Design support system, training data generation system, and trained model generation system

Abstract

This system provides a design support system suitable for understanding multiple elements being edited sequentially and their editing order. [Solution] The drawing creation support device 100 acquires the created or edited element ID, and uses a trained model, which is trained on learning data including the created or edited element ID, a plurality of edited elements that are edited consecutively after that edited element and element order information relating to their editing order, and evaluation values ​​relating to the plurality of edited elements, which include evaluation values ​​relating to the quality or risk of the building, to estimate the plurality of element order information from the acquired element ID. This makes it possible to understand the plurality of edited elements that are edited consecutively and their editing order. It also makes it possible to understand the element order according to the evaluation values ​​relating to the quality or risk of the building.

Description

Technical Field

[0001] The present invention relates to a design support system, and particularly relates to a design support system, a learning data generation system, and a trained model generation system suitable for grasping a plurality of continuously edited elements and the editing order thereof.

Background Art

[0002] Conventionally, as a technique for supporting design using AI (Artificial Intelligence), for example, the technique described in Patent Document 1 is known.

[0003] The technique described in Patent Document 1 constructs a trained model by assigning a reward R to a combination of a state S determined depending on whether a design activity is executed or not and an action A which is a selectable activity under the state, and maximizing a value. Then, the trained model is used to infer the next action to be performed from the current state.

Prior Art Literature

Patent Literature

[0004]

Patent Document 1

Summary of the Invention

Problem to be Solved by the Invention

[0005] In architectural design, a plurality of elements related to a certain unit may be edited continuously. For example, in the case of toilet design, the design of a toilet bowl and the design of a handwasher are performed successively. However, the technique described in Patent Document 1 infers the next action to be performed from the current state, thus there has been a problem that it cannot grasp the plurality of continuously edited elements and the editing order thereof.

[0006] Therefore, the present invention has been made in view of the unresolved problems of the conventional technology, and aims to provide a design support system, a training data generation system, and a trained model generation system suitable for understanding multiple elements that are edited sequentially and their editing order. [Means for solving the problem]

[0007] [Invention 1] To achieve the above objective, the design support system of Invention 1 comprises: element information acquisition means for acquiring element information relating to created or edited elements in design information; and estimation means for estimating the element order information from the element information acquired by the element information acquisition means, using a trained model trained on training data including element information relating to created or edited elements, a plurality of elements edited consecutively after the element and their editing order (hereinafter, the plurality of elements and their editing order are referred to as the "element order"), and evaluation values ​​relating to the plurality of elements, wherein the evaluation value is an evaluation value relating to the quality or risk of the design object.

[0008] In this configuration, element information is acquired by the element information acquisition means. Then, the estimation means uses a trained model to estimate element order information from the acquired element information.

[0009] Here, a trained model is one that has been trained on training data that includes at least element information, element order information, and evaluation values, and also includes models that have been trained on training data that includes element information, element order information, evaluation values, and other information.

[0010] Furthermore, the element information acquisition means may, for example, input element information from an input device, acquire or receive element information from an external terminal, read element information from a storage device or storage medium, or generate or calculate element information through information processing. Therefore, acquisition includes at least input, acquisition, reception, reading (including retrieval), generation, and calculation. The concept of acquisition remains the same hereafter.

[0011] Furthermore, element information can consist of the element itself, as well as information for identifying the element (e.g., name, number, ID, code, link information such as URL), or as feature information relating to the element's overview, statistics, or other characteristics. Element information can also consist of characters, numbers, figures, codes, symbols, images, sounds, or other information. Additionally, element information can consist of keywords related to the element (e.g., one or more keywords indicating part of the element's name). The same applies hereafter to the trained model generation system of Invention 10.

[0012] Furthermore, element sequence information can consist of, for example, the element sequence itself, or it can be composed of information for identifying elements and editing sequences (e.g., name, number, ID, code, URL or other link information), or it can be composed of feature information relating to an overview, statistics, or other characteristics of elements and editing sequences. Element sequence information can also be composed of, for example, characters, numbers, figures, codes, symbols, images, sounds, or other information. Furthermore, element sequence information can be composed of keywords relating to elements and editing sequences (e.g., one or more keywords indicating part of the names of elements and editing sequences). The same applies hereafter to the learning data generation system of Invention 8 and the trained model generation system of Invention 10.

[0013] Furthermore, this system may be implemented as a single device, apparatus, terminal, or other device, or as a network system in which multiple devices, apparatus, terminals, or other devices are connected in a communicative manner. In the latter case, each component may belong to any of the multiple devices, as long as they are connected in a communicative manner. The same applies hereafter to the learning data generation system of Invention 8 and the trained model generation system of Invention 10.

[0014] [Invention 2] Furthermore, in the design support system of Invention 2, the evaluation value for the quality of the design target is an evaluation value for the safety, functionality, convenience, aesthetics, environmental friendliness, economic efficiency, or legal compliance of the design target.

[0015] [Invention 3] Furthermore, in the design support system of Invention 3, the evaluation value for the risk of the design target is an evaluation value for the risk of the quality of the design target deteriorating, the risk of the design target collapsing or being damaged, the risk of the design target deteriorating, the risk of the cost of the design target increasing, the risk of the design target being affected by natural disasters, the risk of the asset value of the design target decreasing, or the risk of the design target causing social problems.

[0016] [Invention 4] Furthermore, the design support system of Invention 4 is a design support system of any one of Inventions 1 to 3, wherein the trained model is trained to maximize the evaluation value based on training data including the element information, the element sequence information, and the evaluation value.

[0017] [Invention 5] Furthermore, the design support system of Invention 5 is the design support system of Invention 4, wherein the trained model includes a first trained model trained to maximize the evaluation value based on training data including the element information, the element sequence information, and the evaluation value based on a first index indicating an index of the value of the evaluation value, and a second trained model trained to maximize the evaluation value based on training data including the element information, the element sequence information, and the evaluation value based on a second index different from the first index, and comprises an index information acquisition means for acquiring index information relating to the first index or the second index, and a trained model selection means for selecting either the first trained model or the second trained model based on the index information acquired by the index information acquisition means, and the estimation means estimates the element sequence information using the trained model selected by the trained model selection means.

[0018] In this configuration, the index information acquisition means acquires index information, and the trained model selection means selects a trained model based on the acquired index information. Then, the estimation means estimates the element order information using the selected trained model.

[0019] Here, the first trained model only needs to be one that has been trained on training data that includes at least element information, element order information, and evaluation values ​​based on the first metric, and also includes models that have been trained on training data that includes element information, element order information, evaluation values ​​based on the first metric, and other information.

[0020] Furthermore, the second trained model only needs to be one that has been trained on training data that includes at least element information, element order information, and evaluation values ​​based on the second indicator, and also includes models that have been trained on training data that includes element information, element order information, evaluation values ​​based on the second indicator, and other information.

[0021] [Invention 6] Furthermore, the design support system of Invention 6 is the design support system of Invention 1, wherein the plurality of elements are elements whose setting or modification requires human judgment, and whose setting or modification affects other elements.

[0022] [Invention 7] Furthermore, in the design support system of Invention 7, the design information is design information for designing buildings, as described in the design support system of Invention 1.

[0023] [Invention 8] On the other hand, in order to achieve the above objective, the learning data generation system of Invention 8 comprises a calculation means for calculating an evaluation value of element sequence information based on first design information which includes element sequence information relating to a plurality of elements that have been edited in succession and the editing order thereof (hereinafter, "a plurality of elements and the editing order thereof" is referred to as "element sequence") and a first evaluation value has been set, and second design information which includes element sequence information relating to the element sequence and a second evaluation value has been set, and a generation means for generating learning data which includes the element sequence information contained in the first design information and the second design information and the evaluation value calculated by the calculation means.

[0024] With such a configuration, the generation means calculates an evaluation value based on the first design information and the second design information, and the generation means generates learning data including the element order information included in the first design information and the second design information and the calculated evaluation value.

[0025] [Invention 9] Furthermore, the learning data generation system according to Invention 9 is the learning data generation system according to Invention 8, wherein the calculation means calculates the evaluation value of the element order information included in the first design information based on the first evaluation value, calculates the evaluation value of the element order information included in the second design information based on the second evaluation value, and calculates the evaluation value of the element order information commonly included in the first design information and the second design information based on the first evaluation value and the second evaluation value.

[0026] With such a configuration, the calculation means calculates the evaluation value of the element order information included in the first design information based on the first evaluation value. Further, the evaluation value of the element order information included in the second design information is calculated based on the second evaluation value. Furthermore, the evaluation value of the element order information commonly included in the first design information and the second design information is calculated based on the first evaluation value and the second evaluation value.

[0027] [Invention 10] On the other hand, in order to achieve the above object, the trained model generation system according to Invention 10 assigns an evaluation value for a plurality of elements to a combination of element information relating to a created or edited element and element order information relating to a plurality of elements continuously edited next to said element and the editing order thereof (hereinafter, "the plurality of elements and the editing order thereof" is referred to as "element order"), and comprises generation means for generating a trained model by performing learning such that the evaluation value is maximized, wherein the evaluation value is an evaluation value relating to the quality or risk of a design object.

[0028] With such a configuration, an evaluation value is assigned to a combination of element information and element order information, and a trained model is generated by performing learning such that the evaluation value is maximized. [Effects of the Invention]

[0029] As explained above, the design support system of Invention 1 allows for the understanding of multiple elements being edited sequentially and their editing order. Furthermore, it allows for the understanding of the element order according to evaluation values ​​related to the quality or risk of the design target.

[0030] Furthermore, according to the design support system of Invention 4, it is possible to identify the sequence of elements with high evaluation values ​​regarding the quality or risk of the design target.

[0031] Furthermore, according to the design support system of Invention 5, it is possible to identify the element order with the highest evaluation value based on the first or second indicator.

[0032] Furthermore, according to the design support system of Invention 6, it is possible to grasp the order of elements, taking into account the relationship between elements that require human judgment for setting or changing and other elements that are affected by that setting or changing.

[0033] On the other hand, according to the learning data generation system of Invention 8, learning data including element sequence information and its evaluation value can be generated based on design information in which evaluation values ​​have been set.

[0034] On the other hand, according to the trained model generation system of invention 10, it is possible to obtain a trained model that has an element order with a high evaluation value. [Brief explanation of the drawing]

[0035] [Figure 1] This figure shows the hardware configuration of the drawing creation support device 100. [Figure 2] This diagram shows the structure of CAD data for a cross-section drawing. [Figure 3] This diagram shows the structure of the edit history data. [Figure 4] This is a flowchart showing the training data generation process. [Figure 5] This diagram shows the structure of the training data. [Figure 6] This is a flowchart showing the process of generating a pre-trained model. [Figure 7] This block diagram shows the process of generating and using a pre-trained model. [Figure 8] This is a flowchart showing the process for estimating element order information. [Figure 9] This is a cross-section drawing that anticipates the delivery of a piano. [Figure 10] This diagram shows the structure of the edit history data. [Figure 11] This diagram shows the structure of the training data. [Figure 12] This is a flowchart showing the process for estimating element order information. [Figure 13] This diagram illustrates how to calculate the evaluation value of element sequence information from the evaluation value of a building. [Figure 14] This diagram shows three pre-trained models, 460-464, connected in parallel. [Figure 15] This diagram shows three pre-trained models, 470-474, connected in series. [Figure 16] This is a block diagram showing the configuration of the network system according to this embodiment. [Figure 17] This is a functional block diagram of the generation AI server 120. [Figure 18] This is a flowchart showing the training data registration process. [Figure 19] This is a flowchart showing the process for obtaining element order information. [Figure 20] This is a flowchart showing the process for obtaining element order information. [Modes for carrying out the invention]

[0036] [First Embodiment] The first embodiment of the present invention will be described below. Figures 1 to 9 show this embodiment.

[0037] [Configuration of this embodiment] First, the configuration of this embodiment will be described. Figure 1 shows the hardware configuration of the drawing creation support device 100.

[0038] As shown in Figure 1, the drawing creation support device 100 consists of a CPU (Central Processing Unit) 30 that controls calculations and the entire system based on a control program, a ROM (Read Only Memory) 32 that stores the control program for the CPU 30 in a predetermined area, a RAM (Random Access Memory) 34 for storing data read from the ROM 32 and other memory, as well as calculation results necessary for the calculation process of the CPU 30, and an I / F (Interface) 38 that mediates data input and output to external devices. These components are connected to each other and enable data exchange via a bus 39, which is a signal line for data transfer.

[0039] I / F38 is connected to an external device, which includes an input device 40 consisting of a keyboard and mouse that can input data as a human interface, a storage device 42 that stores data and tables as files, and a display device 44 that displays a screen based on an image signal.

[0040] The storage device 42 has CAD (Computer-Aided Design) software and BIM (Building Information Modeling) software (hereinafter collectively referred to as "CAD software") installed on it. CAD software is software that assists in the creation of drawings according to the designer's operations. When the startup of CAD software is requested, the CPU 30 starts the program for the CAD software stored in a predetermined area of ​​the ROM 32 and executes processing according to that program. The designer can start the CAD software and create cross-section drawings, floor plan details, and other architectural drawings.

[0041] Next, we will explain the data structure of the storage device 42. The storage device 42 stores CAD data of cross-section drawings, detailed floor plans, and other architectural drawings.

[0042] Figure 2 shows the structure of the CAD data for the cross-section drawing. As shown in Figure 2, the CAD data for a cross-section drawing is data that constitutes a drawing that depicts a detailed cross-section of a building, and is composed of data that includes one or more createable or editable elements (hereinafter referred to as "editable elements"). The CAD data for a cross-section drawing is created by the designer using CAD software. The designer creates the cross-section drawing by creating, setting, changing, or deleting (hereinafter referred to as "editing") the editable elements in the CAD software. In the example in Figure 2, the editable elements for the floor, walls, and ceiling of the area labeled "internal corridor" and the editable elements for the floor, walls, and ceiling of the area labeled "vestibule" are respectively placed.

[0043] The same applies to CAD data for floor plans and other architectural drawings, which are composed of data containing one or more editing elements.

[0044] The storage device 42 stores editing history data for each CAD data set, showing the history of editing the editing elements. The CAD data reflects the final editing results related to the editing history data.

[0045] Figure 3 shows the structure of the editing history data. As shown in Figure 3, the editing history data includes, for each edited element, element information 400 related to that element and the editing time 402 required for editing that element, in the order of editing. The element information 400 includes the element ID for identifying the edited element, the area to be edited by the edited element, and the edited element itself. The editing time 402 can be calculated, for example, by subtracting the editing start time from the editing end time.

[0046] In the example in Figure 3, the first related group of editing elements for the toilet—"toilet bowl," "handwashing counter," "mirror," and "towel rack"—are edited in that order consecutively. These editing elements are assigned element IDs "52," "53," "54," and "55," and the editing times are 35, 38, 70, and 94 minutes, respectively.

[0047] Furthermore, a second related group is shown, indicating that the editing elements of the toilet—"ventilation fan," "lighting," "storage," and "outlet"—were edited in that order consecutively. These editing elements were assigned element IDs "56," "57," "58," and "59," and the editing times were 94, 48, 42, and 74 minutes, respectively.

[0048] Furthermore, a third related group is shown, indicating that the closet editing elements "shelf board," "hanger pipe," "drawer," and "basket" are edited in that order consecutively. These editing elements are assigned element IDs "60," "61," "62," and "63," and the editing times were 96, 74, 84, and 22 minutes, respectively.

[0049] Furthermore, a fourth related group is shown, indicating that the kitchen editing elements "flooring," "wall covering," "countertop," and "sink" are edited in that order consecutively. These editing elements are assigned element IDs "64," "65," "66," and "67," and the editing times were 42, 44, 35, and 32 minutes, respectively.

[0050] Since the editing history data is used to create training data, the storage device 42 stores a large amount of editing history data that has been created in the past.

[0051] [Operation of this embodiment] Next, the operation of this embodiment will be described. [Training data generation process] Figure 4 is a flowchart showing the training data generation process.

[0052] The training data generation process is a process that generates training data, and when it is executed on the CPU 30, it proceeds to step S100, as shown in Figure 4.

[0053] In step S100, the unprocessed editing history data is retrieved from the storage device 42, the process moves to step S102, the variable n is set to "2", and the process moves to step S104.

[0054] In steps S104 to S108, the element IDs of created or edited editing elements (hereinafter abbreviated as "created or edited element IDs"), the n editing elements (the number indicated by the value of variable n) that were edited consecutively after that editing element, their editing order, and their editing times are obtained from the editing history data acquired in step S100. Hereafter, multiple editing elements and their editing order may be referred to as "element order". Using Figure 3 as an example, we will explain the case when the value of variable n is "2". In the editing history data in Figure 3, each row is arranged in editing order.

[0055] When the second row is targeted, the element IDs "01" to "51" of previously edited elements are obtained as created or edited element IDs. Since the value of variable n is "2", "toilet" and "handwashing counter" are obtained as edited elements, and "35" and "38" are obtained as editing times, respectively.

[0056] When the third row is targeted, the element IDs "01" to "52" of the previously edited elements are obtained as created or edited element IDs. Since the value of variable n is "2", "Handwashing Counter" and "Mirror" are obtained as edited elements, and "38" and "70" are obtained as editing times, respectively.

[0057] Next, the process moves to step S110, where the evaluation value is calculated by multiplying the sum of the editing times obtained in step S108 by "-1". In the example in the second line above, editing times of "35" and "38" are obtained, so the evaluation value is calculated as (35 + 38) × -1 = -73. The reason for multiplying by "-1" is to set a higher evaluation value for shorter editing times.

[0058] Next, the process moves to step S112, where the element IDs and element order obtained in steps S104 to S108, along with the evaluation value calculated in step S110, are registered as training data. Then, the process moves to step S114, where "1" is added to the value of variable n, and finally, the process moves to step S116.

[0059] In step S116, it is determined whether the value of variable n is greater than "4". If it is determined to be less than or equal to "4" (NO), the process proceeds to step S104. Then, steps S104 to S114 are repeated until the value of variable n becomes "4".

[0060] Using Figure 3 as an example, the process from steps S104 to S110 will be explained for the case where the value of variable n is "3".

[0061] When the second row is targeted, the element IDs of previously edited elements, from "01" to "51", are obtained as created or edited element IDs. Since the value of variable n is "3", "toilet," "handwashing counter," and "mirror" are obtained as edited elements, and "35," "38," and "70" are obtained as editing times. The evaluation value is calculated as (35 + 38 + 70) × -1 = -143.

[0062] When the third row is targeted, the element IDs "01" to "52" of the previously edited elements are obtained as created or edited element IDs. Since the value of variable n is "3", "Handwashing Counter", "Mirror", and "Towel Rack" are obtained as edited elements, and "38", "70", and "94" are obtained as editing times. The evaluation value is calculated as (38 + 70 + 94) × -1 = -202.

[0063] Furthermore, using Figure 3 as an example, the process of steps S104 to S110 will be explained for the case where the value of variable n is "4".

[0064] When the second row is targeted, the element IDs of previously edited elements, from "01" to "51", are obtained as created or edited element IDs. Since the value of variable n is "4", "toilet," "handwashing counter," "mirror," and "towel rack" are obtained as edited elements, and "35," "38," "70," and "94" are obtained as editing times. The evaluation value is calculated as (35 + 38 + 70 + 94) × -1 = -237.

[0065] When the third row is targeted, the element IDs "01" to "52" of previously edited elements are obtained as created or edited element IDs. Since the value of variable n is "4", "handwashing counter", "mirror", "towel rack", and "ventilation fan" are obtained as edited elements, and "38", "70", "94", and "94" are obtained as editing times. The evaluation value is calculated as (38 + 70 + 94 + 94) × -1 = -296.

[0066] On the other hand, if it is determined in step S116 that the value of variable n is greater than "4" (YES), the process proceeds to step S118 to determine whether the processing in steps S100 to S116 has been completed for all edit history data. If it is determined that the processing has been completed for all edit history data (YES), the process proceeds to step S120.

[0067] In step S120, the learning data in which element IDs and other information were registered in step S112 is stored in the storage device 42.

[0068] Figure 5 shows the structure of the training data. As shown in Figure 5, the training data contains, for each row, the created or edited element ID 410, element order information 412, and evaluation value 414. The element order information 412 includes the area and element order that will be edited for the edited element.

[0069] The second row of Figure 5 shows two editing elements that were edited consecutively after the editing element in the first row of Figure 3, along with their editing order and evaluation value. The third row of Figure 5 shows two editing elements that were edited consecutively after the editing element in the second row of Figure 3, along with their editing order and evaluation value. The fourth row of Figure 5 shows two editing elements that were edited consecutively after the editing element in the third row of Figure 3, along with their editing order and evaluation value. Furthermore, the sixth row of Figure 5 shows two editing elements that were edited consecutively after the editing element in the fifth row of Figure 3, along with their editing order and evaluation value. The seventh row of Figure 5 shows two editing elements that were edited consecutively after the editing element in the sixth row of Figure 3, along with their editing order and evaluation value. The eighth row of Figure 5 shows two editing elements that were edited consecutively after the editing element in the seventh row of Figure 3, along with their editing order and evaluation value.

[0070] Furthermore, the 10th row of Figure 5 shows the three editing elements that were edited consecutively after the editing element in the 1st row of Figure 3, along with their editing order and evaluation value. The 11th row of Figure 5 shows the three editing elements that were edited consecutively after the editing element in the 2nd row of Figure 3, along with their editing order and evaluation value. Additionally, the 13th row of Figure 5 shows the three editing elements that were edited consecutively after the editing element in the 5th row of Figure 3, along with their editing order and evaluation value. The 14th row of Figure 5 shows the three editing elements that were edited consecutively after the editing element in the 6th row of Figure 3, along with their editing order and evaluation value.

[0071] Furthermore, row 16 of Figure 5 shows the four editing elements that were edited consecutively after the editing element in row 1 of Figure 3, along with their editing order and evaluation values. Similarly, row 18 of Figure 5 shows the four editing elements that were edited consecutively after the editing element in row 5 of Figure 3, along with their editing order and evaluation values.

[0072] [Trained model generation process] Figure 6 is a flowchart showing the process of generating a trained model.

[0073] Figure 7 is a block diagram showing the process of generating and using a trained model. The trained model generation process is a process performed to generate a trained model. When executed on the CPU 30, it proceeds to step S200, as shown in Figure 6, to perform the training data analysis process. In the training data analysis process, training data is read from the storage device 42, and the created or edited element ID 410, element sequence information 412, and evaluation value 414 are extracted from the read training data.

[0074] Next, the process moves to step S202. In step S202, as shown in Figure 7, a training dataset is generated based on the information extracted in step S200. The process then moves to step S204, where the generated training dataset is input into the training program, and the training program generates a trained model. The training program includes pre-training parameters and hyperparameters, and performs training based on the input training dataset and hyperparameters, updating the pre-training parameters. As a training method, for example, reinforcement learning (e.g., supervised reinforcement learning, imitation learning) can be employed. In reinforcement learning, an evaluation value is assigned to the editing of multiple editing elements for combinations of created or edited element IDs and element order information relating to multiple editing elements that were edited consecutively after that editing element, and training is performed to maximize the evaluation value. Finally, a trained model is output as the training result.

[0075] The trained model is trained to maximize the evaluation value 414 based on the created or edited element ID 410, element order information 412, and evaluation value 414. The trained model comprises trained parameters, which are updated from pre-training parameters, and an inference program. The inference program takes the created or edited element ID as input, estimates the element order information from the input element ID based on the trained parameters, and outputs the estimated element order information. Note that the relationship between the input element ID and the output element order information is determined by the AI's training, so while it shows a similar trend to the content of past training data, there is an ambiguity that prevents it from being an exact match. However, this ambiguity can be reduced by increasing the amount of training data and the training accuracy.

[0076] Next, the process moves to step S206, where the trained model generated in step S204 is stored in the memory device 42, and the series of processes ends.

[0077] [Element Order Information Estimation Process] Figure 8 is a flowchart showing the process for estimating element order information.

[0078] The element sequence information estimation process is performed in response to requests from designers or other users. When executed on the CPU 30, it first proceeds to step S300, as shown in Figure 8.

[0079] In step S300, the element IDs of created or edited elements are obtained from the CAD data currently being edited, and the process proceeds to step S302.

[0080] In step S302, the trained model in the memory device 42 is used to estimate multiple element order information with different editing elements or editing orders from the element ID obtained in step S300. Estimation is performed by inputting the element ID into the trained model and obtaining the element order information output from the trained model. The trained model may be configured to obtain multiple outputs from one input, or it may be configured to obtain multiple outputs by repeating the input and output of one multiple times.

[0081] Next, the process moves to step S304, where, based on the multiple element order information estimated in step S302, one of the multiple element orders is displayed on the display device 44 using one of the multiple display rules. The choice of which display rule to use may be set, for example, by the designer or other user, or by a predetermined algorithm.

[0082] The first display rule is to display the element order with the most editable elements among the estimated multiple element orders.

[0083] The second display rule is to display element sequences that form related groups among the estimated multiple element sequences. For example, if the estimated multiple element sequences from the current editing state are (1) A, B, (2) A, B, C, (3) A, B, C, D, (4) A, B, C, D, E, (5) A, B, C, D, E, F, and A to D form a related group, then according to the second display rule, one of (3) to (5) will be displayed. Whether or not a group is related can be determined, for example, by identifying two or more editing elements and their editing sequences from A to F that appear more than a predetermined number of times in the training data. The same applies to the third and fourth display rules below.

[0084] The third display rule is to display the same element order as a related group among the multiple estimated element orders. For example, if the multiple element orders estimated from the current editing state are (1) to (5) above, according to the third display rule, (3) will be displayed.

[0085] The fourth display rule is to display related groups and related groups from among the estimated multiple element orders, including other editing elements. For example, if the multiple element orders estimated from the current editing state are (1), (2), and (5) above, according to the fourth display rule, A to D will be displayed starting from (5).

[0086] Next, the process moves to step S306. If the designer creates or edits an editable element based on the displayed element order, the created or edited element IDs, including the element ID of that editable element, are obtained from the currently edited CAD data, and the process moves to step S308.

[0087] In step S308, based on the element order displayed in steps S304 and S314 and the element ID obtained in step S306, it is determined whether the edited element created or edited by the designer differs from the edited element or editing order displayed in steps S304 and S314. If it is determined that the editing result differs from the estimated result (YES), the process proceeds to step S310.

[0088] In step S310, the display rules used for display in steps S304 and S314 are changed. For example, the display rules are changed to the optimal ones so that the edited results match the estimated results. The timing of the change in display rules does not have to be after each creation or edit, but may be after multiple creations or edits.

[0089] Next, the process moves to step S312, where, similar to the process in step S302, the trained model in the storage device 42 is used to estimate multiple element order information from the element IDs obtained in step S306, and then the process moves to step S314.

[0090] In step S314, similar to the process in step S304, one of the multiple element order information estimated in step S312 is displayed on the display device 44, and the process proceeds to step S316.

[0091] In step S316, it is determined whether the designer has finished editing. If it is determined that editing is complete (YES), the series of processes is terminated.

[0092] On the other hand, if it is determined in step S316 that the designer has not finished editing (NO), the process proceeds to step S306.

[0093] On the other hand, if it is determined in step S308 that the editing result matches the estimated result (NO), the process proceeds to step S312.

[0094] [When considering the delivery of a piano] Next, we will explain the procedure when considering the delivery of a piano.

[0095] Figure 9 is a cross-section drawing that shows the planned delivery of a piano. If a designer wants to install a piano with a width of 120 mm in the cross-section drawing shown in Figure 9 using CAD software, they need to shorten the width of the toilet to prevent interference between the piano and the toilet wall during transport. However, changing the width of the toilet necessitates editing other elements of the toilet. Therefore, if the designer requests the estimation of the order of elements to be edited for the toilet after changing its width, the system will go through steps S300 to S304, retrieve the IDs of created or edited elements from the currently edited CAD data, and estimate and display the element order from the retrieved element IDs. For example, if the element order information for row 16 in Figure 5 is estimated, "toilet bowl → handwashing counter → mirror → towel rack" will be displayed. If these editing elements are already included in the CAD data as shown in Figure 9, they will be highlighted (for example, displayed in a specific color or pattern) and connected with arrows to display the editing order. If they are not included in the CAD data, they will be displayed as ghost images (for example, a wireframe of the ghost image of the editing element) and connected with arrows to display the editing order.

[0096] Then, when the designer creates or edits the "toilet" element based on the estimated and displayed element order, the next element order is estimated and displayed via steps S306 to S314. If the designer's editing result differs from the estimated result, the display rule is changed via step S310.

[0097] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, created or edited element IDs are obtained, and a trained model is used to estimate multiple element order information from the obtained element IDs.

[0098] This allows you to understand multiple editing elements being edited sequentially and their editing order.

[0099] Furthermore, in this embodiment, multiple element sequence information with different editing elements or editing orders is estimated, and one of the multiple element sequences is displayed based on the estimated multiple element sequence information.

[0100] This allows for the identification of multiple possible editing sequences for several elements and their editing order.

[0101] Furthermore, in this embodiment, the element order with the largest number of editable elements among the multiple estimated element orders is displayed.

[0102] This allows designers to edit while visualizing larger editing units. Furthermore, in this embodiment, element sequences that appear more than a predetermined number of times in the training data are identified as related groups, and the element sequences that have related groups among the multiple estimated element sequences are displayed.

[0103] This allows designers to edit while visualizing related groups of elements.

[0104] Furthermore, in this embodiment, element sequences that appear more than a predetermined number of times in the training data are identified as related groups, and element sequences identical to those of the related groups are displayed from among the multiple estimated element sequences.

[0105] This allows designers to edit while visualizing related groups of elements.

[0106] Furthermore, in this embodiment, element sequences that appear more than a predetermined number of times in the training data are identified as related groups, and from among the multiple estimated element sequences, the portion containing related groups and other editing elements is displayed.

[0107] This allows designers to edit while visualizing related groups of elements.

[0108] Furthermore, in this embodiment, based on the displayed element order, the element IDs of the edited elements created or edited by the designer are obtained, and multiple element order information is estimated from the obtained element IDs using the trained model.

[0109] This allows the element order to be estimated and displayed for the result of creation or editing based on the displayed element order, making it possible to understand multiple editing elements being edited consecutively and their editing order.

[0110] Furthermore, in this embodiment, the element IDs of created or edited elements, including the element IDs of edited elements created or edited by the designer, are obtained based on the displayed element order, and the display rules are changed based on the displayed element order and the obtained element IDs.

[0111] This allows you to change the display rules according to the result of the element order display and the result of subsequent creation or editing.

[0112] Furthermore, in this embodiment, the trained model is trained to maximize the evaluation value based on training data that includes created or edited element IDs, element order information relating to a plurality of edited elements that are edited consecutively after that edited element and their editing order, and evaluation values ​​relating to the editing of the plurality of edited elements.

[0113] This allows us to identify the order of elements that receive the highest evaluation scores. Furthermore, in this embodiment, an evaluation value is assigned to a combination of a created or edited element ID and element order information relating to a plurality of edited elements that were edited consecutively after that edited element, and the editing order of those elements. A trained model is generated by training the model to maximize the evaluation value.

[0114] This allows us to obtain a trained model that yields a high evaluation score for element order. [Second Embodiment] Next, a second embodiment of the present invention will be described. Figures 10 to 12 show this embodiment. Figures 3, 5, 6 and 9 will also be referenced.

[0115] This embodiment differs from the first embodiment in that it performs estimation using a first pre-trained model and a second pre-trained model, each trained using evaluation values ​​based on different indicators. Below, only the parts that differ from the first embodiment will be described, and the overlapping parts will be omitted.

[0116] [Configuration of this embodiment] First, the configuration of this embodiment will be described. Figure 10 shows the structure of the editing history data.

[0117] The storage device 42 stores the editing history data shown in Figure 3, as well as the editing history data shown in Figure 10, for each CAD data file.

[0118] As shown in Figure 10, the editing history data includes, for each edited element, element information 420 related to that element and the number of editing items 422 required to edit that element, in the order of editing.

[0119] In the example in Figure 10, the first related group of editing elements for the toilet, "toilet bowl," "handwashing counter," "mirror," and "towel rack," are edited in that order consecutively. These editing elements are assigned element IDs "52," "53," "54," and "55," and the number of editable items is 8, 5, 6, and 9, respectively.

[0120] Furthermore, a second related group is shown, indicating that the editing elements for the toilet—"ventilation fan," "lighting," "storage," and "outlet"—are edited in that order consecutively. These editing elements are assigned element IDs "56," "57," "58," and "59," with the number of editable items being 3, 7, 8, and 6 respectively.

[0121] Furthermore, a third related group is shown, indicating that the closet editing elements "shelves," "hanger pipes," "drawers," and "baskets" are edited in that order consecutively. These editing elements are assigned element IDs "60," "61," "62," and "63," with the number of editable items being 8, 7, 2, and 4 respectively.

[0122] Furthermore, a fourth related group is shown, indicating that the kitchen editing elements "flooring," "wall covering," "countertop," and "sink" are edited in that order consecutively. These editing elements are assigned element IDs "64," "65," "66," and "67," with 7, 4, 3, and 8 editing items respectively.

[0123] [Operation of this embodiment] Next, the operation of this embodiment will be described. [Training data generation process] In the training data generation process, training data is generated based on the editing history data in Figure 10, similar to the training data generation in Figure 5.

[0124] Figure 11 shows the structure of the training data. As shown in Figure 11, the training data includes, for each row, the created or edited element ID 430, element order information 432, and evaluation value 434.

[0125] The second row of Figure 11 shows two editing elements that were edited consecutively after the editing element in the first row of Figure 10, along with their editing order and evaluation value. The third row of Figure 11 shows two editing elements that were edited consecutively after the editing element in the second row of Figure 10, along with their editing order and evaluation value. The fourth row of Figure 11 shows two editing elements that were edited consecutively after the editing element in the third row of Figure 10, along with their editing order and evaluation value. Furthermore, the sixth row of Figure 11 shows two editing elements that were edited consecutively after the editing element in the fifth row of Figure 10, along with their editing order and evaluation value. The seventh row of Figure 11 shows two editing elements that were edited consecutively after the editing element in the sixth row of Figure 10, along with their editing order and evaluation value. The eighth row of Figure 11 shows two editing elements that were edited consecutively after the editing element in the seventh row of Figure 10, along with their editing order and evaluation value.

[0126] Furthermore, the 10th row of Figure 11 shows the three editing elements that were edited consecutively after the editing element in the 1st row of Figure 10, along with their editing order and evaluation value. The 11th row of Figure 11 shows the three editing elements that were edited consecutively after the editing element in the 2nd row of Figure 10, along with their editing order and evaluation value. Additionally, the 13th row of Figure 11 shows the three editing elements that were edited consecutively after the editing element in the 5th row of Figure 10, along with their editing order and evaluation value. The 14th row of Figure 11 shows the three editing elements that were edited consecutively after the editing element in the 6th row of Figure 10, along with their editing order and evaluation value.

[0127] Furthermore, row 16 of Figure 11 shows the four editing elements that were edited consecutively after the editing element in row 1 of Figure 10, along with their editing order and evaluation values, and row 18 of Figure 11 shows the four editing elements that were edited consecutively after the editing element in row 5 of Figure 10, along with their editing order and evaluation values.

[0128] [Trained model generation process] In the pre-trained model generation process, a first pre-trained model is generated by performing training based on the training data shown in Figure 5, following steps S200 to S204. The first pre-trained model is the same as the pre-trained model in the first embodiment described above.

[0129] In the trained model generation process, following steps S200 to S204, a second trained model is generated by training based on the training data shown in Figure 11, similar to the generation of the first trained model. The second trained model is trained to maximize the evaluation value 434 based on the created or edited element ID 430, element order information 432, and evaluation value 434.

[0130] Then, the process moves to step S206, where the first trained model and the second trained model generated in step S204 are stored in the memory device 42, and the series of processes ends.

[0131] [Element Order Information Estimation Process] Figure 12 is a flowchart showing the element order information estimation process.

[0132] The element sequence information estimation process is performed in response to requests from designers or other users. When executed on the CPU 30, it first proceeds to step S330, as shown in Figure 12.

[0133] In step S330, indicator information is obtained regarding the first indicator "editing time" or the second indicator "number of edited items," which indicate the value of the evaluation value. The process then proceeds to step S332, where the first trained model is selected if the indicator related to the obtained indicator information is the first indicator, and the second trained model is selected if the indicator related to the obtained indicator information is the second indicator.

[0134] Next, the process moves to step S334, where the element IDs of created or edited elements are obtained from the CAD data currently being edited, and then the process moves to step S336.

[0135] In step S336, the trained model selected in step S332 from the first and second trained models in the storage device 42 (hereinafter referred to as the "selected trained model") is used to estimate multiple element order information from the element IDs obtained in step S334. The estimation method is the same as the process in step S302 in the first embodiment described above.

[0136] Next, the process moves to step S338, where, similar to the process in step S304, one of the multiple element order information estimated in step S336 is displayed on the display device 44, and the process moves to step S340.

[0137] In step S340, if the designer creates or edits an editable element based on the displayed element order, the created or edited element IDs, including the element ID of that editable element, are obtained from the currently edited CAD data, and the process proceeds to step S342.

[0138] In step S342, based on the element order displayed in steps S338 and S348 and the element ID obtained in step S340, it is determined whether the edited element created or edited by the designer differs from the edited element or editing order displayed in steps S338 and S348. If it is determined that the editing result differs from the estimated result (YES), the process proceeds to step S344.

[0139] In step S344, similar to the process in step S310, the display rules used for display in steps S338 and S348 are changed, and the process proceeds to step S346.

[0140] In step S346, similar to the process in step S336, the selected and trained model is used to estimate multiple element order information from the element IDs obtained in step S340, and then the process proceeds to step S348.

[0141] In step S348, similar to the process in step S338, one of the multiple element order information estimated in step S346 is displayed on the display device 44, and the process proceeds to step S350.

[0142] In step S350, it is determined whether the designer has finished editing. If it is determined that editing is complete (YES), the series of processes is terminated.

[0143] On the other hand, if it is determined in step S350 that the designer has not finished editing (NO), the process proceeds to step S340.

[0144] On the other hand, if it is determined in step S342 that the editing result matches the estimated result (NO), the process proceeds to step S346.

[0145] [When considering the delivery of a piano] Next, we will explain the procedure when considering the delivery of a piano.

[0146] If a designer wants to install a piano with a width of 120 mm in the cross-section drawing shown in Figure 9 using CAD software, they need to shorten the width of the toilet to prevent interference between the piano and the toilet wall during transport. However, changing the width of the toilet necessitates editing other elements of the toilet. Therefore, the designer requests an estimation of the order in which elements should be edited for the toilet after changing its width. If the designer wants to obtain an element order that minimizes editing time, they select "editing time" as the indicator, and after steps S330 to S332, the first trained model is selected. Then, after steps S334 to S338, the IDs of created or edited elements are obtained from the currently edited CAD data, and the element order is estimated and displayed by the first trained model based on the obtained element IDs.

[0147] In contrast, if the designer wants to obtain an element order that minimizes the number of editing items, they select "number of editing items" as the indicator, and after steps S330 to S332, the second trained model is selected. Then, after steps S334 to S338, the element IDs of created or edited elements are obtained from the CAD data currently being edited, and the second trained model estimates and displays the element order from the obtained element IDs.

[0148] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, index information relating to the first or second index is acquired, either the first trained model or the second trained model is selected based on the acquired index information, the created or edited element IDs are acquired, and multiple element sequence information is estimated from the acquired element IDs using the selected trained model.

[0149] This allows us to identify the order of elements that have high evaluation values ​​based on the first or second indicator.

[0150] [Third Embodiment] Next, a third embodiment of the present invention will be described. Figure 13 shows this embodiment.

[0151] This embodiment differs from the first and second embodiments in that it employs evaluation values ​​other than editing time or the number of editing items. Below, only the parts that differ from the first and second embodiments will be described, and the overlapping parts will be omitted from the explanation.

[0152] [Configuration of this embodiment] First, the configuration of this embodiment will be described. As evaluation values ​​to be included in the training data, we will adopt evaluation values ​​related to the quality or risk of buildings.

[0153] As qualities of a building, for example, the following can be considered: (1) safety of the building (e.g., structural safety, fire safety, safety for use), (2) functionality or convenience of the building (e.g., livability, comfort), (3) aesthetics of the building (e.g., design), (4) environmental friendliness of the building (e.g., reduction of environmental impact, energy saving), (5) economic efficiency of the building (e.g., construction costs, maintenance costs, life cycle costs, and other costs), and (6) legal compliance of the building (e.g., the degree to which it conforms to laws and standards, or whether it conforms or not).

[0154] Risks associated with buildings may include, for example, (1) the risk of a decline in the quality of the building (e.g., safety, functionality, convenience, aesthetics, environmental friendliness, economic efficiency, or legal compliance), (2) the risk of the building collapsing or being damaged due to insufficient strength, durability, seismic resistance, etc., (3) the risk of the building deteriorating due to aging, etc., (4) the risk of increased costs related to the building, (5) risks to the building caused by fire, flood, windstorm, earthquake, or other natural disasters, (6) the risk of a decrease in the asset value of the building, and (7) the risk of the building causing social problems.

[0155] The quality or risk of a building can be evaluated as a numerical value within a predetermined range (e.g., 0 to 10), and this can be used as the evaluation value. Furthermore, it is preferable to standardize the evaluation values ​​for each indicator (mean value 0, standard deviation 1) so that evaluation values ​​based on different indicators can be compared to a similar degree. The evaluation value for the quality or risk of a building may be set for the element sequence information as in the first and second embodiments described above, or it may be set for the building being designed. When setting an evaluation value for a building, the learning data requires an evaluation value corresponding to each element sequence information, so the evaluation value of the element sequence information related to the design of that building is calculated from the evaluation value for the building.

[0156] Figure 13 is a diagram illustrating the process of calculating the evaluation value of element sequence information from the evaluation value of a building.

[0157] In the example shown in Figure 13, there are three CAD data sets, 450-454, and one evaluation value is assigned to each CAD data set as a whole. It should be assumed that the evaluation value is standardized.

[0158] CAD data 450 is CAD data related to the design of building A, and an evaluation value for index 1 (e.g., building safety) is set for building A. For example, for CAD data 450, the evaluation value based on index 1 is set to "0.1".

[0159] CAD data 452 is CAD data related to the design of building B, and an evaluation value for index 2 (e.g., the functionality of the building) is set for building B. For example, for CAD data 452, an evaluation value of "0.2" is set based on index 2.

[0160] CAD data 454 is CAD data related to the design of building C, and an evaluation value based on indicator 3 (e.g., building risk) is set for building C. For example, the evaluation value based on indicator 3 for CAD data 454 is set to "0.3".

[0161] CAD data 450-454 contains multiple elements of sequence information. In step S110, the evaluation value is calculated as "0.1" for the elements of sequence information contained only in CAD data 450, "0.2" for the elements of sequence information contained only in CAD data 452, and "0.3" for the elements of sequence information contained only in CAD data 454. Furthermore, the evaluation value is calculated as 0.1 + 0.2 = "0.3" for the elements of sequence information contained in common in CAD data 450 and 452, 0.2 + 0.3 = "0.5" for the elements of sequence information contained in common in CAD data 452 and 454, and 0.1 + 0.3 = "0.4" for the elements of sequence information contained in common in CAD data 450 and 454. Furthermore, the evaluation value is calculated as 0.1 + 0.2 + 0.3 = "0.6" for the elements of sequence information contained in common in CAD data 450-454.

[0162] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, a created or edited element ID is obtained, and a trained model, which has been trained on training data including the created or edited element ID, a plurality of edited elements that were edited consecutively after that edited element and element sequence information relating to the editing order, and evaluation values ​​relating to the plurality of edited elements, which include evaluation values ​​relating to the quality or risk of the building, is used to estimate the plurality of element sequence information from the obtained element ID.

[0163] This allows for the tracking of multiple editing elements being edited sequentially and their editing order. Furthermore, it enables the tracking of element order based on evaluation values ​​related to the quality or risk of the building.

[0164] Furthermore, in this embodiment, the trained model is trained to maximize the evaluation value based on training data including element IDs, element order information, and evaluation values.

[0165] This makes it possible to identify the order of elements that have high evaluation scores regarding the quality or risk of a building.

[0166] In this embodiment, step S110 corresponds to the calculation means of invention 8 or 9, step S120 corresponds to the generation means of invention 8 or 10, steps S300, S306, S334, and S340 correspond to the element information acquisition means of invention 1, and steps S302, S312, S336, and S346 correspond to the estimation means of invention 1 or 5. Furthermore, step S330 corresponds to the index information acquisition means of invention 5, step S332 corresponds to the model selection means of invention 5, the CAD data corresponds to the design information of inventions 7 to 9, and the created or edited element IDs correspond to the element information of inventions 1, 4, 5, or 10.

[0167] [Fourth Embodiment] Next, a fourth embodiment of the present invention will be described. Figure 14 is a diagram showing this embodiment. Figures 5 and 11 will also be referenced.

[0168] This embodiment differs from the third embodiment in that it estimates element order information using multiple trained models, each trained based on evaluation values ​​derived from different indicators. Below, only the parts that differ from the first to third embodiments will be described, and the overlapping parts will be omitted.

[0169] [Configuration of this embodiment] First, the configuration of this embodiment will be described. Figure 14 shows three pre-trained models 460-464 connected in parallel.

[0170] As shown in Figure 14, the memory device 42 stores three trained models 460 to 464.

[0171] In the training data shown in Figure 5 or Figure 11, training data 1 is prepared with evaluation values ​​based on a first indicator (for example, building safety). The trained model 460 is trained based on training data 1. The trained model 460 takes an element ID as input, estimates one or more element sequence information with high evaluation values ​​based on the first indicator from the input element ID, and outputs the estimated element sequence information.

[0172] In the training data shown in Figure 5 or Figure 11, training data 2 is prepared with evaluation values ​​based on a second indicator (for example, the functionality of the building). The trained model 462 is trained based on training data 2. The trained model 462 takes an element ID as input, estimates one or more element order information with high evaluation values ​​based on the second indicator from the input element ID, and outputs the estimated element order information.

[0173] In the training data shown in Figure 5 or Figure 11, training data 3 is prepared with evaluation values ​​based on a third indicator (for example, building risk). The trained model 464 is trained based on training data 3. The trained model 464 takes an element ID as input, estimates one or more element order information with high evaluation values ​​based on the third indicator from the input element ID, and outputs the estimated element order information.

[0174] The method for generating the trained models 460 to 464 is the same as in the first and second embodiments described above.

[0175] In step S302, the element IDs obtained in step S300 are input to each of the trained models 460-464, and estimation is performed by obtaining the element order information output by the decision unit 466. The same procedure is followed for steps S312, S336, and S346.

[0176] The decision unit 466 determines one or more element order information from the element order information output by the trained models 460 to 464. For example, the decision unit 466 can be configured as follows.

[0177] The first configuration determines the element order information based on the number of occurrences of the element order information. For example, if element order information A and B are output from trained model 460, element order information A, B, and C are output from trained model 462, and element order information A is output from trained model 464, the number of occurrences of element order information A to C will be "3", "2", and "1" respectively. Therefore, if the top one is element order information A, then element order information A is output; if the top two are element order information A and B are output.

[0178] The second configuration determines the element order information based on the number of occurrences of the editing elements included in the element order information. For example, in the above example in the first configuration, if element order information A includes editing elements a and b, element order information B includes editing elements b and c, and element order information C includes editing elements a and c, then the number of occurrences of editing elements a, b, and c will be "4", "5", and "3", respectively. If we calculate the number of occurrences of the editing elements included in element order information A to C by summing up the number of occurrences of the editing elements included in that element order information, the number of occurrences of element order information A to C will be "9", "8", and "3", respectively. Therefore, if there is one element order information, element order information A is output; if there are two element order information information A and B are output.

[0179] The trained models 460-464 are configured to output evaluation values ​​and fitting values ​​along with element order information. The third configuration determines the element order information according to the evaluation values, etc. In the example above in the first configuration, if the evaluation values, etc. of element order information A-C are "0.2", "0.4", and "0.5" respectively, then when the evaluation values, etc. of element order information A-C are corrected by multiplying them by the number of occurrences, the evaluation values, etc. of element order information A-C become "0.2 × 3 = 0.6", "0.4 × 2 = 0.8", and "0.5 × 1 = 0.5" respectively. Therefore, if there is one top value, element order information B is output, and if there are two top values, element order information A and B are output.

[0180] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, multiple trained models, each trained based on evaluation values ​​derived from different indicators, are connected in parallel. Element IDs are input to each trained model, and element order information is estimated by determining the element order information from the output results of each trained model.

[0181] This makes it possible to understand the order of elements according to evaluation values ​​based on multiple different indicators.

[0182] [Fifth Embodiment] Next, a fifth embodiment of the present invention will be described. Figure 15 shows this embodiment. Figures 5 and 11 will also be referenced.

[0183] This embodiment differs from the third embodiment in that it estimates element order information using multiple trained models, each trained based on evaluation values ​​derived from different indicators. Below, only the parts that differ from the first to third embodiments will be described, and the overlapping parts will be omitted.

[0184] [Configuration of this embodiment] First, the configuration of this embodiment will be described. Figure 15 shows three pre-trained models 470-474 connected in series.

[0185] As shown in Figure 15, the memory device 42 stores three trained models 470 to 474.

[0186] In the training data shown in Figure 5 or Figure 11, training data 1 is prepared with evaluation values ​​based on a first indicator (for example, building safety). The trained model 470 is trained based on training data 1. The trained model 470 takes an element ID as input, estimates one or more element order information with high evaluation values ​​based on the first indicator from the input element ID, and outputs the estimated element order information.

[0187] In the training data shown in Figure 5 or Figure 11, element IDs 410 and 430 are removed, and training data 2 is prepared by setting evaluation values ​​based on a second indicator (for example, the functionality of the building). The trained model 472 is trained based on training data 2. The trained model 472 takes the element order information output by the trained model 470 as input, estimates one or more element order information with high evaluation values ​​based on the second indicator from the input element order information, and outputs the estimated element order information.

[0188] In the training data shown in Figure 5 or Figure 11, element IDs 410 and 430 are removed, and training data 3 is prepared with evaluation values ​​based on a third indicator (e.g., building risk). The trained model 474 is trained based on training data 3. The trained model 474 takes the element order information output by the trained model 472 as input, estimates one or more element order information with high evaluation values ​​based on the third indicator from the input element order information, and outputs the estimated element order information.

[0189] The method for generating the trained models 470 to 474 is the same as in the first and second embodiments described above.

[0190] In step S302, the element IDs obtained in step S300 are input to the trained model 470, and estimation is performed by obtaining the element order information output by the trained model 474. The same procedure is followed for steps S312, S336, and S346.

[0191] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, multiple trained models, each trained based on evaluation values ​​derived from different indicators, are connected in series. Element IDs are input to the first-stage trained model, and element order information is estimated by obtaining the element order information output by the final-stage trained model.

[0192] This makes it possible to understand the order of elements according to evaluation values ​​based on multiple different indicators.

[0193] [Sixth Embodiment] Next, a sixth embodiment of the present invention will be described. Figures 16 to 19 show this embodiment. Figures 5 and 9 will also be referenced.

[0194] This embodiment differs from the first embodiment in that it utilizes a large language model. Below, only the differences from the first embodiment will be described, and the overlapping parts will be omitted.

[0195] [Configuration of this embodiment] First, the configuration of this embodiment will be described. Figure 16 is a block diagram showing the configuration of the network system according to this embodiment.

[0196] As shown in Figure 16, the Internet 199 is connected to a drawing creation support device 100 and a generation AI server 120 that generates response information using an AI (Artificial Intelligence) model in response to requests, enabling communication between them.

[0197] [Generating AI Server 120] Next, we will explain the configuration of the generation AI server 120. The generation AI server 120, like the drawing creation support device 100, has a hardware configuration similar to a general computer with a CPU, ROM, RAM, and I / F connected via a bus, and is configured, for example, as a cloud server.

[0198] Figure 17 is a functional block diagram of the generation AI server 120. As shown in Figure 17, the generating AI server 120 is configured to include a plurality of AI models 50, an AI model control unit 52 that controls the AI ​​models 50, and a knowledge base 54 that registers data that the AI ​​models 50 refer to for inference.

[0199] AI Model 50 is an AI model trained on a large dataset and is a highly versatile model capable of performing various tasks. For example, AI Model 50 can employ a large-scale language model. A large-scale language model is a deep learning model that pre-trains a language model, which models human spoken language based on its occurrence probability, on a vast amount of data. When a prompt is input, the large-scale language model statistically infers the probability of generating the next word from the sentence contained in the input prompt and outputs the inference result. For example, publicly known techniques described on the internet sites "https: / / chatgpt-lab.com / n / n418d3aa56f0b" and "https: / / agirobots.com / chatgpt-mechanism-and-problem / " can be used as large-scale language models. More specifically, for example, Titan Text G1 - Express, Titan Text G1 - Lite, Titan Image Generator G1, Titan Embeddings G1 - Text, Titan Embeddings Text V2, Titan Multimodal Embeddings G1, Claude, Claude Instant, Claude 3 Sonnet, Claude 3 Haiku, Claude 3 Opus, Jurassic-2 Mid, Jurassic-2 Ultra, Command, Command Light, Command R, Command R+, Embed English, Embed Multilingual, Llama 2 Chat 13B, Llama 2 Chat 70B, Llama 2 13B, Llama 2 70B, Llama 3 8b Instruct, Llama 3 70b Instruct, Mistral 7B Instruct, Mixtral 8X7B Instruct, Mistral Large, and Stable Diffusion XL can be adopted.

[0200] The AI ​​model control unit 52 selects one of the multiple AI models 50 to be used for inference in response to a selection request from the request processing unit 58. When a reference request is input from the request processing unit 58, the AI ​​model control unit 52 causes the selected AI model 50 (hereinafter referred to as the "selected AI model") to access data in the knowledge base 54 according to the input reference request. When a prompt is input from the request processing unit 58, the AI ​​model control unit 52 inputs the prompt to the selected AI model. Finally, when an execution request is input from the request processing unit 58, the AI ​​model control unit 52 causes the selected AI model to perform inference according to the input execution request, obtains the inference result from the selected AI model, and outputs the obtained inference result to the request processing unit 58.

[0201] Knowledge base 54 can register training data. The information registered in knowledge base 54 is in a data format that the AI ​​model 50 can access (for example, vector data).

[0202] The generation AI server 120 is further configured to include a request receiving unit 56 that receives requests, a request processing unit 58 that processes the requests received by the request receiving unit 56, and a response information transmission unit 60 that transmits response information for the requests received by the request receiving unit 56 to the drawing creation support device 100.

[0203] The request receiving unit 56 receives a request from the drawing creation support device 100 for the generation of response information and outputs the received request to the request processing unit 58. The request includes (1) an element ID that has been created or edited, (2) a generation request to generate element sequence information relating to multiple editing elements to be edited consecutively after an edited element that has been created or edited, and the editing order thereof, (3) a selection request to select an AI model 50, and (4) a reference request to refer to the learning data of the knowledge base 54. Items (3) and (4) are not mandatory but are included as additional items.

[0204] If the request received by the request receiving unit 56 includes a selection request or a reference request, the request processing unit 58 outputs the selection request or reference request to the AI ​​model control unit 52. It also generates a prompt to instruct the AI ​​model 50 based on the request received by the request receiving unit 56. For example, the prompt may request the AI ​​model 50 to generate element sequence information relating to multiple editing elements to be edited consecutively after a created or edited editing element, and their editing order, based on the created or edited element ID, provided that the evaluation value is above a predetermined value. The generated prompt and execution request are then output to the AI ​​model control unit 52, and if the AI ​​model control unit 52 inputs an inference result in response to the execution request, the input inference result is output to the response information transmission unit 60.

[0205] The response information transmission unit 60 transmits the response information, including the inference result input from the request processing unit 58, to the drawing creation support device 100.

[0206] The generation AI server 120 is further configured to include a request receiving unit 62 that receives requests and a learning data registration unit 64 that registers learning data in the knowledge base 54.

[0207] The request receiving unit 62 receives a request from the drawing creation support device 100 to register learning data, and outputs the received request to the learning data registration unit 64. The request includes (1) learning data.

[0208] The learning data registration unit 64 stores the learning data included in the request received by the request receiving unit 62 in storage (not shown) and converts it into a data format that the AI ​​model 50 can access (for example, vector data). Vector data can be generated by a technique (embedding) that converts data including characters, images, and audio into numerical vectors. The converted learning data is then registered in the knowledge base 54. The AI ​​model control unit 52 allows the AI ​​model 50 to access the learning data in response to a reference request from the request processing unit 58.

[0209] [Operation of this embodiment] Next, the operation of this embodiment will be described. [Training data registration process] Figure 18 is a flowchart showing the training data registration process.

[0210] The learning data registration process is performed in response to requests from designers and other users. When it is executed on the CPU 30, it first proceeds to step S400, as shown in Figure 18.

[0211] In step S400, the learning data shown in Figure 5 is acquired from the storage device 42, and the process proceeds to step S402.

[0212] In step S402, a request is sent to the generating AI server 120 requesting the registration of training data. The request includes (1) the training data acquired in step S400.

[0213] Once the process in step S402 is completed, the series of processes ends. [Element Order Information Acquisition Process] Figure 19 is a flowchart showing the process for obtaining element order information.

[0214] The element sequence information acquisition process is performed in response to requests from designers or other users. When executed on the CPU 30, it first proceeds to step S500, as shown in Figure 19.

[0215] In step S500, the element IDs of created or edited elements are obtained from the CAD data currently being edited, and the process proceeds to step S502.

[0216] In step S502, a request is sent to the generation AI server 120 requesting the generation of response information based on the element ID obtained in step S500. The request includes (1) the element ID obtained in step S500, (2) a generation request to generate multiple element sequence information relating to multiple editing elements to be edited consecutively after an already created or edited editing element and their editing order, wherein the element sequence information will be generated for multiple editing elements or editing order that differs, (3) a selection request to select a predetermined AI model 50, and (4) a reference request to refer to the training data in the knowledge base 54.

[0217] Next, the process moves to step S504, where response information is received from the generating AI server 120. Then, the process moves to step S506, where, similar to the process in step S304, one of the multiple element sequences is displayed on the display device 44 based on the multiple element sequence information contained in the received response information, and the process moves to step S508.

[0218] In step S508, if the designer creates or edits an editable element based on the displayed element order, the created or edited element IDs, including the element ID of that editable element, are obtained from the currently edited CAD data, and the process proceeds to step S510.

[0219] In step S510, based on the element order displayed in steps S506 and S518 and the element ID obtained in step S508, it is determined whether the edited element created or edited by the designer differs from the edited element or editing order displayed in steps S506 and S518. If it is determined that the editing result differs from the inference result (YES), the process proceeds to step S512.

[0220] In step S512, similar to the process in step S310, the display rules used for display in steps S506 and S518 are changed, and the process proceeds to step S514.

[0221] In step S514, similar to the process in step S502, a request is sent to the generation AI server 120 requesting the generation of response information based on the element ID obtained in step S508, and the process proceeds to step S516.

[0222] In step S516, similar to the process in step S504, response information is received from the generating AI server 120, and the process moves to step S518, where, similar to the process in step S304, one of the multiple element order information contained in the received response information is displayed on the display device 44, and the process moves to step S520.

[0223] In step S520, it is determined whether the designer has finished editing. If it is determined that editing is complete (YES), the series of processes is terminated.

[0224] On the other hand, if it is determined in step S520 that the designer has not finished editing (NO), the process proceeds to step S508.

[0225] On the other hand, if it is determined in step S510 that the editing result matches the inference result (NO), the process proceeds to step S514.

[0226] [When considering the delivery of a piano] Next, we will explain the procedure when considering the delivery of a piano.

[0227] If a designer wants to install a piano with a width of 120 mm in the cross-section drawing of Figure 9 using CAD software, they need to shorten the width of the toilet to prevent interference between the piano and the toilet wall when bringing the piano in. However, changing the width of the toilet necessitates editing other editing elements of the toilet. Therefore, after changing the width of the toilet, the designer requests inference of the order of elements to be edited for the toilet. After steps S500 to S506, the created or edited element IDs are obtained from the CAD data currently being edited, and the element order is inferred and displayed from the obtained element IDs. In the inference, the AI ​​model 50 refers to the training data of the knowledge base 54. The method of displaying the element order is the same as in the first embodiment described above.

[0228] Then, when the designer creates or edits the editing element "toilet" based on the inferred and displayed element order, the next element order is inferred and displayed via steps S508 to S518. If the designer's editing result differs from the inferred result, the display rule is changed via step S512.

[0229] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, a request is input to the AI ​​model 50 that includes a generation request to generate multiple element sequence information that includes created or edited element IDs and has different edited elements or editing order, and in response to the request, the multiple element sequence information output from the AI ​​model 50 is obtained.

[0230] This allows you to understand multiple editing elements being edited sequentially and their editing order.

[0231] Furthermore, in this embodiment, a request including a generation request to generate multiple element order information with different editing elements or editing orders is input to the AI ​​model 50, and in response to the request, multiple element order information output from the AI ​​model 50 is obtained, and one of the multiple element orders is displayed based on the obtained multiple element order information.

[0232] This allows for the identification of multiple possible editing sequences for several elements and their editing order.

[0233] Furthermore, in this embodiment, learning data including created or edited element IDs, multiple edited elements that are edited consecutively after the edited element, element order information relating to the editing order, and evaluation values ​​relating to the editing of the multiple edited elements is registered in the knowledge base 54, and in response to a request, the learning data in the knowledge base 54 is referenced to obtain the multiple element order information output from the AI ​​model 50.

[0234] This allows us to identify the order of elements that receive the highest evaluation scores. Furthermore, in this embodiment, based on the displayed element order, the element IDs of the edited elements created or edited by the designer are obtained, and a request including a generation request that generates multiple element order information with different edited elements or editing orders, including the obtained element IDs, is input to the AI ​​model 50, and in response to the request, the multiple element order information output from the AI ​​model 50 is obtained.

[0235] This allows the element order to be inferred and displayed for the result of creation or editing based on the displayed element order, making it possible to understand multiple editing elements being edited consecutively and their editing order.

[0236] Furthermore, in this embodiment, based on the displayed element order, the AI ​​model 50 receives a request that includes the element IDs of edited elements created or edited by the user, the element IDs of the edited elements created or edited by the user, and the generation request that includes the obtained element IDs and generates multiple element order information with different edited elements or editing orders. In response to the request, the AI ​​model 50 outputs multiple element order information, and based on the obtained multiple element order information, it displays one of the multiple element orders.

[0237] As a result, multiple element orders are inferred and displayed based on the displayed element order when creating or editing, allowing you to understand multiple candidates for consecutively edited elements and their editing order.

[0238] [Seventh Embodiment] Next, a seventh embodiment of the present invention will be described. Figure 20 is a diagram showing this embodiment. Figures 5, 9, 11, and 18 will also be referenced.

[0239] This embodiment differs from the second and sixth embodiments described above in that it causes the AI ​​model 50 to refer to first training data including evaluation values ​​based on the first indicator or second training data including evaluation values ​​based on the second indicator. Below, only the parts that differ from the second and sixth embodiments described above will be explained, and the overlapping parts will be omitted from the explanation.

[0240] [Operation of this embodiment] Next, the operation of this embodiment will be described. [Training data registration process] In the training data registration process, after going through steps S400 to S402, the training data shown in Figure 5 and the training data shown in Figure 11 are retrieved from the storage device 42, and a request for registration of the training data is sent to the generating AI server 120. The request includes (1) the training data shown in Figure 5 as the first training data, and (2) the training data shown in Figure 11 as the second training data.

[0241] [Element Order Information Acquisition Process] Figure 20 is a flowchart showing the process for obtaining element order information.

[0242] The element sequence information acquisition process is performed in response to requests from designers or other users. When executed on the CPU 30, it first proceeds to step S530, as shown in Figure 20.

[0243] In step S530, indicator information related to the first indicator "editing time" or the second indicator "number of edited items," which indicate the value of the evaluation value, is obtained. The process then moves to step S532, where the IDs of created or edited elements are obtained from the CAD data currently being edited, and then moves to step S534.

[0244] In step S534, a request is sent to the generation AI server 120 requesting the generation of response information based on the metric information and element IDs obtained in steps S530 and S532. The request includes (1) the element ID obtained in step S532, (2) a generation request to generate multiple element sequence information relating to multiple editing elements to be edited consecutively after an already created or edited editing element and their editing order, wherein the element sequence information is different in editing elements or editing order, (3) a selection request to select a predetermined AI model 50, and (4) a reference request to refer to the first learning data in the knowledge base 54 if the metric related to the metric information obtained in step S530 is the first metric, or a reference request to refer to the second learning data in the knowledge base 54 if the metric related to the metric information obtained in step S530 is the second metric.

[0245] Next, the process moves to step S536, where response information is received from the generating AI server 120. Then, the process moves to step S538, where, similar to the process in step S304, one of the multiple element sequences is displayed on the display device 44 based on the multiple element sequence information contained in the received response information, and the process moves to step S540.

[0246] In step S540, if the designer creates or edits an editable element based on the displayed element order, the created or edited element IDs, including the element ID of that editable element, are obtained from the currently edited CAD data, and the process proceeds to step S542.

[0247] In step S542, based on the element order displayed in steps S538 and S550 and the element ID obtained in step S540, it is determined whether the edited element created or edited by the designer differs from the edited element or editing order displayed in steps S538 and S550. If it is determined that the editing result differs from the inference result (YES), the process proceeds to step S544.

[0248] In step S544, similar to the process in step S310, the display rules used for display in steps S538 and S550 are changed, and the process proceeds to step S546.

[0249] In step S546, similar to the process in step S534, a request is sent to the generation AI server 120 requesting the generation of response information based on the indicator information and element IDs obtained in steps S530 and S540, and the process proceeds to step S548.

[0250] In step S548, similar to the process in step S536, response information is received from the generating AI server 120, and the process moves to step S550, where, similar to the process in step S304, one of the multiple element order information contained in the received response information is displayed on the display device 44, and the process moves to step S552.

[0251] In step S552, it is determined whether the designer has finished editing. If it is determined that editing is complete (YES), the series of processes is terminated.

[0252] On the other hand, if it is determined in step S552 that the designer has not finished editing (NO), the process proceeds to step S540.

[0253] On the other hand, if it is determined in step S542 that the editing result matches the inference result (NO), the process proceeds to step S546.

[0254] [When considering the delivery of a piano] Next, we will explain the procedure when considering the delivery of a piano.

[0255] When a designer wants to place a piano with a width of 120 mm in the sectional plan view of FIG. 9 in CAD software, it is necessary to reduce the width of the toilet in order to prevent interference between the piano and the toilet wall when carrying the piano in. However, changing the width of the toilet makes it necessary to edit other editing elements of the toilet as well. Therefore, after changing the width of the toilet, the designer requests inference of the order of elements to be edited for the toilet. In this case, if the designer wants to obtain an element order that reduces editing time, when the designer selects "editing time" as an indicator, through steps S530 to S538, created or edited element IDs are acquired from the currently edited CAD data, and the element order is inferred from the acquired element IDs and displayed. In the inference, the AI model 50 refers to the first learning data in the knowledge base 54.

[0256] In contrast, if the designer wants to obtain an element order that reduces the number of editing items, when the designer selects "number of editing items" as an indicator, through steps S530 to S538, created or edited element IDs are acquired from the currently edited CAD data, and the element order is inferred from the acquired element IDs and displayed. In the inference, the AI model 50 refers to the second learning data in the knowledge base 54.

[0257] [Effects of the Present Embodiment] Next, effects of the present embodiment will be described. In the present embodiment, first learning data including evaluation values based on a first indicator and second learning data including evaluation values based on a second indicator are registered in the knowledge base 54, indicator information relating to the first indicator or the second indicator is acquired, and a request including a request to refer to either the first learning data or the second learning data in the knowledge base 54 based on the acquired indicator information is input to the AI model 50.

[0258] This makes it possible to grasp an element order having a high evaluation value based on the first indicator or the second indicator.

[0259] [Eighth Embodiment] Next, an eighth embodiment of the present invention will be described. This embodiment differs from the sixth and seventh embodiments in that it employs evaluation values ​​other than editing time or the number of editing items. Below, only the parts that differ from the sixth and seventh embodiments will be described, and the overlapping parts will be omitted from the explanation.

[0260] The storage device 42 stores the learning data (hereinafter referred to as "learning data 1") in the third embodiment described above.

[0261] In the learning data registration process, after going through steps S400 to S402, learning data 1 is obtained from the storage device 42 and a request to register learning data 1 is sent to the generating AI server 120. When the generating AI server 120 infers element sequence information, the AI ​​model 50 refers to learning data 1 in the knowledge base 54, so it can grasp the element sequence according to the evaluation value regarding the quality or risk of the building.

[0262] [Ninth Embodiment] Next, a ninth embodiment of the present invention will be described. This embodiment differs from the eighth embodiment in that it infers elemental order information using multiple training data sets, each containing evaluation values ​​based on different indicators. Below, only the parts that differ from the sixth to eighth embodiments will be described, and the overlapping parts will be omitted.

[0263] [Configuration of this embodiment] First, the configuration of this embodiment will be described. The storage device 42 stores the learning data 1 to 3 in the fourth embodiment described above.

[0264] In the training data registration process, after going through steps S400 to S402, training data 1 to 3 are obtained from the storage device 42, and a request to register training data 1 to 3 is sent to the generating AI server 120.

[0265] Steps S502 and S504 consist of the first to fourth processes. The first process sends a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. The request includes (1) the element ID obtained in step S500, (2) a generation request to generate multiple element sequence information with different editing elements or editing order, (3) a selection request to select a predetermined AI model 50, and (4) a reference request to refer to the training data 1 in the knowledge base 54. The system then receives answer information from the generation AI server 120 in response to the request and obtains multiple element sequence information from the received answer information.

[0266] The second process involves sending a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. In addition to (1), (2), and (3) above, the request includes (4) a reference request to refer to the learning data 2 of the knowledge base 54. The system then receives the answer information from the generation AI server 120 in response to the request and obtains multiple element sequence information from the received answer information.

[0267] The third process sends a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. In addition to (1), (2), and (3) above, the request includes (4) a reference request to refer to the training data 3 of the knowledge base 54. The generation AI server 120 then receives the answer information in response to the request, and multiple element order information is obtained from the received answer information.

[0268] The fourth process determines one or more element order information from the element order information obtained in the first to third processes. As the fourth process, for example, a process equivalent to the determination unit 466 in the third embodiment described above can be adopted.

[0269] The same applies to steps S514, S516, S534, S536, and S546, S548.

[0270] [Effects of this embodiment] Next, effects of the present embodiment will be described. In the present embodiment, element order information is obtained by causing the AI model 50 to output element order information with reference to a plurality of pieces of learning data each including evaluation values based on different indicators, and determining the element order information from the obtained results.

[0271] This makes it possible to grasp an element order corresponding to evaluation values based on a plurality of different indicators.

[0272] [Tenth Embodiment] Next, a tenth embodiment of the present invention will be described. The present embodiment differs from the eighth embodiment described above in that element order information is inferred using a plurality of pieces of learning data each including evaluation values based on different indicators. Hereinafter, only portions different from the sixth to eighth embodiments described above will be described, and descriptions of overlapping portions will be omitted.

[0273] [Configuration of Present Embodiment] First, the configuration of the present embodiment will be described. A storage device 42 stores learning data 1 to 3 in the fifth embodiment described above.

[0274] In learning data registration processing, through steps S400 to S402, learning data 1 to 3 are acquired from the storage device 42, and a request for requesting registration of the learning data 1 to 3 is transmitted to a generative AI server 120.

[0275] Steps S502 and S504 consist of first to third processes. The first process sends a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. The request includes (1) the element ID obtained in step S500, (2) a generation request to generate multiple element sequence information with different editing elements or editing order, (3) a selection request to select a predetermined AI model 50, and (4) a reference request to refer to the training data 1 in the knowledge base 54. The system then receives answer information from the generation AI server 120 in response to the request and obtains multiple element sequence information from the received answer information.

[0276] The second process sends a request to the generation AI server 120 requesting the generation of answer information based on the element order information obtained in the first process. In addition to (2) and (3) above, the request includes (1) the element order information obtained in the first process and (4) a reference request to refer to the training data 2 of the knowledge base 54. The generation AI server 120 then receives the answer information in response to the request and obtains multiple pieces of element order information from the received answer information.

[0277] The third process sends a request to the generation AI server 120 requesting the generation of answer information based on the element order information obtained in the second process. In addition to (2) and (3) above, the request includes (1) the element order information obtained in the second process and (4) a reference request to refer to the training data 3 of the knowledge base 54. The generation AI server 120 then receives the answer information in response to the request and obtains multiple pieces of element order information from the received answer information.

[0278] In step S506, one of several element sequences is displayed on the display device 44 based on the element sequence information obtained in the third process.

[0279] The same applies to steps S514-S518, S534-S538, and S546-S550.

[0280] [Effects of this embodiment] Next, the effects of this embodiment will be described. In this embodiment, for a request including an element ID and a generation request, element sequence information is obtained from the AI ​​model 50 by referring to training data including evaluation values ​​based on a first indicator. For a request including the obtained element sequence information and a generation request, element sequence information is obtained from the AI ​​model 50 by referring to training data including evaluation values ​​based on a second indicator.

[0281] This allows us to understand the order of elements according to the evaluation values ​​based on the first and second indicators.

[0282] Furthermore, in this embodiment, in response to a request including element sequence information and a generation request, element sequence information output from the AI ​​model 50 is obtained by referring to training data including evaluation values ​​based on a second indicator, and in response to a request including the acquired element sequence information and a generation request, element sequence information output from the AI ​​model 50 is obtained by referring to training data including evaluation values ​​based on a third indicator.

[0283] This allows us to understand the order of elements according to the evaluation values ​​based on the second and third indicators.

[0284] [Variation] In the first, second, sixth, and seventh embodiments and their variations described above, editing time or the number of editing items was used as the evaluation value, but the system is not limited to these, and other evaluation values ​​related to the design of buildings can be used.

[0285] Furthermore, in the embodiments described in the third to fifth and eighth to tenth embodiments and their modifications, the evaluation values ​​were standardized for each indicator. However, the invention is not limited to this, and normalization or other scaling processes may be performed, or standardization, normalization, or other scaling processes may not be performed.

[0286] Furthermore, while the fourth embodiment and its variations described above use building quality or risk as indicators, the system is not limited to these, and user information relating to the user can also be used as an indicator. For example, the following configuration can be adopted.

[0287] Each row of training data 1 to 3 contains a user ID to identify the user who edited the editing elements related to element IDs 410 and 430 (hereinafter referred to as the "editing user"). To illustrate with the example in Figure 5, the second row of Figure 5 contains two editing elements that were edited consecutively after the editing element in the first row of Figure 3, their editing order, the user ID of the editing user, and their evaluation value. The third row of Figure 5 contains two editing elements that were edited consecutively after the editing element in the second row of Figure 3, their editing order, the user ID of the editing user, and their evaluation value. The fourth row of Figure 5 contains two editing elements that were edited consecutively after the editing element in the third row of Figure 3, their editing order, the user ID of the editing user, and their evaluation value.

[0288] In step S302, the user ID of the user currently editing is obtained, and the element ID and user ID obtained in step S300 are input to each of the trained models 460 to 464. Estimation is then performed by obtaining the element order information output by the decision unit 466. This makes it possible to estimate element order information that is suitable for the user.

[0289] In this modified example, the editing history data and training data are composed of user IDs, but are not limited to user IDs; they can also be composed of other identification information, or feature information relating to the user's overview, statistics, and other characteristics.

[0290] This modification can also be applied to the first to third and fifth to tenth embodiments and their modifications.

[0291] Furthermore, in the fourth embodiment and its modified form described above, three trained models 460 to 464 were connected in parallel, but the configuration is not limited to this, and two trained models or four or more trained models can be connected in parallel.

[0292] Furthermore, in the fifth embodiment and its modified form described above, three trained models 470 to 474 were connected in series, but the configuration is not limited to this, and two trained models or four or more trained models can be connected in series.

[0293] Furthermore, in the ninth and tenth embodiments and their modifications described above, the AI ​​model 50 was made to refer to three training data sets, but it is not limited to this, and it is also possible to refer to two training data sets or four or more training data sets.

[0294] Furthermore, in the fourth and fifth embodiments and their modifications described above, multiple trained models are configured by connecting them in either parallel or series, but the system is not limited to this, and can be configured by arbitrarily combining parallel and series connections. In the node that aggregates the output results of multiple trained models, a configuration can be adopted in which a determination unit 466 is provided when determining element order information from the output results of multiple trained models, or a configuration can be adopted in which the output results of multiple trained models are directly input to the next trained model.

[0295] Furthermore, in the fourth embodiment and its modified form, the determination unit 466 determined one or more element order information from the element order information output by the trained models 460 to 464, but is not limited to this, and any method can be adopted as a method for determining the element order information. The same applies to the fourth process in the ninth embodiment.

[0296] Furthermore, in the first to fifth embodiments and their variations described above, multiple element order information was estimated in steps S302, S312, S336, and S346, but the method is not limited to this, and one element order information can be estimated.

[0297] Furthermore, in the sixth to tenth embodiments and their modifications described above, steps S502, S514, S534, and S546 included a generation request for generating multiple element sequence information in the request, but the request is not limited to this, and a generation request for generating one element sequence information can also be included in the request.

[0298] Furthermore, while the first to tenth embodiments and their variations employ the first to fourth display rules, the system is not limited to these. For example, any display rule can be adopted that displays the fewest number of editing elements, an element order greater than or equal to a predetermined value, less than or equal to a predetermined value, or any other arbitrary display rule.

[0299] Furthermore, in the first to tenth embodiments and their variations described above, the element order is displayed according to one of several display rules. However, the system is not limited to this, and the element order can be displayed according to a predetermined display rule without providing multiple display rules.

[0300] Furthermore, while one element order was displayed in the first to tenth embodiments and their variations described above, the system is not limited to this, and multiple element orders can be displayed.

[0301] Furthermore, while the labeling rules were changed in the first to tenth embodiments and their variations described above, the system is not limited to these, and configurations that do not change the labeling rules can also be adopted.

[0302] Furthermore, in the first to fifth embodiments and their variations described above, estimation and display were repeated multiple times, but the system is not limited to this, and a configuration in which estimation and display are performed only once can also be adopted. The same applies to the sixth to tenth embodiments and their variations described above.

[0303] Furthermore, while reinforcement learning was employed as the learning method in the first to fifth embodiments and their variations described above, the method is not limited to this, and supervised learning, semi-supervised learning, unsupervised learning, deep learning, or any other learning method can be adopted.

[0304] Furthermore, in the first to tenth embodiments and their variations described above, the training data was configured to include evaluation values, but it is not limited to this and can be configured without including evaluation values.

[0305] Furthermore, in the first to tenth embodiments and their variations described above, the editing history data and learning data are composed of element IDs, but are not limited to this and may be composed of identification information other than element IDs, or feature information relating to the summary, statistics, and other characteristics of the editing elements.

[0306] Furthermore, in the second embodiment and its modified form, the first trained model and the second trained model can be configured as a single trained model. Similarly, in the third to fifth embodiments and their modified forms, m (m≧2) trained models can be configured as n (1≦n≦m) trained models.

[0307] Furthermore, in the sixth to tenth embodiments and their variations described above, the AI ​​model 50 can be configured as the trained model in the first to fifth embodiments and their variations described above.

[0308] Furthermore, in the first to tenth embodiments and their variations described above, element sequences with a predetermined or greater frequency of appearance in the training data were identified. However, the invention is not limited to this, and element sequences with a predetermined or greater frequency of appearance in the editing history data can also be identified.

[0309] Furthermore, while the first to tenth embodiments and their variations described above handled a maximum of four editing elements and their editing order, the system is not limited to this, and can handle five or more editing elements and their editing order. There is no need to limit the number of editing elements; any number of editing elements is acceptable. In addition, the number of editing elements for the training data in Figures 5 and 11 can be set independently.

[0310] Furthermore, in the first to tenth embodiments and their variations described above, training data was generated by obtaining all combinations of 2 to 4 editing elements from the editing history data regarding element order. However, the model is not limited to this, and training data can be generated by obtaining element orders that appear more than a predetermined number of times from the editing history data. In this case, a trained model that has been trained on multiple editing history data sets may be used to estimate the element orders with high occurrence frequencies. This makes it possible to learn or infer the element orders with high occurrence frequencies within the editing history data.

[0311] Furthermore, in the first to tenth embodiments and their modifications described above, the editing history data was configured as separate data from the CAD data, but this is not limited to this configuration; it can also be included in the CAD data and configured as an integrated entity.

[0312] Furthermore, in the sixth to tenth embodiments and their modifications described above, the AI ​​model 50 was made to refer to the learning data in the knowledge base 54. However, the learning data to be referenced in the knowledge base 54 can be included in the request. This allows the system to be applied even in configurations that do not have a knowledge base 54. Specifically, for example, the following configuration can be adopted.

[0313] [Invention A1] An element information acquisition means for acquiring element information relating to created or edited elements in design information, Input means for inputting a request to the AI ​​model that includes element information acquired by the element information acquisition means, element information relating to created or edited elements, element sequence information relating to a plurality of elements edited consecutively after the element and their editing order (hereinafter, "a plurality of elements and their editing order" is referred to as "element sequence"), and reference information including evaluation values ​​relating to the plurality of elements, and includes a request to generate element sequence information relating to the element sequence. The system includes an acquisition means for acquiring the element sequence information output from the AI ​​model in response to the aforementioned request, The aforementioned evaluation values ​​are evaluation values ​​related to the quality or risk of the design object.

[0314] This allows for the tracking of multiple elements being edited sequentially and their editing order. Furthermore, it enables the tracking of element order based on evaluation values ​​related to the quality or risk of the design target.

[0315] [Invention A2] In Invention A1, The system includes an indicator information acquisition means for acquiring indicator information relating to a first indicator that shows an indicator of the value of the aforementioned evaluation value, or a second indicator different from the first indicator. The input means inputs the request to the AI ​​model, which includes either first reference information including the element information, the element sequence information, and the evaluation value based on the first indicator, or second reference information including the element information, the element sequence information, and the evaluation value based on the second indicator, based on the indicator information acquired by the indicator information acquisition means.

[0316] This allows us to identify the order of elements that have high evaluation values ​​based on the first or second indicator.

[0317] [Invention A3] In inventions A1 and A2, The evaluation values ​​for the quality of the design object are evaluation values ​​for the safety, functionality, convenience, aesthetics, environmental friendliness, economic efficiency, or legal compliance of the design object.

[0318] [Invention A4] In inventions A1 and A2, The assessment values ​​for the risks of the design object include the risk of a decline in the quality of the design object, the risk of the design object collapsing or being damaged, the risk of the design object deteriorating, the risk of an increase in costs related to the design object, the risk of the design object being affected by natural disasters, the risk of a decrease in the asset value related to the design object, or the risk of the design object causing social problems.

[0319] Embodiments A1 to A4 of the invention will be described as modified examples of the ninth embodiment described above. The first process sends a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. The request includes (1) the element ID obtained in step S500, (2) a generation request to generate multiple element sequence information with different editing elements or editing order, (3) a selection request to select a predetermined AI model 50, and (4) training data 1. The system then receives answer information from the generation AI server 120 in response to the request and obtains multiple element sequence information from the received answer information.

[0320] The second process involves sending a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. The request includes (1), (2), and (3) above, as well as (4) training data 2. The system then receives the answer information from the generation AI server 120 in response to the request and obtains multiple element order information from the received answer information.

[0321] The third process involves sending a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. The request includes (1), (2), and (3) above, as well as (4) training data 3. The generation AI server 120 then receives the answer information in response to the request, and multiple element order information is obtained from the received answer information.

[0322] The fourth process determines one or more element order information from the element order information obtained in the first to third processes. As the fourth process, for example, a process equivalent to the determination unit 466 in the third embodiment described above can be adopted.

[0323] The same applies to steps S514, S516, S534, S536, and S546, S548.

[0324] Next, embodiments A1 to A4 of the invention will be described as modified examples of the tenth embodiment described above.

[0325] The first process sends a request to the generation AI server 120 requesting the generation of answer information based on the element ID obtained in step S500. The request includes (1) the element ID obtained in step S500, (2) a generation request to generate multiple element sequence information with different editing elements or editing order, (3) a selection request to select a predetermined AI model 50, and (4) training data 1. The system then receives answer information from the generation AI server 120 in response to the request and obtains multiple element sequence information from the received answer information.

[0326] The second process sends a request to the generation AI server 120 requesting the generation of answer information based on the element order information obtained in the first process. The request includes (2) and (3) above, as well as (1) the element order information obtained in the first process and (4) the training data 2. The generation AI server 120 then receives the answer information in response to the request, and multiple pieces of element order information are obtained from the received answer information.

[0327] The third process sends a request to the generation AI server 120 requesting the generation of answer information based on the element order information obtained in the second process. The request includes (2) and (3) above, as well as (1) the element order information obtained in the second process and (4) the training data 3. The generation AI server 120 then receives the answer information in response to the request and obtains multiple pieces of element order information from the received answer information.

[0328] The same applies to steps S514-S518, S534-S538, and S546-S550.

[0329] [Invention A5] In Invention A1, The aforementioned elements are elements whose setting or modification requires human judgment, and whose setting or modification affects other elements.

[0330] [Invention A6] In Invention A1, The aforementioned design information is design information used for designing buildings.

[0331] Furthermore, in the sixth to tenth embodiments and their modifications described above, the AI ​​model 50 was made to refer to information in the knowledge base 54. However, the AI ​​model 50 is not limited to this, and information to be referenced in the knowledge base 54 (for example, the training data in Figure 5 or Figure 11) can be obtained by web search or the like, and the AI ​​model 50 can be made to refer to the search results and perform inference. This configuration can be realized, for example, by RAG (Retrieval Augmented Generation).

[0332] Furthermore, while the first to tenth embodiments and their modifications described above used a trained model or AI model 50, the system is not limited to these, and for example, the following configurations can be adopted.

[0333] [Invention B1] An element information acquisition means for acquiring element information relating to created or edited elements in design information, The system includes a storage means that stores element sequence information relating to a plurality of elements edited consecutively after a created or edited element and the order in which they are edited (hereinafter, "a plurality of elements and their editing order" is referred to as "element sequence") in association with element information relating to the created or edited element and the evaluation value relating to the plurality of elements, and a search means that searches for element sequence information corresponding to element information acquired by the element information acquisition means, wherein the evaluation value is equal to or greater than a predetermined value. The aforementioned evaluation values ​​are evaluation values ​​related to the quality or risk of the design object.

[0334] This allows for the tracking of multiple elements being edited sequentially and their editing order. Furthermore, it enables the tracking of element order based on evaluation values ​​related to the quality or risk of the design target.

[0335] [Invention B2] In Invention B1, The evaluation values ​​for the quality of the design object are evaluation values ​​for the safety, functionality, convenience, aesthetics, environmental friendliness, economic efficiency, or legal compliance of the design object.

[0336] [Invention B3] In Invention B1, The assessment values ​​for the risks of the design object include the risk of a decline in the quality of the design object, the risk of the design object collapsing or being damaged, the risk of the design object deteriorating, the risk of an increase in costs related to the design object, the risk of the design object being affected by natural disasters, the risk of a decrease in the asset value related to the design object, or the risk of the design object causing social problems.

[0337] Embodiments B1 to B3 of the invention will be described as modified examples of the fourth embodiment described above. The storage device 42 stores element order information tables 1 to 3, which have the same data structure as the learning data 1 to 3 in the fourth embodiment described above. In step S302, one or more element order information corresponding to the element ID acquired in step S300, with an evaluation value of a predetermined value or higher, is searched from element order information tables 1 to 3, and one or more element order information is determined from the retrieved element order information. The same method as the determination method used by the determination unit 466 can be used for this determination. In step S304, the element order is displayed on the display device 44 based on the element order information determined in step S302. The same applies to steps S312, S134, S336, S338, S346, and S348.

[0338] Next, embodiments B1 to B3 of the invention will be described as modified examples of the fifth embodiment described above. The storage device 42 stores element order information tables 1 to 3, which have the same data structure as the learning data 1 to 3 in the fifth embodiment described above. In step S302, one or more element order information corresponding to the element ID acquired in step S300, with an evaluation value of a predetermined or higher, is acquired from element order information table 1. Next, one or more element order information corresponding to the acquired element order information, with an evaluation value of a predetermined or higher, is acquired from element order information table 2. Then, one or more element order information corresponding to the acquired element order information, with an evaluation value of a predetermined or higher, is acquired from element order information table 3. In step S304, the element order is displayed on the display device 44 based on the element order information acquired in step S302. The same applies to steps S312, S314, S336, S338, S346, and S348.

[0339] [Invention B4] In Inventions B1 to B3, The storage means includes a first storage means for storing the element sequence information in association with the element information and the evaluation value based on a first index indicating an index of the value of the evaluation value, and a second storage means for storing the element sequence information in association with the element information and the evaluation value based on a second index different from the first index. An indicator information acquisition means for acquiring indicator information relating to the first indicator or the second indicator, The system includes a storage means selection means that selects either the first storage means or the second storage means based on the index information acquired by the index information acquisition means, The search means retrieves the element sequence information from the storage means selected by the storage means selection means.

[0340] This allows us to identify the order of elements that have high evaluation values ​​based on the first or second indicator.

[0341] An embodiment of Invention B4 will be described as a modified example of the fourth embodiment described above. The storage device 42 stores element sequence information tables 1 and 2, which have the same data structure as the learning data 1 and 2 in the fourth embodiment described above. In step S332, if the index related to the index information acquired in step S330 is the first index, element sequence information table 1 is selected, and if the index related to the acquired index information is the second index, element sequence information table 2 is selected. In step S336, one or more element sequence information corresponding to the element ID acquired in step S334, with an evaluation value of a predetermined value or higher, is searched from the element sequence information table selected in step S332. In step S346, one or more element sequence information corresponding to the element ID acquired in step S340, with an evaluation value of a predetermined value or higher, is searched from the element sequence information table selected in step S332.

[0342] [Invention B5] In Inventions B1 to B3, The storage means stores the element sequence information in association with the element information, the evaluation value, and index information relating to a first index indicating an index of the value of the evaluation value or a second index different from the first index. The system includes an indicator information acquisition means for acquiring indicator information relating to the first indicator or the second indicator, The search means searches for the element sequence information corresponding to the element information obtained by the element information acquisition means and the index information obtained by the index information acquisition means.

[0343] This allows us to identify the order of elements that have high evaluation values ​​based on the first or second indicator.

[0344] An embodiment of Invention B5 will be described as a modified example of the fourth embodiment described above. The storage device 42 stores an element sequence information table in which, for each row, created or edited element IDs 430, element sequence information 432, evaluation values ​​414, 434, and index information related to the first or second index are registered. In step S336, one or more element sequence information corresponding to the element IDs and index information obtained in steps S330 and S334, with evaluation values ​​equal to or greater than a predetermined value, are searched from the element sequence information table. In step S346, one or more element sequence information corresponding to the element IDs and index information obtained in steps S330 and S340, with evaluation values ​​equal to or greater than a predetermined value, are searched from the element sequence information table.

[0345] [Invention B6] In Invention B1, The aforementioned elements are elements whose setting or modification requires human judgment, and whose setting or modification affects other elements.

[0346] [Invention B7] In Invention B1, The aforementioned design information is design information used for designing buildings.

[0347] Furthermore, in inventions B1 to B7 and their modifications, storing element sequence information in association with element information, etc., includes, for example, (1) storing element sequence information and element information, etc., in the same record, or directly associating them, and (2) storing them via one or more intermediate pieces of information, such as providing a table for storing element sequence information and intermediate information in association, and another table for storing element information, etc., and intermediate information in association. In other words, any data structure can be adopted as long as it is possible to trace element sequence information from element information, etc. Note that element sequence information only needs to be stored in the storage means in association with element information, etc., and it is not necessarily required to store element information, etc., in the storage means.

[0348] Furthermore, in inventions B1 to B7 and their modifications, the storage means stores element sequence information by any means and at any time. The element sequence information may be stored in advance, or it may not be stored in advance, but rather stored by external input or the like during the operation of the drawing creation support device 100.

[0349] Furthermore, in the sixth to tenth embodiments and their variations described above, the prompt is set to request the AI ​​model 50 to generate element sequence information relating to a plurality of editing elements to be edited consecutively after a created or edited editing element, and their editing order, based on the created or edited element ID, provided that the evaluation value is above a predetermined value. However, the prompt is not limited to this, and the prompt can be set to request the AI ​​model 50 to generate element sequence information relating to a plurality of editing elements to be edited consecutively after a created or edited editing element, and their editing order.

[0350] Furthermore, in the sixth to tenth embodiments and their modifications described above, the training data shown in Figure 5 or Figure 11 was registered in the knowledge base 54. However, the knowledge base 54 can also be used to register (1) editing history data, (2) reference information including element information relating to created or edited elements, and element order information relating to multiple elements to be edited consecutively after the element and their editing order, or (3) reference information including element information relating to created or edited elements, element order information relating to multiple elements to be edited consecutively after the element and their editing order, and evaluation values ​​relating to the editing of the multiple elements. The reference information in (2) or (3) may include element order information estimated using the trained model in the first to fifth embodiments and their modifications described above.

[0351] Furthermore, in the sixth to tenth embodiments and their variations described above, vector data was registered in the knowledge base 54, but the invention is not limited to this, and data in any format can be registered.

[0352] Furthermore, while the first to tenth embodiments and their modifications described above employ a configuration that includes one or more of the following: estimation processing using a trained model, acquisition processing from the AI ​​model 50, and search processing using a table, the system is not limited to these, and a configuration that includes multiple of these processes can be adopted. Specifically, for example, the following configuration can be adopted.

[0353] The first configuration is one in which the processing in step S302 or step S312 is performed by one of the following: estimation processing, acquisition processing, or search processing. For the second and subsequent processing in step S312, it can be performed by the same or different processing as the processing in step S302 or the first processing in step S312.

[0354] The second configuration is one in which the processing in step S336 or step S346 is performed by one of the following: estimation processing, acquisition processing, or search processing. For the second and subsequent processing in step S346, it can be performed by the same or different processing as the processing in step S336 or the first processing in step S346.

[0355] The third configuration is one in which the processing in steps S502 and S504 or the processing in steps S514 and S516 is performed by one of the following: estimation processing, acquisition processing, or search processing. For the second and subsequent processing in steps S514 and S516, the processing can be the same as or different from the processing in steps S502 and S504 or the first processing in steps S514 and S516.

[0356] The fourth configuration is one in which the processing in steps S534 and S536 or the processing in steps S546 and S548 is performed by one of the following: estimation processing, acquisition processing, or search processing. For the second and subsequent processing in steps S546 and S548, the processing can be the same as or different from the processing in steps S534 and S536 or the first processing in steps S546 and S548.

[0357] The fifth configuration is one that controls which processes are prioritized. For example, one can adopt a configuration that (1) prioritizes processes with the lowest current load, (2) prioritizes processes whose results have been adopted by users to a high degree (number of adoptions, percentage, etc.), or (3) prioritizes processes that are used by users to a high degree (number of uses, percentage, etc.).

[0358] Furthermore, in the first to tenth embodiments and their variations described above, multiple editing elements and their editing order were learned or inferred. However, the "multiple editing elements" to be learned or inferred can be editing elements A whose setting or modification requires human judgment, and whose setting or modification affects other editing elements B. This makes it possible to grasp the element order considering the relationship between editing elements A and B. For editing elements that do not require human judgment, editing can be automated using, for example, the technology described in Japanese Patent Publication No. 7341580. Therefore, by targeting editing elements that are difficult to automate, the editing work can be made more efficient.

[0359] Furthermore, while the first to fifth embodiments and their variations described above are implemented as a single device, the system is not limited to this and can also be implemented as a network system. As an example of a network system, some or all of the functions of the drawing support device 100 can be configured as virtual servers on a server that provides cloud computing services.

[0360] Furthermore, in the sixth to tenth embodiments and their modifications described above, the generating AI server 120 integrates the functions of the AI ​​model 50, AI model control unit 52, knowledge base 54, request receiving unit 56, request processing unit 58, response information transmission unit 60, request receiving unit 62, and learning data registration unit 64. However, it is not limited to this configuration, and some functions can be configured on separate servers or the like.

[0361] Furthermore, while the sixth to tenth embodiments and their modifications described above were implemented as a network system, the system is not limited to this and can also be implemented as a single device or application.

[0362] Furthermore, while the sixth to tenth embodiments and their modifications described above describe the case where the invention is applied to a network system consisting of the Internet 199, the invention is not limited to this, and may also be applied to, for example, a so-called intranet that communicates using the same method as the Internet 199. Of course, it is not limited to networks that communicate using the same method as the Internet 199, but can be applied to any network using any communication method.

[0363] Furthermore, in the first to tenth embodiments and their variations described above, the drawing creation support device 100 is configured to utilize the storage device 42, but it is not limited to this, and can also be configured to utilize an external storage device such as a database server.

[0364] Furthermore, in the first to tenth embodiments and their modifications described above, the process shown in the flowcharts of Figures 4, 6, 8, 12, and 18 to 20 was described in the case of executing a program pre-stored in ROM 32. However, the invention is not limited to this, and the program describing these procedures may be read into RAM 34 from a storage medium in which the program is stored and then executed.

[0365] Here, "storage medium" refers to any storage medium that can be read by a computer, regardless of whether it is an electronic, magnetic, or optical reading medium, including semiconductor storage mediums such as RAM and ROM, magnetic storage mediums such as FD and HD, optical reading mediums such as CD, CDV, LD, and DVD, and magnetic / optical reading mediums such as MO.

[0366] Furthermore, the first to tenth embodiments and their variations described above are mutually applicable. Furthermore, the present invention is applicable not only to the first to tenth embodiments and their variations described above, but also to other cases without departing from the spirit of the present invention. For example, the present invention can be applied to a wide range of design work, such as automobile design, machine design, and circuit design (when automobiles, machines, circuits, etc. are the design targets). [Explanation of Symbols]

[0367] 100…Drawing creation support device, 30…CPU, 32…ROM, 34…RAM, 38…I / F, 39…Bus, 40…Input device, 42…Storage device, 44…Display device, 120…Generating AI server, 50…AI model, 52…AI model control unit, 54…Knowledge base, 56, 62…Request receiving unit, 58…Request processing unit, 60…Answer information transmission unit, 64…Learning data registration unit, 199…Internet, 400, 420…Element information, 402…Editing time, 422…Number of editing items, 410, 430…Element ID, 412, 432…Element order information, 414, 434…Evaluation value, 450~454…CAD data, 460~464, 470~474…Trained model, 466…Decision unit

Claims

1. An element information acquisition means for acquiring element information relating to created or edited elements in design information, The system comprises element information relating to created or edited elements, element sequence information relating to a plurality of elements edited consecutively after the element and their editing order (hereinafter, "a plurality of elements and their editing order" is referred to as "element sequence"), and estimation means for estimating the element sequence information from the element information acquired by the element information acquisition means, using a trained model trained on training data including evaluation values ​​relating to the plurality of elements. The design support system is characterized in that the aforementioned evaluation value is an evaluation value relating to the quality or risk of the design target.

2. In claim 1, A design support system characterized in that the evaluation values ​​for the quality of the design target are evaluation values ​​for the safety, functionality, convenience, aesthetics, environmental friendliness, economic efficiency, or legal compliance of the design target.

3. In claim 1, A design support system characterized in that the evaluation value for the risks of the design target is an evaluation value for the risks of the quality of the design target deteriorating, the risk of the design target collapsing or being damaged, the risk of the design target deteriorating, the risk of the costs of the design target increasing, the risks to the design target due to natural disasters, the risk of the asset value of the design target decreasing, or the risk of the design target causing social problems.

4. In any one of claims 1 to 3, A design support system characterized in that the trained model is trained to maximize the evaluation value based on training data including the element information, the element sequence information, and the evaluation value.

5. In claim 4, The trained model includes a first trained model trained to maximize the evaluation value based on training data including the element information, the element sequence information, and the evaluation value based on a first index indicating an index of the value of the evaluation value, and a second trained model trained to maximize the evaluation value based on training data including the element information, the element sequence information, and the evaluation value based on a second index different from the first index. An indicator information acquisition means for acquiring indicator information relating to the first indicator or the second indicator, The system includes a trained model selection means that selects either the first trained model or the second trained model based on the index information acquired by the index information acquisition means, The estimation means is characterized by estimating the element sequence information using the trained model selected by the trained model selection means.

6. In claim 1, The design support system is characterized in that the aforementioned plurality of elements are elements whose setting or modification requires human judgment, and whose setting or modification affects other elements.

7. In claim 1, The design support system is characterized in that the aforementioned design information is design information for designing buildings.

8. A calculation means for calculating an evaluation value of element sequence information based on first design information which includes element sequence information relating to a plurality of elements edited in succession and the editing order thereof (hereinafter referred to as "a plurality of elements and their editing order" as "element sequence") and a first evaluation value is set, and second design information which includes element sequence information relating to the element sequence and a second evaluation value is set, A learning data generation system characterized by comprising a generation means for generating learning data that includes the element sequence information contained in the first design information and the second design information, and evaluation values ​​calculated by the calculation means.

9. In claim 8, The calculation means is Based on the first evaluation value, the evaluation value of the element sequence information included in the first design information is calculated. Based on the second evaluation value, the evaluation value of the element sequence information included in the second design information is calculated. A learning data generation system characterized by calculating an evaluation value of the element sequence information that is commonly included in the first design information and the second design information, based on the first evaluation value and the second evaluation value.

10. The system includes a generation means that generates a trained model by assigning evaluation values ​​to a combination of element information relating to an already created or edited element and element sequence information relating to a plurality of elements that were edited consecutively after that element and their editing order (hereinafter, "a plurality of elements and their editing order" is referred to as "element sequence"), and training the system to maximize the evaluation values. A trained model generation system characterized in that the aforementioned evaluation value is an evaluation value relating to the quality or risk of the design target.

Images