divide", "radius", "rise", "divide2", and "type". Furthermore, shape parameters other than "depth", "divide", "radius", "rise", "divide2", and "type" may also be shown on the vertical axis.

[0072] The design support system 100 of the first embodiment, configured in this way, can acquire first relational information and obtain the parameter contribution when the shape parameter values ​​are changed for a design target based on the first relational information. Therefore, a designer using the design support system 100 does not necessarily need to analyze all candidate designs when designing. For example, if the designer sees that the estimation results of the design support system 100 indicate that a change in shape parameters is unlikely to yield a favorable result, the designer can choose not to consider it further. As a result, the design support system 100 can reduce the burden required for designing buildings.

[0073] Furthermore, the first relational information does not necessarily have to be information obtained by machine learning methods such as a pre-trained first learning model. However, if the first relational information is obtained by machine learning methods, it is preferable from the standpoint of higher accuracy in estimating parameter contributions compared to first relational information obtained by methods other than machine learning methods. Machine learning methods are capable of obtaining mappings expressed by functions that are expressed by multiple nonlinear functions and cannot be expressed in closed form, while mappings obtained by methods other than machine learning are mappings expressed by linear functions or equations expressed in closed form.

[0074] Therefore, first relational information obtained by methods other than machine learning is more likely to have lower approximation accuracy than first relational information obtained by machine learning methods. Consequently, when the design support system 100 estimates the parameter contribution using first relational information obtained by machine learning methods, the estimation accuracy is higher than when it estimates the parameter contribution using first relational information obtained by methods other than machine learning methods.

[0075] Furthermore, the design support system 100 of the first embodiment configured in this way can acquire first relational information and obtain the parameter contribution when the shape parameter values ​​of the design object are changed based on the first relational information. Therefore, once the first relational information has been obtained, when acquiring shape parameter values ​​for the design object, it is not necessary to perform processing according to the rules for acquiring training data such as structural analysis, cross-sectional analysis, static analysis, or dynamic analysis each time the shape parameters are estimated. Incidentally, as mentioned above, the first relational information is information that shows the relationship between shape parameters and property parameters. The first relational information can be obtained, for example, by machine learning methods.

[0076] When obtained by machine learning methods, the obtained first relational information is information that shows the relationship between the input and output of analyses such as structural analysis, cross-sectional analysis, static analysis, or dynamic analysis, and is information that omits the analysis process. Therefore, a design support system 100 that uses a first learning model that has been trained as the first relational information can reduce the amount of computation required when obtaining the values ​​of property parameters based on the values ​​of shape parameters.

[0077] In the first embodiment configured in this way, the relationship information acquisition device 1 acquires first relationship information. Once the first relationship information is obtained, the support device 2 can obtain the parameter contribution when the shape parameter value is changed. Therefore, the relationship information acquisition device 1 can reduce the burden required for design when designing buildings.

[0078] In the first embodiment configured in this way, the support device 2 obtains the parameter contribution when the shape parameter value is changed using the first relational information. Therefore, the support device 2 can reduce the burden required for design when designing buildings.

[0079] (Second Embodiment) Figure 9 is an explanatory diagram illustrating the outline of the design support system 100a of the second embodiment. For the sake of simplicity, components having the same functions as those described in Figures 1 to 6 will be denoted by the same reference numerals as in Figures 1 to 6, and their descriptions will be omitted. The design support system 100a differs from the design support system 100 in that it includes a relational information acquisition device 1a instead of relational information acquisition device 1, and a support device 2a instead of support device 2.

[0080] The relational information acquisition device 1a differs from the relational information acquisition device 1 in that it acquires second relational information instead of first relational information. The second relational information is information that shows the relationship between shape parameters and parameter contributions.

[0081] The relational information acquisition device 1a acquires second relational information, for example, by machine learning. That is, the relational information acquisition device 1a acquires a mathematical model showing the relationship between shape parameters and parameter contributions by machine learning. The design support system 100a will be described below using the case where the relational information acquisition device 1a acquires second relational information by machine learning as an example.

[0082] When the relational information acquisition device 1a acquires second relational information using a machine learning method, the relational information acquisition device 1a updates a learning model (hereinafter referred to as the "second learning model") that shows the relationship between shape parameters and parameter contributions through learning. The relational information acquisition device 1a continues learning until the learning termination condition is met. The second learning model at the point when the learning termination condition is met is used for the analysis of the design target. In other words, the trained second learning model is used for the analysis of the design target.

[0083] The relational information acquisition device 1a performs training on the second training dataset when training the second learning model. The second training dataset is a collection of one or more second unit training sets. Each second unit training set is a pair of shape parameters and parameter contributions. More specifically, when training the second learning model, the relational information acquisition device 1a performs training using shape parameters as training data and parameter contributions as training data. In other words, when training the second learning model, the relational information acquisition device 1a performs training using shape parameters as explanatory variables and parameter contributions as the target variable.

[0084] The training data in the second training dataset may be measurement results of already constructed buildings, or it may be obtained using training data acquisition rules. The training data in the second training dataset may be obtained, for example, by performing a first contribution estimation process using a trained first training model. That is, the training data in the second training dataset may be obtained, for example, by a design support system 100.

[0085] In the design support system 100a, the training data corresponding to the learning data may be input to the relational information acquisition device 1a by an external device or user. In the design support system 100a, the training data corresponding to the learning data may be calculated by the relational information acquisition device 1a by executing a first contribution estimation process using a first learning model that has been trained based on the input learning data.

[0086] For the sake of simplicity, the design support system 100a will be described below using the example of a case where the relational information acquisition device 1a acquires corresponding training data based on the input learning data. In particular, the design support system 100a will be described using the example of a case where the relational information acquisition device 1a obtains training data by executing the first contribution estimation process. Hereinafter, the process by which the relational information acquisition device 1a acquires training data based on the learning data in accordance with the training data acquisition rules will be referred to as the second training data acquisition process.

[0087] Furthermore, if training data is input to the relational information acquisition device 1a by an external device or user, the relational information acquisition device 1a only needs to use the input training data for learning and does not need to acquire the training data by executing a second training data acquisition process. In other words, if training data is input to the relational information acquisition device 1a by an external device or user, the relational information acquisition device 1a only needs to use the input training data for learning and does not need to execute a second training data acquisition process.

[0088] The support device 2a accepts input of multiple shape parameter values ​​(i.e., first analysis target data) for the design object. Based on the input first analysis target data, the support device 2a uses second relational information, such as a trained second learning model acquired by the relational information acquisition device 1a, to estimate the parameter contribution for each shape parameter for some or all of the shape parameters included in the first analysis target data with respect to the design object. Hereinafter, the process of estimating the parameter contribution using the second relational information based on the input shape parameter values ​​will be referred to as the second contribution estimation process.

[0089] Thus, the support device 2a differs from the support device 2 in that it estimates the parameter contribution using second relational information, such as a pre-trained second learning model, instead of first relational information, such as a pre-trained first learning model. The support device 2a comprises an output control unit 200 and an output unit 20.

[0090] Figure 10 shows an example of the hardware configuration of the relationship information acquisition device 1a in the second embodiment. The relationship information acquisition device 1a differs from the relationship information acquisition device 1 in that it has a control unit 11a instead of a control unit 11.

[0091] Figure 11 shows an example of the functional configuration of the control unit 11a in the second embodiment. The control unit 11a differs from the control unit 11 in that it includes a learning dataset acquisition unit 110a instead of the learning dataset acquisition unit 110, and a relational information acquisition unit 120a instead of the relational information acquisition unit 120.

[0092] The training dataset acquisition unit 110a differs from the training dataset acquisition unit 110 in that it acquires a second training dataset instead of the first training dataset. The training dataset acquisition unit 110a acquires training data in the same way as the training dataset acquisition unit 110.

[0093] The training dataset acquisition unit 110a acquires training data for each training dataset by executing a second training data acquisition process based on the acquired training data. More specifically, the training dataset acquisition unit 110a acquires training data corresponding to each training dataset by executing a second training data acquisition process for each acquired training dataset. Specifically, the second training data acquisition process executed by the training dataset acquisition unit 110a is, for example, a first contribution estimation process.

[0094] The relational information acquisition unit 120a acquires second relational information using the second training dataset acquired by the training dataset acquisition unit 110a. For example, the relational information acquisition unit 120a uses the second training dataset acquired by the training dataset acquisition unit 110a to train the second training model until the training completion condition is met. In this way, the relational information acquisition unit 120a acquires the trained second training model. The trained second training model is an example of second relational information.

[0095] Thus, the relational information acquisition unit 120a differs from the relational information acquisition unit 120 in that, instead of acquiring first relational information using the first training dataset, it acquires second relational information using the second training dataset.

[0096] Figure 12 is a flowchart showing an example of the processing flow performed by the relational information acquisition device 1a in the second embodiment.

[0097] The learning dataset acquisition unit 110a acquires the second learning dataset (step S101a). The learning data of the second learning dataset acquired by the learning dataset acquisition unit 110a is, for example, the learning data input to the communication unit 12. The learning data of the second learning dataset acquired by the learning dataset acquisition unit 110 may be, for example, the learning data input to the input unit 13. The learning data of the first learning dataset acquired by the learning dataset acquisition unit 110a may be, for example, the learning data read from the storage unit 14 in accordance with the first read instruction.

[0098] Next, the relational information acquisition unit 120a uses one second unit learning set included in the second learning dataset acquired in step S101a, which has not yet been used to train the second learning model, to train the second learning model (step S102a). Next, the relational information acquisition unit 120a determines whether the learning termination condition has been met (step S103a). If the learning termination condition has been met (step S103a: YES), the memory control unit 130 records the trained second learning model in the memory unit 14 (step S104a). Note that the processing in step S104a may be replaced by the communication control unit 160 transmitting the trained second learning model to the support device 2a instead of the memory control unit 130 recording the trained second learning model in the memory unit 14. On the other hand, if the learning termination condition has not been met (step S103a: NO), the process returns to step S102a.

[0099] Figure 12 shows an example of the processing flow executed by the relational information acquisition device 1a, using the case where the second relational information is a second learned model that has already been learned. Therefore, the processing in steps S102a and S103a is an example of the processing flow for acquiring the second relational information.

[0100] Figure 13 shows an example of the hardware configuration of the support device 2a in the second embodiment. The support device 2a differs from the support device 2 in that it has a control unit 21a instead of a control unit 21.

[0101] Figure 14 shows an example of the functional configuration of the control unit 21a in the second embodiment. The control unit 21a differs from the control unit 21 in that it includes a contribution estimation unit 220a instead of the contribution estimation unit 220. The contribution estimation unit 220a differs from the contribution estimation unit 220 in that it performs a second contribution estimation process instead of a first contribution estimation process.

[0102] Figure 15 is a flowchart showing an example of the processing flow performed by the support device 2a in the second embodiment. For simplicity of explanation, the same processes as in Figure 7 will be denoted by the same reference numerals as in Figure 7, and their explanation will be omitted below.

[0103] The process in step S201 is executed. Next, the contribution estimation unit 220a performs the second contribution estimation process (step S202a). Then, the output control unit 200 outputs the parameter contribution estimated in step S202a to the output unit 20 (step S203a).

[0104] The design support system 100a of the second embodiment, configured in this way, can acquire second relational information and, based on the second relational information, obtain the parameter contribution when the shape parameter values ​​are changed for the design target. Therefore, a designer using the design support system 100 does not necessarily need to analyze all candidate designs when designing. For example, if the designer sees that the estimation results of the design support system 100 indicate that a change in shape parameters is unlikely to yield a favorable result, the designer can choose not to consider it further. As a result, the design support system 100a can reduce the burden required for designing buildings.

[0105] In the second embodiment configured in this way, the relationship information acquisition device 1a acquires second relationship information. Once the second relationship information is obtained, the support device 2a can obtain the parameter contribution when the shape parameter value is changed. Therefore, the relationship information acquisition device 1a can reduce the burden required for design when designing buildings.

[0106] In the second embodiment configured in this way, the support device 2a obtains the parameter contribution when the shape parameter value is changed using the second relational information. Therefore, the support device 2a can reduce the burden required for design when designing buildings.

[0107] (Third embodiment) Figure 16 is an explanatory diagram illustrating the outline of the design support system 100b of the third embodiment. For the sake of simplicity, components having the same functions as those described in Figures 1 to 6 will be denoted by the same reference numerals as in Figures 1 to 6, and their descriptions will be omitted. The design support system 100b differs from the design support system 100 in that it includes a relational information acquisition device 1b instead of relational information acquisition device 1, and a support device 2b instead of support device 2.

[0108] The relation information acquisition device 1b differs from the relation information acquisition device 1 in that it acquires third relation information instead of first relation information. The third relation information is information that shows the relationship between property parameters and parameter contributions.

[0109] The relational information acquisition device 1b acquires third relational information, for example, by machine learning. That is, the relational information acquisition device 1b acquires a mathematical model showing the relationship between property parameters and parameter contributions by machine learning. The design support system 100b will be described below using the case where the relational information acquisition device 1b acquires third relational information by machine learning as an example.

[0110] When the relational information acquisition device 1b acquires third relational information using a machine learning method, the relational information acquisition device 1b updates a learning model (hereinafter referred to as the "third learning model") that shows the relationship between property parameters and parameter contributions through learning. The relational information acquisition device 1b continues learning until the learning termination condition is met. The third learning model at the point when the learning termination condition is met is used for the analysis of the design target. In other words, the trained third learning model is used for the analysis of the design target.

[0111] The relational information acquisition device 1b performs training on the third learning model based on the third learning dataset. The third learning dataset is a collection of one or more third unit learning sets. Each third unit learning set is a pair of property parameters and parameter contributions. More specifically, when training the third learning model, the relational information acquisition device 1b uses property parameters as training data and parameter contributions as training data. In other words, when training the third learning model, the relational information acquisition device 1b uses property parameters as explanatory variables and parameter contributions as the target variable.

[0112] The relational information acquisition device 1b acquires a third training dataset. The third training dataset may consist of measurement results of completed buildings for both the training data and the training data. The third training dataset may consist of measurement results of buildings for the training data and training data calculated using training data acquisition rules. The third training dataset may be acquired by performing a first contribution estimation process and a second contribution estimation process on multiple shape parameter values, using the results of the first contribution estimation process as training data and the results of the second contribution estimation process as training data.

[0113] In the design support system 100b, the third training dataset may be input to the relational information acquisition device 1b by an external device or user. In the design support system 100b, the training dataset may be input by an external device or user, and the training data may be calculated by the relational information acquisition device 1b using training data acquisition rules.

[0114] For the sake of simplicity, the design support system 100b will be described below using the example where both the training data and the teacher data of the third training dataset are input by an external device or user to the relational information acquisition device 1b.

[0115] The support device 2b accepts input of multiple property parameter values ​​for the design object (hereinafter referred to as "second analysis target data"). Based on the input second analysis target data, the support device 2b uses third relational information, such as a trained third learning model acquired by the relational information acquisition device 1b, to estimate the parameter contribution for each property parameter for some or all of the property parameters included in the second analysis target data with respect to the design object. Hereinafter, the process of estimating the parameter contribution using the third relational information based on the input property parameter values ​​is referred to as the third contribution estimation process.

[0116] Thus, the support device 2b differs from the support device 2 in that it estimates the parameter contribution using third relational information, such as a pre-trained third learning model, instead of first relational information, such as a pre-trained first learning model. The support device 2b comprises an output control unit 200 and an output unit 20.

[0117] Figure 17 shows an example of the hardware configuration of the relationship information acquisition device 1b in the third embodiment. The relationship information acquisition device 1b differs from the relationship information acquisition device 1 in that it has a control unit 11b instead of a control unit 11.

[0118] Figure 18 shows an example of the functional configuration of the control unit 11b in the third embodiment. The control unit 11b differs from the control unit 11 in that it includes a learning dataset acquisition unit 110b instead of a learning dataset acquisition unit 110, and a relational information acquisition unit 120b instead of a relational information acquisition unit 120.

[0119] The training dataset acquisition unit 110b differs from the training dataset acquisition unit 110 in that it acquires a third training dataset instead of the first training dataset.

[0120] The relational information acquisition unit 120b acquires third relational information using the third training dataset acquired by the training dataset acquisition unit 110b. For example, the relational information acquisition unit 120b uses the third training dataset acquired by the training dataset acquisition unit 110b to train the third learning model until the training completion condition is met. In this way, the relational information acquisition unit 120b acquires the trained third learning model. The trained third learning model is an example of third relational information.

[0121] Thus, the relational information acquisition unit 120b differs from the relational information acquisition unit 120 in that, instead of acquiring first relational information using the first training dataset, it acquires third relational information using the third training dataset.

[0122] Figure 19 is a flowchart showing an example of the processing flow performed by the relational information acquisition device 1b in the third embodiment.

[0123] The training dataset acquisition unit 110b acquires the third training dataset (step S101b). Next, the relational information acquisition unit 120b uses one third unit learning set included in the third training dataset acquired in step S101b, which has not yet been used to train the third training model, to train the third training model (step S102b). Next, the relational information acquisition unit 120b determines whether the training completion condition has been met (step S103b).

[0124] If the learning completion condition is met (step S103b: YES), the memory control unit 130 records the learned third learning model in the storage unit 14 (step S104b). Note that the process in step S104b may be replaced by the communication control unit 160 transmitting the learned third learning model to the support device 2b instead of the memory control unit 130 recording the learned third learning model in the storage unit 14. On the other hand, if the learning completion condition is not met (step S103b: NO), the process returns to step S102b.

[0125] Figure 19 shows an example of the processing flow executed by the relational information acquisition device 1b, using the case where the third relational information is a pre-learned third learning model as an example. Therefore, the processing in steps S102b and S103b is an example of the processing flow for acquiring the third relational information.

[0126] Figure 20 shows an example of the hardware configuration of the support device 2b in the third embodiment. The support device 2b differs from the support device 2 in that it has a control unit 21b instead of a control unit 21.

[0127] Figure 21 shows an example of the functional configuration of the control unit 21b in the third embodiment. The control unit 21b differs from the control unit 21 in that it includes an analysis target data acquisition unit 210b instead of the analysis target data acquisition unit 210, and a contribution estimation unit 220b instead of the contribution estimation unit 220.

[0128] The data acquisition unit 210b differs from the data acquisition unit 210b in that it acquires the second data to be analyzed instead of the first data to be analyzed. The contribution estimation unit 220b differs from the contribution estimation unit 220 in that it performs a third contribution estimation process instead of a first contribution estimation process.

[0129] Figure 22 is a flowchart showing an example of the processing flow performed by the support device 2b in the third embodiment. The data acquisition unit 210b acquires the second data to be analyzed (step S201b). The second data to be analyzed acquired by the data acquisition unit 210b is, for example, the second data to be analyzed that has been input to the communication unit 22. The second data to be analyzed acquired by the data acquisition unit 210b may also be, for example, the second data to be analyzed that has been input to the input unit 23. The second data to be analyzed acquired by the data acquisition unit 210b may also be, for example, the second data to be analyzed that has been read from the storage unit 24 in accordance with the third read instruction. The third read instruction is an instruction to use a portion of the second data to be analyzed already stored in the storage unit 24 for the third contribution estimation process. The third read instruction is input to the communication unit 22 or the input unit 23.

[0130] Next, the contribution estimation unit 220b performs the third contribution estimation process (step S202b). Then, the output control unit 200 outputs the parameter contribution estimated in step S202b to the output unit 20 (step S203b).

[0131] The design support system 100b of the third embodiment, configured in this way, can acquire third-party relational information and, based on this third-party relational information, obtain the parameter contribution when the values ​​of property parameters are changed for the design object. For example, if a designer wants to change the values ​​of property parameters to approach the desired shape, the design support system 100b can be used to understand the influence of each shape parameter and obtain guidance on how to change the shape parameters. Therefore, the design support system 100b can reduce the burden required for designing buildings.

[0132] The relationship information acquisition device 1b in the third embodiment configured in this way acquires third relationship information. Once the third relationship information is obtained, the support device 2b can obtain the parameter contribution when the value of the property parameter is changed. For example, if a designer wants to change the value of a property parameter to get closer to the desired shape, the design support system 100b can be used to find out the effect of each shape parameter and get guidance on how to change the shape parameters. Therefore, the relationship information acquisition device 1b can reduce the burden required for design when designing buildings.

[0133] In the third embodiment configured in this way, the support device 2b uses third relational information to obtain the parameter contribution when the value of a property parameter is changed. For example, if a designer wants to change the value of a property parameter to approach the desired shape, the design support system 100b can be used to find out the effect of each shape parameter and get guidance on how to change the shape parameters. Therefore, the support device 2b can reduce the burden required for design when designing buildings.

[0134] (modified version) The input unit 13 may receive information instructing the output unit 20 to output information (hereinafter referred to as "output control information"). When output control information is input to the input unit 13, the input control unit 140 acquires the input output control information. When the input control unit 140 acquires the output control information, the output control unit 150 controls the operation of the output unit 20 to cause the output unit 20 to output information that conforms to the instructions of the output control information. The output control information is, for example, information instructing the conditions for the output parameter contribution. In such a case, the output control unit 150 causes the output unit 20 to output information that satisfies the conditions instructed by the output control information. The conditions for the output parameter contribution are, for example, that the height of the parameter contribution is greater than or equal to a predetermined height.

[0135] Support devices 2 and 2a may perform a process to assist the user in inputting multiple shape parameter values ​​(hereinafter referred to as "input assistance processing") when multiple shape parameter values ​​are input by the user. Input assistance processing is a process that causes a predetermined device that outputs images or sounds prompting the input or selection of multiple shape parameter values ​​to be output according to predetermined rules (hereinafter referred to as "output rules"). The predetermined device that outputs images or sounds is, for example, the output unit 20. Input assistance processing is a process performed by the output control unit 150.

[0136] Output rules are rules that execute the processing flow shown in the flowcharts in Figures 23 and 24 below. In other words, the processing flow shown in the flowcharts in Figures 23 and 24 is an example of input assistance processing. Output rules are rules that execute the processing flow shown in the flowchart in Figure 25 below. In other words, the processing flow shown in the flowchart in Figure 25 is an example of input assistance processing.

[0137] Figure 23 is a first flowchart showing a first example of the input assistance processing flow performed by the output control unit 150 in a modified example. Figure 24 is a second flowchart showing a first example of the input assistance processing flow performed by the output control unit 150 in a modified example.

[0138] The output control unit 150 displays a screen on the output unit 20 prompting the user to input whether or not to add a new roof structure (step S301). The user inputs information corresponding to the displayed image via the input unit 23. If a new roof structure is to be added (step S301: YES), the output control unit 150 displays a screen on the output unit 20 prompting the user to input whether or not the added roof structure is a truss structure (step S302). The user inputs information corresponding to the displayed image via the input unit 23.

[0139] If the additional roof structure is a truss structure (step S302: YES), the output control unit 150 switches its own operating mode to the input control mode for the truss structure (step S303). The input control mode for the truss structure is an operating mode in which predetermined displays related to the truss structure are displayed on the output unit 20 in a predetermined order when predetermined conditions are met.

[0140] Next, the output control unit 150 displays a screen on the output unit 20 prompting the user to set the number of lines (step S304). The user inputs information corresponding to the displayed image via the input unit 23. Next, the output control unit 150 displays a screen on the output unit 20 prompting the user to set the number of divisions (step S305). The user inputs information corresponding to the displayed image via the input unit 23.

[0141] Next, the output control unit 150 determines whether a new division position has been set based on the information entered by the user in response to the display in step S305 (step S306). If a new division position has been set (step S306: YES), the output control unit 150 then determines whether there is a new diagonal member based on the information entered by the user in response to the display in step S305 (step S307). If there is a new diagonal member (step S307: YES), the output control unit 150 displays a screen on the output unit 20 prompting the user to set the shape of the new diagonal member arrangement (step S308). The user inputs information in response to the displayed image via the input unit 23.

[0142] Next, the output control unit 150 displays a screen on the output unit 20 prompting the user to set whether or not there are new bundle materials (step S309). The user inputs information corresponding to the displayed image via the input unit 23. Next, the output control unit 150 determines whether or not all settings for the truss structure have been completed (step S310). If all settings are completed (step S310: YES), the process ends. On the other hand, if there are still items that have not been set (step S310: NO), the process returns to step S303.

[0143] If there are no new diagonal members in step S307 (step S307: NO), proceed to step S309. If no new division positions are set in step S306 (step S306: NO), proceed to step S310.

[0144] If the roof structure added in step S302 is not a truss structure (step S302: NO), the output control unit 150 switches its own operating mode to the input control mode for a tensioned beam structure (step S311). The input control mode for a tensioned beam structure is an operating mode in which predetermined displays related to the tensioned beam structure are displayed on the output unit 20 in a predetermined order when predetermined conditions are met.

[0145] Next, the output control unit 150 displays a screen on the output unit 20 prompting the user to set the number of wires (step S312). The user inputs information corresponding to the displayed image via the input unit 23. Next, the output control unit 150 displays a screen on the output unit 20 prompting the user to set the number of divisions (step S313). The user inputs information corresponding to the displayed image via the input unit 23. Next, the output control unit 150 determines whether all settings for the tensioned beam structure have been completed (step S314). If all settings have been completed (step S314: YES), the process ends. On the other hand, if there are still settings that have not been set (step S314: NO), the process returns to step S312.

[0146] If a new division position is set in step S306 (step S306: YES), the output control unit 150 displays an image connecting the division positions on the output unit 20. A division position refers to the (n+1) nodal coordinates that result when the linear members constituting the upper chord and lower chord are divided into n sections (where n is a natural number). Connecting the division positions means connecting two independent nodal coordinates with a line such as a diagonal member or a bundle member that connects them with a straight line. A node refers to a point where two or more members are joined together.

[0147] Figure 25 is a flowchart showing a second example of the input assistance processing flow performed by the output control unit 150 in a modified example. The output control unit 150 displays a screen on the output unit 20 prompting the user whether or not to add a new central tension ring (step S401). The user inputs information corresponding to the displayed image via the input unit 23. Next, the output control unit 150 displays a screen on the output unit 20 prompting the user whether or not the new central tension ring is polygonal (step S402). The user inputs information corresponding to the displayed image via the input unit 23.

[0148] If the new central tension ring is polygonal (step S402: YES), the output control unit 150 displays a screen on the output unit 20 prompting the user to set the number of corners (step S403). The user inputs information corresponding to the displayed image via the input unit 23. Next, the output control unit 150 determines whether all settings for the central tension ring have been completed (step S404). If all settings are completed (step S404: YES), the process ends. On the other hand, if there are still settings that have not been set (step S404: NO), the process returns to step S401.

[0149] If, in step S402, the new central tension ring is not polygonal (i.e., circular) (step S402: NO), the output control unit 150 displays a screen on the output unit 20 prompting the user to set the curvature (step S405). The user inputs information corresponding to the displayed image via the input unit 23. The process then proceeds to step S404.

[0150] If there is no new central tension ring in step S401 (step S401: NO), proceed to step S404.

[0151] Thus, at least some of the values ​​of multiple shape parameters are required to be entered or selected only when certain conditions are met. Therefore, at least some of the values ​​of multiple shape parameters must be entered or selected according to predetermined rules.

[0152] Thus, in the input assistance processing, shape parameters are classified and displayed as structural parameters and member parameters. Structural parameters are variables that indicate the structure of the building, while member parameters are variables that indicate information about the members used in the structure. Structural parameters indicate, for example, whether the roof structure is a truss structure or a tensioned beam structure. Member parameters are variables that indicate information about, for example, upper chord members, lower chord members, diagonal members, bracing members and nodes.

[0153] The input assistance process, as shown in Figures 23 to 25, assists in the input or selection of multiple shape parameter values. Therefore, the input assistance process assists in the generation of training data.

[0154] Up to this point, we have explained an example of input assistance processing using the case where shape parameters are input, but input assistance processing may also be performed for the input of property parameters. Hereafter, when we do not distinguish between the values ​​of shape parameters and the values ​​of property parameters, we will refer to them as parameter values.

[0155] Thus, the support devices 2, 2a, and 2b, each equipped with an output control unit 150, can reduce the frequency or time spent by the user being confused about the order in which to input or select multiple parameter values ​​by outputting images or sounds that prompt the user to input or select multiple parameter values ​​according to output rules. In other words, the support devices 2, 2a, and 2b, each equipped with an output control unit 150 that performs input assistance processing, can reduce the user's time and effort required to input information.

[0156] Therefore, support devices 2, 2a, and 2b, which are equipped with an output control unit 150 that performs such input assistance processing, can reduce the user's time and effort required for information input. Accordingly, a design support system 100 equipped with support device 2 equipped with an output control unit 150 that performs input assistance processing can reduce the user's time and effort required for information input. A design support system 100a equipped with support device 2a equipped with an output control unit 150 that performs input assistance processing can reduce the user's time and effort required for information input. A design support system 100b equipped with support device 2b equipped with an output control unit 150 that performs input assistance processing can reduce the user's time and effort required for information input.

[0157] Figures 26 to 28 will be used to explain truss structures, tensioned beam structures, upper chord members, lower chord members, diagonal members, bracing members, and nodes.

[0158] Figure 26 shows an example of a truss structure in a modified example. The truss structure is composed of upper chords, lower chords, diagonal members, and bracing members.

[0159] Figure 27 shows an example of a tensioned beam structure in a modified example. The tensioned beam structure is composed of an upper chord, a lower chord, and a brace. In the tensioned beam structure, the lower chord is composed of, for example, a tension member. The tension member may be, for example, a rod or a rope.

[0160] Figure 28 is an explanatory diagram illustrating the upper chord, lower chord, diagonal member, brace member, and nodes in a modified example. The upper chord and lower chord are both members parallel to the X-axis in Figure 26, although their positions in the Y-axis direction differ. The brace member is a member parallel to the Y-axis in Figure 26 and connects the upper chord and the lower chord. Nodes are the intersections of the brace member with the upper chord or lower chord. Diagonal members are members located on the diagonals of the rectangle enclosed by four nodes.

[0161] Furthermore, the design support systems 100, 100a, and 100b, the related information acquisition devices 1, 1a, and 1b, and the support devices 2, 2a, and 2b may each be implemented using multiple information processing devices connected to each other via a network. In this case, the functional units of the design support systems 100, 100a, and 100b, the related information acquisition devices 1, 1a, and 1b, and the support devices 2, 2a, and 2b may be distributed and implemented across multiple information processing devices.

[0162] Support devices 2, 2a, and 2b are examples of output devices. The trained first learning model, the trained second learning model, and the trained third learning model are examples of trained models. Relational information acquisition devices 1, 1a, and 1b are examples of learning devices.

[0163] Furthermore, all or part of the functions of the design support systems 100, 100a, 100b, the related information acquisition devices 1, 1a, 1b, and the support devices 2, 2a, 2b may be implemented using hardware such as ASICs (Application Specific Integrated Circuits), PLDs (Programmable Logic Devices), or FPGAs (Field Programmable Gate Arrays). The program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program may also be transmitted via telecommunications lines.

[0164] As mentioned above, the problem that design support systems 100, 100a, and 100b aim to solve is to reduce the burden of design when designing buildings. This problem of reducing the burden of design when designing buildings includes, for example, reducing the burden of design by facilitating communication between designers and engineers. Communication is smoother the shorter the time it takes for designers and engineers to understand the effects that arise from changes in building-related parameters such as shape parameters and property parameters.

[0165] Therefore, a device that does not estimate the parameter contribution in response to changes in building-related parameters such as shape parameters and property parameters cannot facilitate communication between designers and engineers. On the other hand, design support systems 100, 100a, and 100b can facilitate communication because they estimate the parameter contribution.

[0166] To elaborate further, the design support system 100, design support system 100a, and design support system 100b facilitate communication between designers and engineers, using design support system 100 as an example. Design support system 100 uses property parameters such as weight as training data and shape parameters as learning data, but depth can be used as learning data. Since depth is a quantity defined above, it is a type of parameter that indicates the shape of the exterior or interior of a building (hereinafter referred to as "appearance parameter"). Rise is also an example of an appearance parameter. As learning data, parameters that have a stronger influence on the physical properties of a building, such as the cross-sectional area, length, or material of columns, than on the appearance or interior of the building (hereinafter referred to as "physical shape parameter") may be used.

[0167] While physical shape parameters certainly influence the appearance or interior, from a designer's perspective, their impact on the design is smaller than that of visual parameters such as depth. The reasons are as follows: With materials, the design can be changed indefinitely by processing the surface. Also, for example, when a designer designs a small house and a large house, if the shapes of both are the same, the designer can propose the same design for both houses, so parameters related to size such as cross-sectional area and length have little impact on the design. Therefore, from the perspective of facilitating smooth communication between designers and engineers, it is preferable to use visual parameters rather than physical shape parameters in the training data.

[0168] While appearance parameters such as depth have a stronger influence on the appearance or interior than physical properties compared to physical shape parameters, they do not necessarily only affect the appearance or interior, but also influence physical properties.

[0169] In this way, the design support system 100, design support system 100a, and design support system 100b perform learning using visual parameters such as depth and estimate the parameter contribution, thereby facilitating smoother communication with designers in design processes involving designers compared to simply estimating the parameter contribution.

[0170] Note that the user input shown in Figures 23 to 25 does not necessarily have to be performed by the user; for example, it may be performed automatically by a computer according to predetermined rules. Automatic input may be performed by, for example, the input control unit 240. Automatic input may also be performed by, for example, the output control unit 200 itself.

[0171] Generally, numerical analysis algorithms for calculating the contribution of shape parameters of spatial structures are estimated to involve complex nonlinear polynomials. Therefore, generating such algorithms manually is considered practically impossible, and no numerical analysis algorithms for calculating the contribution of shape parameters of spatial structures have been reported in conventional technology. The design support system 100 enables the calculation of shape parameter contributions by using machine learning. This allows the design support system 100 to quantitatively explain to clients and designers, along with the reasons, the proposed designs adopted by the designers regarding the design of spatial structures.

[0172] If the designer can quantitatively explain to the designer the reasoning behind the design of the spatial structure, it becomes easier to obtain agreement between the designer and the client. Furthermore, it becomes easier to obtain agreement not only between the designer and the client, but also between the designer and the client. In other words, the design support system 100 can alleviate the difficulty in obtaining agreement between the designer and the client. Therefore, the problem that the design support system 100 solves may be to alleviate the difficulty in obtaining agreement between the designer and the client.

[0173] Incidentally, in design fields other than spatial structures, the correlation between training data and target data is simple, and it is sometimes easy to generate numerical analysis algorithms. By simple, I mean that it is not nonlinear. Therefore, in design fields other than spatial structures, the correlation is simple, making it easy to obtain agreement between designers and engineers, and the problem of wanting to alleviate the difficulty of obtaining agreement between designers and engineers does not exist in the first place. Thus, the problem that design support system 100 solves, which is to alleviate the difficulty of obtaining agreement between designers and engineers, is a problem unique to design including spatial structures and does not exist in design fields other than spatial structures.

[0174] Furthermore, even within the same field of architecture, when a building is a multi-story structure, the concept of calculating the contribution of each element doesn't even arise because the structure of a multi-story structure is simple. And, as mentioned above, because the structure of a multi-story structure is simple, even if the contribution of each element were to be calculated for a multi-story building, the numerical analysis algorithm for calculating the contribution of the shape parameters of a multi-story structure is expected to be relatively simple. Therefore, even when a building is a multi-story structure, the challenge of mitigating the difficulty of obtaining agreement between the designer and the architect does not exist, as is the case with other fields of design that do not include spatial structures. A spatial structure is, for example, a structure that has columns only on the perimeter and no columns inside, covering a large space. Examples of spatial structures include gymnasiums and domes. On the other hand, a multi-story structure is a structure in which spaces are stacked in multiple layers. Examples of multi-story structures include multi-story buildings.

[0175] <Experiment to evaluate performance> This section describes an experiment to evaluate the performance of the design support system 100 (hereinafter referred to as the "performance evaluation experiment"). Figures 29 to 37 show an example of the results of the performance evaluation experiment. The training data and training data used in the experiment were obtained using well-known analytical methods such as structural analysis.

[0176] In the performance evaluation experiment, the performance of the design support system 100 was evaluated by comparing the results obtained using well-known analytical methods such as structural analysis for the gymnasium roof with the results obtained using the design support system 100. For simplicity of explanation, the experimental results for design support system 100 will be described, but similar results can be obtained for design support systems 100a and 100b. Furthermore, the gymnasium roof is just one example of a structure used for performance evaluation, and the results of the performance evaluation will be similar for other structures.

[0177] In the performance evaluation experiment, shape parameters such as depth, rise, and the number of main beams were set as variables during the shape data generation process for generating the training data and target data. The shape data generation process is the process of generating shape data for the building (i.e., the gymnasium roof) used in the performance evaluation experiment. The range of possible values ​​for the shape parameters during shape data generation was set to predetermined values. For example, the rise was set to a value between 0 meters and 10 meters. For example, the depth was set to a value between 0.5 meters and 10 meters. Hereinafter, the process of generating the training data and target data will be referred to as the training set generation process.

[0178] In the training set generation process, each value of the shape parameter is selected according to a predetermined rule within a range set in advance for each shape parameter, and data representing the structural shape is generated using the selected values. For example, a rise of 5 meters and a depth of 6 meters were selected, and a structural shape with a rise of 5 meters and a depth of 6 meters was generated.

[0179] In the training set generation process, data representing the frame shape was created for all combinations of shape parameter values. All of the generated frame shape data was used as training data in the performance evaluation experiment.

[0180] In the training set generation process, information on material, cross-section, support points, and load was added to the obtained training data. Note that the information on material, cross-section, support points, and load was the same regardless of the training data.

[0181] In the training set generation process, property parameters such as stress, displacement, weight, and strain were calculated by performing a well-known stress analysis using training data to which information on material, cross-section, support points, and loads had been added. In the performance evaluation experiment, the calculated property parameters were used as training data. In this way, the training set generation process generated the training data and training data used in the performance evaluation experiment.

[0182] In the performance evaluation experiment, learning was performed using the obtained training data and training data. The learning was performed by relational information acquisition device 1. In the performance evaluation experiment, support device 2 was used to estimate results according to the input information using the results of the learning performed by relational information acquisition device 1. The estimation results by support device 2 are the estimation results of the design support system 100. In the performance evaluation experiment, the shape parameters of the building to be estimated were input to support device 2. In the performance evaluation experiment, the deformation, strain energy, and weight of the building to be estimated were estimated based on the shape parameters of the building to be estimated input to support device 2.

[0183] Figure 29 is the first figure showing an example of the results of a performance evaluation experiment in a modified example. Figure 29 shows the results of deformation estimation by the design support system 100 compared with experimental values. The horizontal axis of Figure 29 shows the estimation results by the design support system 100. The vertical axis of Figure 29 shows the experimental values. Figure 29 shows that the estimation results of the design support system 100 show good agreement with the experimental values. The coefficient of determination was 99.9%.

[0184] Figure 30 is a second figure showing an example of the results of a performance evaluation experiment in a modified example. Figure 30 shows the results of strain energy estimation by the design support system 100 compared with experimental values. The horizontal axis of Figure 30 shows the estimation results by the design support system 100. The vertical axis of Figure 30 shows the experimental values. Figure 30 shows that the estimation results of the design support system 100 show good agreement with the experimental values. The coefficient of determination was 99.9%.

[0185] Figure 31 is the third figure showing an example of the results of a performance evaluation experiment in a modified example. Figure 31 shows the results of weight estimation by the design support system 100 compared with experimental values. The horizontal axis of Figure 31 shows the estimation results by the design support system 100. The vertical axis of Figure 31 shows the experimental values. Figure 31 shows that the estimation results of the design support system 100 show good agreement with the experimental values. The coefficient of determination was 99.9%.

[0186] Figure 32 is the fourth figure, showing an example of the results of a performance evaluation experiment in a modified example. Figure 32 is an example of the deformation estimation results by the design support system 100. The vertical axis of Figure 32 represents the shape parameters. The horizontal axis of Figure 32 shows the degree of influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in vertical deformation Z. As described above, the degree of influence that changes in the values ​​of shape parameters have on the change in vertical deformation Z is a type of parameter contribution. Therefore, the results in Figure 32 show an example of the parameter contribution obtained as a result of deformation estimation by the design support system 100.

[0187] Figure 33 is the fifth figure showing an example of the results of a performance evaluation experiment in a modified example. Figure 33 is an example of the strain energy estimation results by the design support system 100. The vertical axis of Figure 33 represents the shape parameters. The horizontal axis of Figure 33 shows the degree of influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in vertical deformation Z. Therefore, the results in Figure 33 show an example of the parameter contribution obtained as a result of strain energy estimation by the design support system 100.

[0188] Figure 34 is the sixth figure showing an example of the results of a performance evaluation experiment in a modified example. Figure 34 is an example of the results of weight estimation by the design support system 100. The vertical axis of Figure 34 represents the shape parameters. The horizontal axis of Figure 34 shows the degree of influence that changes in the values ​​of each shape parameter on the vertical axis have on the change in vertical deformation Z. Therefore, the results in Figure 34 show an example of the parameter contribution obtained as a result of weight estimation by the design support system 100.

[0189] Figure 35 is the seventh figure showing an example of the results of a performance evaluation experiment in a modified example. More specifically, Figure 35 shows an example of the change in the evaluation function value R2 in weight estimation by the design support system 100. The horizontal axis of Figure 35 represents the number of training cycles. The number of training cycles is, for example, the number of trees. The number of trees is the convergence number in the gradient boosting method. The vertical axis of Figure 35 represents the value of the evaluation function value R2. In Figure 35, "train" means the training cycle. In Figure 35, "eval" means the number of trees in the gradient boosting method that the trained mathematical model possesses. Figure 35 shows that "train" and "eval" show good agreement.

[0190] Figure 36 is the eighth figure showing an example of the results of a performance evaluation experiment in a modified example. More specifically, Figure 36 shows an example of the change in the evaluation function value R2 in the estimation of strain energy by the design support system 100. The horizontal axis of Figure 36 shows the number of training cycles. The number of training cycles is, for example, the number of trees. The number of trees is the convergence number in the gradient boosting method. The vertical axis of Figure 36 shows the value of the evaluation function value R2. In Figure 36, "train" means the training cycle. In Figure 36, "eval" means the number of trees in the gradient boosting method that the trained mathematical model has. Figure 36 shows that "train" and "eval" show good agreement.

[0191] Figure 37 is the ninth figure showing an example of the results of a performance evaluation experiment in a modified example. More specifically, Figure 37 shows an example of the change in the evaluation function value R2 in the estimation of strain energy by the design support system 100. The horizontal axis of Figure 37 represents the number of training cycles. The number of training cycles is, for example, the number of trees. The number of trees is the convergence number in the gradient boosting method. The vertical axis of Figure 37 represents the value of the evaluation function value R2. In Figure 37, "train" means the training cycle. In Figure 37, "eval" means the number of trees in the gradient boosting method that the trained mathematical model possesses. Figure 37 shows that "train" and "eval" show good agreement.

[0192] As shown in Figures 29 to 37, the design support system 100's estimations are highly accurate, and it is capable of estimating results that show good agreement with experimental values. Furthermore, the correlation between the appearance parameters and property parameters, for example, is a new finding of the present inventor, obtained by performing a contribution analysis of numerous shape parameters to the property parameters.

[0193] Although embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to this first embodiment and includes designs and the like that do not depart from the spirit of this invention. [Explanation of Symbols]

[0194] 100, 100a, 100b…Design support system, 1, 1a, 1b…Relational information acquisition device, 2, 2a, 2b…Support device, 11, 11a, 11b…Control unit, 12…Communication unit, 13…Input unit, 14…Storage unit, 15…Output unit, 110, 110a, 110b…Training dataset acquisition unit, 120, 120a, 120b…Relational information acquisition unit, 130…Storage control unit, 140…Input control unit, 150…Output control unit, 160…Communication control unit, 20…Output unit, 21, 21a, 21b…Control unit, 22…Communication unit, 23…Input unit, 24…Storage unit, 210, 210b…Analysis target data acquisition unit, 220, 220a, 220b…Contribution estimation unit, 91…Processor, 92...Memory, 93...Processor, 94...Memory

Claims

1. A contribution estimation unit estimates the parameter contribution, which is the degree of influence that changes in the multiple shape parameters have on changes in the properties, using a trained model obtained by a machine learning method that uses a first training dataset containing multiple shape parameters representing the shape of a building as training data and property parameters representing the properties of the building as training data. An output device equipped with the following features.

2. The contribution estimation unit estimates the parameter contribution based on the importance of the features obtained from the trained model, specifically the importance of the features for each of the multiple shape parameters. The output device according to claim 1.

3. A contribution estimation unit estimates the parameter contributions using a trained model obtained by a machine learning method that uses a second training dataset containing multiple shape parameters representing the shape of a building as training data, and parameter contributions, which are the degree to which changes in the multiple shape parameters have an effect on changes in the properties of the building, as training data. An output device.

4. A contribution estimation unit estimates the parameter contributions using a trained model obtained by a machine learning method that uses a third training dataset containing multiple property parameters indicating the properties of a building as training data, and parameter contributions, which are the degree of influence that changes in multiple shape parameters indicating the shape of the building have on changes in the properties of the building, as training data. An output device equipped with the following features.

5. The output control unit causes the parameter contribution estimated by the contribution estimation unit to be output to the controlled object. An output device according to any one of claims 1 to 4, comprising:

6. The output control unit, based on the output control information which is information that instructs the controlled object to output information, causes the controlled object to output information in accordance with the instructions of the output control information. The output device according to claim 5.

7. The maximum distance between the upper and lower chords is defined as depth, and the distance between the column base and the upper chord is defined as rise. The aforementioned set of shape parameters includes at least one of the following: depth, rise, number of main beams, number of bracing members, and number of tensioned beams of the building. The output device according to any one of claims 1 to 6.

8. The property parameters indicating the properties of the aforementioned building represent at least one of the following related to the building: reaction force, deformation, stress, strain energy, and weight. The output device according to any one of claims 1 to 7.

9. An acquisition step to obtain a first training dataset which includes multiple shape parameters indicating the shape of a building as training data and property parameters indicating the properties of the building as training data, A trained model acquisition step is to obtain a trained model using the machine learning method described above with the first training dataset, It has, The trained model is used to estimate the parameter contribution, which is the degree to which changes in the multiple shape parameters have an effect on the changes in the properties. Learning methods.

10. The parameter contribution is the importance of the features obtained from the trained model, and is estimated based on the importance of the features for each of the multiple shape parameters. The learning method according to claim 9.

11. An input assistance step, which is a process that outputs images or sounds prompting the input or selection of values ​​for multiple shape parameters according to predetermined rules, It further possesses, The learning method according to claim 9 or 10.

12. The shape parameters include structural parameters indicating the structure of the building and member parameters relating to the members used in the structure. In the input assistance step, the values ​​of the structural parameter and the member parameter corresponding to the structural parameter are set. The learning method according to claim 11.

13. In the acquisition step, the computer is instructed to input or select values ​​for the multiple shape parameters, the computer is instructed to calculate values ​​for the property parameters corresponding to the input or selected values ​​for the multiple shape parameters, and the first training dataset is acquired, which includes the input or selected values ​​for the multiple shape parameters as training data and the calculated values ​​for the property parameters as training data. The plurality of shape parameters include structural parameters indicating the structure of the building and member parameters relating to the members used in the structure. The learning method according to any one of claims 9 to 11.

14. An acquisition step to obtain a first training dataset which includes multiple shape parameters indicating the shape of a building as training data and property parameters indicating the properties of the building as training data, A pre-trained model generation step in which a pre-trained model is generated by a machine learning method using the first training dataset, It has, The trained model is used to estimate the parameter contribution, which is the degree to which changes in the multiple shape parameters have an effect on the changes in the properties. Learning methods.

15. A program for causing a computer to function as an output device according to any one of claims 1 to 8.

Images