Performance improvement policy prompt system and performance improvement policy prompt method

The system addresses inefficiencies in conventional design optimization by using machine learning to predict and present performance improvement measures, reducing man-hours and enhancing design efficiency.

JP7595622B2Active Publication Date: 2024-12-06HITACHI LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2022151379
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2022-09-22
Publication Date
2024-12-06
Estimated Expiration
2042-09-22

AI Technical Summary

Technical Problem

Conventional design optimization methods require numerous trials and verifications to ensure product characteristics meet target values, leading to increased man-hours and lack of insight into design know-how accumulation.

Method used

A performance improvement plan presentation system utilizing machine learning to predict structural performance and present actionable measures based on past design data, incorporating a learning data section, prediction model, and calculation units to enhance design efficiency.

Benefits of technology

Reduces the number of man-hours required for structural design by providing performance predictions and improvement measures directly to designers, allowing for informed design changes without extensive analyst involvement.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007595622000001
    Figure 0007595622000001
  • Figure 0007595622000002
    Figure 0007595622000002
  • Figure 0007595622000003
    Figure 0007595622000003
Patent Text Reader

Abstract

To provide a performance improving measure presentation system that can present a performance prediction reason and a performance improving measure of a designed product.SOLUTION: A performance improving measure presentation system 100 comprises: a learning data part 110 that has a design parameter which includes an actual value of structural design performed for an object being a target of structural design in the past and an analytic value being a structural performance index of the object executed by structural analysis on the basis of the design parameter as learning data of machine learning; and a program part 120 that predicts the structural performance index of the object from the design parameter input by a design change using a prediction model which is obtained by performing machine learning of the design parameter and analytic value read from the learning data part 110, calculates an interpretation possibility representing explain-ability of a result contributing to the structural performance index predicted by the prediction model, and presents a performance improving measure for improving performance of the object obtained from the interpretation possibility together with the structural performance index predicted by the prediction model.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a system and a method for presenting performance improvement measures. [Background technology]

[0002] In recent years, product design utilizing machine learning has been researched. In product development, in a situation where multiple objective variables (characteristics) for a product are given, the multiple objective variables are optimized in order to improve the multiple characteristics that change according to the design variables. In the design process of a product, etc., a technology described in Patent Document 1 is known that enables the optimization of the characteristics of the product, etc., that constitute the objective variables, and the design variables with a low load.

[0003] This patent document 1 describes a "design support device that determines multiple design parameters that satisfy target values ​​set for each of multiple characteristic items that indicate the characteristics of a product, work-in-progress, semi-finished product, part or prototype, in order to be applied to a method of optimizing design parameters by repeatedly determining design parameters and producing the product, work-in-progress, semi-finished product, part or prototype based on a group of design parameters consisting of multiple design parameters." [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2022-92308 Summary of the Invention [Problem to be solved by the invention]

[0005] By using the technology described in Patent Document 1, a designer would repeatedly input design parameters and past performance data and perform experiments to confirm that any characteristic of a product satisfies a target value or to bring the characteristic of a product closer to the target value, thereby predicting desirable design parameters. Therefore, the designer had to perform multiple trial and verifications until the product satisfied the requirements, which led to an increase in the number of steps required for structural design.

[0006] In addition, in the technology described in Patent Document 1, an acquisition function is calculated to output an index value indicating the suitability of a group of design parameters as a solution. However, simply calculating the acquisition function only shows the solution of the group of design parameters to the designer. For this reason, the conventional technology is unable to provide hidden insight to the designer, and it is difficult to accumulate know-how for design.

[0007] The present invention has been made in view of the above circumstances, and has an object to predict the performance of an object for which structural design is being carried out, and to propose measures for improving the performance. [Means for solving the problem]

[0008] The performance improvement measure presentation system of the present invention comprises a learning data section having, as learning data for machine learning, design parameters including actual values ​​of structural designs performed in the past on an object that is the subject of structural design, and analytical values ​​that are indicators of the structural performance of the object for which a structural analysis was performed based on the design parameters, and a calculation section that predicts an indicator of the structural performance of the object from design parameters input due to a design change using a predictive model obtained by machine learning the design parameters and the analytical values ​​read out from the learning data section, calculates an interpretability that represents the explainability of the results contributing to the structural performance indicator predicted by the predictive model, and presents performance improvement measures for improving the performance of the object obtained from the interpretability, together with the structural performance indicator predicted by the predictive model. Effect of the Invention

[0009] According to the present invention, performance improvement measures for improving the performance of an object can be presented together with indicators of structural performance predicted by a predictive model. Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0010] [Figure 1] 1 is a block diagram showing an example of the overall configuration of a performance improvement measure presentation system in accordance with an embodiment of the present invention, in which performance improvement measures are presented and a product design change is made; [Diagram 2] FIG. 2 is a block diagram showing an example of the hardware configuration of a computer according to an embodiment of the present invention. [Diagram 3] 4 is a diagram showing an example of a table configuration of a design and analysis results DB according to one embodiment of the present invention; FIG. [Figure 4] 4 is a diagram showing an example of a table configuration of a performance improvement measure learning information DB according to one embodiment of the present invention; FIG. [Diagram 5] 1 is a block diagram showing an example of the overall configuration of a performance improvement measure presentation system in a case where a designer finishes a design without requesting the presentation of performance improvement measures, according to an embodiment of the present invention; [Figure 6] 1 is a block diagram showing an example of the overall configuration of a performance improvement measure presentation system in accordance with an embodiment of the present invention, in which a designer rejects a presented performance improvement measure and terminates the design; [Figure 7] 1 is a flowchart showing an example of a series of processes of a performance improvement measure presentation method performed in a performance improvement measure presentation system according to an embodiment of the present invention. [Figure 8] 1 is a flowchart illustrating an example of a learning process of a performance prediction program according to an embodiment of the present invention. [Figure 9] 1 is a flowchart illustrating an example of a prediction process of a performance prediction program according to an embodiment of the present invention. [Figure 10] 11 is a flowchart illustrating an example of a learning process of an interpretability calculation program according to an embodiment of the present invention. [Figure 11]11 is a flowchart illustrating an example of a prediction process of an interpretability calculation program according to an embodiment of the present invention. [Figure 12] 10 is a flowchart showing an example of a learning process of a performance improvement measure calculation program according to an embodiment of the present invention. [Figure 13] 10 is a flowchart illustrating an example of a prediction process of a performance improvement measure calculation program according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functions or configurations are designated by the same reference numerals, and redundant description will be omitted.

[0012] [One embodiment] <Example of the configuration of the performance improvement proposal system> FIG. 1 is a block diagram showing an example of the overall configuration of a system 100 for presenting measures to improve performance in the structural design of a product. FIG. 1 shows an example of the configuration and operation of performance improvement measure presentation system 100 in the case where a designer 130 checks a predicted performance value, requests a program unit 120 to present a performance improvement measure, adopts the performance improvement measure presented by the program unit 120, and modifies the product design.

[0013] A structure (object) that is the subject of structural design is called a product. One of the performance characteristics of a structure is called strength performance. Strength performance is, for example, the seismic resistance performance of a product that can withstand an earthquake, and a product with high strength performance is less likely to be damaged in the event of an earthquake. Performance improvement measures are information predicted by the program unit 120 and presented to the designer 130, and it is expected that the performance of the structure will be improved by reflecting the performance improvement measures in the structural design. However, it is up to the designer 130 to decide whether to incorporate the performance improvement measures and perform the structural design of the product again.

[0014] The performance improvement measure presentation system 100 includes a learning data unit 110 that accumulates learning data for the program unit 120 to perform machine learning, a program unit 120 that performs actual processing, and a designer PC (Personal Computer) 140. The program unit 120 has a function of performing predetermined calculations by executing various programs, and is used as an example of a calculation unit. The configuration of the program unit 120 will be described later.

[0015] Operation input to the program section 120, screens output from the program section 120, etc. are performed on a designer PC 140 used by a designer 130. The designer 130 shown in Fig. 1 and Figs. 5 and 6 described later can check the processing contents of the program section 120 through a screen displayed on a display device 55 (see Fig. 2 described later) of the designer PC 140, and can input instructions to the program section 120 through an input device of the designer PC 140 and make decisions regarding design changes.

[0016] <Example of the configuration of the learning data section> The learning data unit 110 has, as learning data for machine learning, design parameters including performance values ​​of structural designs performed in the past on an object that is the subject of structural design, and analysis values ​​that are indexes of the structural performance of the object for which structural analysis was performed based on the design parameters. The learning data unit 110 has a design and analysis performance database (DB) 11 and a performance improvement measure learning information database 12. In the following description, these are abbreviated as design and analysis performance DB11 and performance improvement measure learning information DB12.

[0017] The design and analysis results DB11 stores the results of structural designs and structural analysis values ​​performed in the past. In other words, the design and analysis results DB11 stores past design results. The past design results are data in which design parameters such as a product ID for identifying a product, the external dimensions and mass of the product, and a fastening method for the product to a base are linked to analysis values ​​of performance indices such as the product's natural frequency. In this embodiment, a performance index for evaluating the strength of a structure is required. In this embodiment, the product's natural frequency is adopted as a performance index so that the product does not resonate at its natural frequency due to the effects of an earthquake or the like.

[0018] In this embodiment, for example, a control panel is assumed as an object of structural design by the designer 130. Control panels are used to control the operation of each facility and system in, for example, social infrastructure facilities (power plants, ironworks, water supply and sewerage, etc.), power generation systems, water supply and sewerage systems, etc. There are also control panels that control the operation of turbines in power plants, etc.

[0019] Usually, the designer 130 who performs the structural design of a product and the analyst who performs the structural analysis are different persons. After the designer 130 performs the structural design, he / she passes the structural design information (such as CAD (Computer Aided Design) data) to an analyst (not shown). The analyst performs the structural analysis based on the structural design information received from the designer 130. The result of the structural analysis is stored in the design and analysis results DB 11 as a structural analysis value. Since the structural design and the structural analysis are performed by different persons in this way, it may take time for the designer 130 to obtain the analysis result even for a slight modification of the structural design. Therefore, by using the performance improvement measure presentation system 100 according to this embodiment, the designer 130 can obtain the performance prediction result and the performance improvement measure after performing the structural design and before requesting the analyst to perform the structural analysis.

[0020] The performance improvement measure learning information DB12 stores learning data to which performance improvement measures have been added. The learning data stored in the performance improvement measure learning information DB12 stores design information before and after a design change and performance values ​​extracted from the information in the design and analysis results DB11 of the performance improvement measures implemented. For this reason, the learning data unit 110 creates the performance improvement measure learning information DB12 including information before and after a product design change and performance improvement measures for the information after the design change, both extracted from the design and analysis results DB11. The learning data stored in the design and analysis results DB11 and the performance improvement measure learning information DB12 are also called learning information.

[0021] <Example of computer hardware configuration> Next, the hardware configuration of the computer 50 constituting each device of the performance improvement measure presentation system 100 will be described. Fig. 2 is a block diagram showing an example of the hardware configuration of calculator 50. Calculator 50 is an example of hardware used as a computer capable of operating as performance improvement measure presentation system 100 according to this embodiment. In performance improvement measure presentation system 100 according to this embodiment, calculator 50 (computer) executes a program to realize a performance improvement measure presentation method in which the functional blocks shown in Fig. 1 work together.

[0022] The computer 50 includes a CPU (Central Processing Unit) 51, a ROM (Read Only Memory) 52, and a RAM (Random Access Memory) 53, each of which is connected to a bus 54. The computer 50 further includes a display device 55, an input device 56, a storage device 57, and a network interface 58.

[0023] The CPU 51 reads out program code of software for realizing each function according to the present embodiment from the ROM 52, loads it into the RAM 53, and executes it. Variables, parameters, etc. generated during the arithmetic processing of the CPU 51 are temporarily written into the RAM 53, and these variables, parameters, etc. are read out by the CPU 51 as appropriate. However, an MPU (Micro Processing Unit) may be used instead of the CPU 51. The functions of each program and each output section of the program section 120 shown in FIG. 1 are realized by the arithmetic processing of the CPU 51.

[0024] The display device 55 is, for example, a liquid crystal display monitor, and displays the results of processing performed by the computer 50 to the designer 130. The input device 56 is, for example, a keyboard, a mouse, etc., and enables the designer 130 to input predetermined operations and give instructions. The performance prediction button 2 and the performance improvement measure presentation button 4 shown in FIG. 1 may be operation icons displayed on the display device 55 instead of physical buttons.

[0025] As the storage device 57, for example, a non-volatile recording medium such as a hard disk drive (HDD), a solid state drive (SSD), a flexible disk, an optical disk, a magneto-optical disk, a CD-ROM, a CD-R, a magnetic tape, or a non-volatile memory is used. In addition to an operating system (OS) and various parameters, a program for making the computer 50 function is recorded in the storage device 57. The ROM 52 and the storage device 57 record programs and data necessary for the CPU 51 to operate, and are used as an example of a computer-readable non-transient storage medium storing a program executed by the computer 50. Each database of the learning data unit 110 shown in FIG. 1 is configured in the storage device 57.

[0026] The network interface 58 may be, for example, a network interface card (NIC), and various data may be transmitted and received between devices via a local area network (LAN) or dedicated line connected to a terminal of the NIC.

[0027] Here, examples of table configurations of the design and analysis results DB 11 and the performance improvement measures learning information DB 12 will be described with reference to FIG. 3 and FIG. FIG. 3 is a diagram showing an example of a table configuration of the design and analysis results DB 11. As shown in FIG.

[0028] The design and analysis results DB11 stores design parameters and analysis values, which are the results of structural designs performed on objects in the past. This design and analysis results DB11 has items such as labels, external dimensions, mass, fastening method, and natural frequency. Information for each item stored in the design and analysis results DB11 is the design values ​​and analysis results of the actual product. The fastening method is, for example, a method of fastening based on the product, and riveting is assumed.

[0029] The label item stores a unique label, such as a product name or model, that is assigned to identify the product. In this embodiment, it is assumed that there are three types of products with labels "AAA", "BBB", and "CCC".

[0030] The outer dimensions item stores outer dimensions (an example of outer dimension information) including the width, height, and depth of the product. The mass item stores the mass of the product. The external dimensions and mass of the product are features and explanatory variables in machine learning, and are used as important factors in predicting product performance. This important factor in predicting performance is related to the natural frequency, which is an index of strength performance.

[0031] The fastening method item stores the fastening method for fastening the product to the base in the form of a code. There are multiple types of fastening methods, and in this embodiment, it is assumed that there are three types of fastening methods, "1" to "3".

[0032] The natural frequency item stores the natural frequency in the left-right direction and the natural frequency in the front-back direction that are predefined for the product as an example of an index of the product's structural performance (called a "performance index"). The explanatory variables in machine learning are design parameters including the external dimensions, weight, and fastening method, and the relationship between the explanatory variables and the natural frequency, which is the objective variable, is calculated by machine learning.

[0033] Next, an example of a table configuration of the performance improvement measure learning information DB 12 will be described with reference to FIG. FIG. 4 is a diagram showing an example of a table configuration of the performance improvement measure learning information DB 12. As shown in FIG.

[0034] The performance improvement measure learning information DB12 stores design parameters and analysis values ​​before a design change (combinations that are in a relationship between before and after a change that was actually performed in the past) extracted for each object from the design and analysis results DB11, and performance improvement measures for the object after a design change (performance improvement measures that were actually performed in the past that caused changes in the design parameters and performance indexes before and after the change in each combination). For this reason, the performance improvement measure learning information DB12 stores data (design parameters) that are in a relationship between before and after the change, extracted from the design and analysis results DB11 shown in Fig. 3. And, analysis results values ​​are stored as data after the change in the performance improvement measure learning information DB12.

[0035] Moreover, the performance improvement measure learning information DB12 stores performance improvement measures implemented in the process of changing design parameters. The performance improvement measures include a number of items such as part addition and processing, and a numerical value is stored for each item of the implemented performance improvement measure. Part addition represents, for example, the measure of adding a cover to protect the product.

[0036] The performance improvement measure learning information DB12 has major items before and after the change shown in the upper part of Fig. 4, and major items of performance improvement measures shown in the lower part of Fig. 4. The major items in the upper and lower parts of Fig. 4 are continuous and represent the contents of one piece of data. In Fig. 4, the information stored in the performance improvement measure learning information DB12 will be explained by focusing on the product with the label "AAA".

[0037] The pre-change items and post-change items store information on each item in the design and analysis results DB11 shown in Fig. 3 (combinations showing the relationship between before and after changes that were actually made in the past). That is, information after the product's structural design has actually been changed is stored in the post-change items, and the performance improvement measures adopted in the product after the change are stored in the performance improvement measures items.

[0038] The information on the product labeled "AAA" stored in the Before Change item is the same for all three records. On the other hand, the information on the product labeled "AAA" stored in the Before Change item is different in the hatched cells in the figure. For example, the external dimensions and fastening method of the product assigned the label "DDD" after the change are the same as the product labeled "AAA", but the mass and the natural frequencies for the left and right and front and rear are different.

[0039] In addition, after the change, the width, depth, and front-to-back natural frequencies of products assigned the "EEE" label will be the same as those of products labeled "AAA," but the height, mass, fastening method, and left-to-right natural frequencies will be different. In addition, the height and depth of products assigned the label "FFF" after the change are the same as those labeled "AAA", but the width, mass, fastening method, and natural frequencies for the left and right and front and rear are different.

[0040] The major items for performance improvement measures include adding parts, processing, reinforcement, fastening method, material, plate thickness, center of gravity, and mounting position. For example, a "1" is stored in the part addition item as a performance improvement measure for a product to which the label "DDD" is assigned. This indicates that adding a part associated with part addition "1" is one of the performance improvement measures. In addition, "2" is stored in the reinforcement item as a performance improvement measure for a product to which the label "EEE" is assigned. This indicates that taking the reinforcement measure associated with reinforcement "2" is one of the performance improvement measures. In addition, as performance improvement measures for a product to which the label "FFF" is assigned, "1" is stored in the part addition item and "1" is stored in the fastening method item. This indicates that adding a part associated with part addition "1" and implementing the fastening method associated with fastening method "1" are performance improvement measures.

[0041] 4 are expressed as numerical values, it is preferable that a message corresponding to the numerical value is displayed when the performance improvement measure is presented to the designer 130. By reading the message displayed on the display device 55, the designer 130 can specifically grasp the performance improvement measure and can determine whether or not to adopt the presented performance improvement measure.

[0042] 3 and 4 show an example in which the natural frequency is used as an index of structural performance to be predicted by the performance prediction program 21 shown in Fig. 1. However, other than the natural frequency, for example, either the weight of the target object or the density of components mounted on the target object may be used as an index of structural performance to be predicted.

[0043] <Program section configuration example> Next, returning to FIG. 1, an example of the configuration of the program unit 120 will be described. The program unit 120 predicts the structural performance index of the object from the design parameters input due to the design change, using a prediction model obtained by machine learning the design parameters and analysis values ​​read from the learning data unit 110. The program unit 120 also calculates interpretability, which indicates the explainability of the results that contribute to the structural performance index predicted by the prediction model. Then, the program unit 120 presents performance improvement measures for improving the performance of the object, which are obtained from the interpretability, together with the structural performance index predicted by the prediction model. In this way, the program unit 120 can read various information from the design and analysis results DB 11 and the performance improvement measure learning information DB 12, and present the performance improvement measures for the structural design to the designer 130.

[0044] The program unit 120 includes a performance prediction program 21 (an example of a performance prediction unit), an interpretability calculation program 22 (an example of an interpretability calculation unit), a performance improvement measure calculation program 23 (an example of a performance improvement measure calculation unit), a prediction result output unit 3, and a performance improvement measure output unit 5. The performance prediction program 21, the interpretability calculation program 22, and the performance improvement measure calculation program 23 are all programs obtained by performing machine learning based on information read out from the design and analysis results DB11 and the performance improvement measure learning information DB12.

[0045] The performance prediction program 21 is a program that performs machine learning based on the design parameters and analysis values ​​before the design change read from the design and analysis results DB 11 to create a first prediction model, and predicts an index of structural performance using the first prediction model. The performance prediction program 21 selects XGboost (eXtreme Gradient Boosting) based on the design and analysis results DB 11 to enable more accurate prediction of performance indexes with less data. Then, the performance prediction program 21 creates the first prediction model using machine learning using XGboost. When the performance prediction program 21 predicts performance, the first prediction model is used.

[0046] XGboost is a machine learning method that uses decision trees as models trained on data. In XGboost, multiple models are trained based on a portion of data extracted from a large amount of data. With XGboost, data that is incorrectly recognized by a certain decision tree is retrained with another decision tree so that it can be prioritized for correct classification, and one model is finally obtained. The one model finally obtained by XGboost is used as the first prediction model.

[0047] The prediction result output unit 3 outputs the structural performance index predicted by the performance prediction program 21 as the prediction result to the display device 55 of the designer PC 140. When the designer 130 checks the prediction result of the performance index output to the display device 55, he / she inputs an instruction to the program unit 120 to present a performance improvement measure of the product by pressing the performance improvement measure presentation button 4 displayed on the display device 55. The instruction to present the performance improvement measure input from the performance improvement measure presentation button 4 is input to the interpretability calculation program 22.

[0048] The interpretability calculation program 22 is a program that performs machine learning based on the design parameters and analysis values ​​before the design change read from the design and analysis results DB 11 to create a second prediction model, and calculates the importance of the feature amount of the first prediction model, the explainability of the first prediction model that influences the prediction result, and the interpretability using the second prediction model. The interpretability calculation program 22 can calculate the contribution and correlation of each design parameter to the performance index predicted by the performance prediction program 21. When the interpretability calculation program 22 calculates the interpretability, the second prediction model is used.

[0049] The interpretability represents a numerical value that is the basis for the performance prediction, and examples of the interpretability include the external dimensions, mass, and natural frequency of the product shown in the design and analysis results DB 11. Note that the section modulus, which is one of the indicators related to the strength of the product, may also be used as the interpretability. The interpretability may also be represented by a message that describes problems with the structural design that correspond to the numerical value. In this case, the designer 130 can grasp the specific problems with the structural design by checking the message instead of the numerical value.

[0050] The program unit 120 uses, for example, SHAP (SHapley Additive exPlanations) as an optimal interpretability technique for the tabular data stored in the design and analysis results DB 11, and creates an interpretability calculation program 22 by machine learning based on the design and analysis results DB 11. SHAP is one of the XAI (Explainable AI) techniques, and can determine how much each element affects a prediction. By using SHAP as the interpretability calculation program 22, the program unit 120 can determine how much each feature of certain data X positively or negatively affects the performance index predicted by the performance prediction program 21.

[0051] The performance improvement measure calculation program 23 performs machine learning based on the design parameters before the design change, the design parameters after the design change, and the performance improvement measures read from the performance improvement measure learning information DB 12 to create a third prediction model. The performance improvement measure calculation program 23 is a program that calculates a performance improvement measure using a third prediction model based on the design parameters before the design change, the design parameters after the design change, and the deviation of the structural performance index predicted by the first prediction model after the design change from the analysis value before the design change. This performance improvement measure calculation program 23 presents a measure (called a "performance improvement measure") for improving the structural performance of the product to the designer based on the difference in the performance index before and after the design change of the product and the change in the design parameters. When the performance improvement measure calculation program 23 calculates the performance improvement measure, the third prediction model is used.

[0052] The performance improvement measure calculation program 23 creates multiple random forests based on the performance improvement measure learning information DB12, and creates a third prediction model by utilizing machine learning using the random forest with the highest accuracy. The random forest is a method that uses decision trees, and the performance improvement measure calculation program 23 randomly selects features and creates multiple decision trees. Then, the output of the third prediction model is determined by obtaining a majority vote using the prediction results of the multiple decision trees. The performance improvement measures calculated by the performance improvement measure calculation program 23 are stored in the performance improvement measure item of the performance improvement measure learning information DB12.

[0053] The performance improvement measure output unit 5 outputs the interpretability and the performance improvement measures calculated by the performance improvement measure calculation program 23 to the display device 55 of the designer PC 140. The designer 130 can check the performance improvement measures and make changes to the product design.

[0054] Next, a description will be given of the designer 130. The block shown as the designer 130 actually represents the functionality of the designer PC 140 operated by the designer 130.

[0055] The designer 130 inputs design change information 1, which represents a change to the design of the product, into the performance prediction program 21 through an input device of the designer PC 140. The design change information 1 includes, for example, at least one piece of information among the product's external dimensions, mass, fastening method, and natural frequency. Then, the designer 130 presses the performance prediction button 2. At the same time that an instruction is input from the performance prediction button 2, execution of the performance prediction program 21 is started.

[0056] Thereafter, as described above, the performance prediction result by the performance prediction program 21 is displayed on the display device 55 of the designer PC 140 through the prediction result output unit 3. The designer 130 checks the performance prediction result, and presses the performance improvement measure presentation button 4 as necessary. At the same time that an instruction is input from this performance improvement measure presentation button 4, the interpretability calculation program 22 is executed. When the performance improvement measure calculation program 23 is executed after the interpretation possibility calculation program 22 is executed, the performance improvement measure calculated by the performance improvement measure calculation program 23 is displayed on the display device 55 of the designer PC 140 through the performance improvement measure output unit 5. The designer 130 checks the performance improvement measure, and again changes the design of the product based on the performance improvement measure. Then, the designer 130 can input the change information of the design change to the performance prediction program 21 again.

[0057] <Process flow for terminating the design without requesting the presentation of performance improvement measures> Next, a process flow in which the designer 130 checks the predicted values ​​of the product's performance and, if the designer 130 determines that the performance is sufficient, ends the design without requesting the presentation of performance improvement measures will be described with reference to FIG.

[0058] Fig. 5 is a diagram showing an example of the overall configuration of the performance improvement measure presentation system 100 when the design is completed without the designer 130 requesting the presentation of performance improvement measures. Since the configuration of the performance improvement measure presentation system 100 is as described with reference to Fig. 1, detailed description of the parts common to Fig. 1 and Fig. 5 will be omitted.

[0059] The designer 130 checks the prediction results of the performance prediction program 21 via the prediction result output unit 3. If the designer 130 determines that the prediction results output from the prediction result output unit 3 satisfy the required performance, the designer 130 ends the design 6. At this time, the designer 130 instructs the program unit 120 to end the design via the designer PC 140. Thereafter, the design data changed according to the design change information 1 is passed to the analyst, who analyzes the structure whose design has been changed.

[0060] <Process flow for rejecting performance improvement measures and terminating the design> Next, a case will be described with reference to FIG. 6 in which the designer 130 checks the predicted values ​​of the product's performance, requests performance improvement measures, and then rejects the performance improvement measures presented by the program unit 120 and terminates the design.

[0061] Fig. 6 is a diagram showing an example of the overall configuration of the performance improvement measure presentation system 100 when the designer 130 rejects the presented performance improvement measures and ends the design. Since the configuration of the performance improvement measure presentation system 100 is as described with reference to Fig. 1, detailed description of the parts common to Fig. 1 and Fig. 6 will be omitted.

[0062] The designer 130 checks the performance improvement measures calculated by the performance improvement measure calculation program 23 and presented via the performance improvement measure output unit 5. Then, if the designer 130 rejects the outputted performance improvement measures because they have already been implemented or because they do not satisfy various design requirements, the designer 130 ends the design 7. At this time, the designer 130 instructs the program unit 120 to end the design via the designer PC 140. Thereafter, the design data that has been modified by the design change information 1 and reflects the performance improvement measures is passed to the analyst, who then analyzes the structure whose design has been modified.

[0063] If the designer 130 accepts the confirmed performance improvement measures, in the subsequent processing, the designer 130 inputs new design change information 1 to the program unit 120. For example, the designer 130 can input the performance improvement measures presented by the program unit 120 through a design support program such as CAD (not shown) to perform a product design that reflects the contents of the performance improvement measures. Conversely, if the designer 130 does not accept the confirmed performance improvement measures, the contents of the performance improvement measures presented by the program unit 120 are not reflected in the design support program such as CAD (not shown).

[0064] Next, an example of the processing of the performance improvement measure presentation system 100 will be described with reference to FIG. 7 is a flowchart showing an example of a series of processes of the performance improvement measure suggestion method performed by the performance improvement measure suggestion system 100. Here, the operation procedure of the designer 130 as a user of the performance improvement measure suggestion system 100 is shown.

[0065] First, the designer 130 starts the performance improvement measure presentation system 100 (for example, a macro function of a spreadsheet software) and inputs design change information 1 for the structure to be designed (S1). The design change information 1 includes various design parameters.

[0066] Next, the designer 130 presses the performance prediction button 2 (S2) to execute the performance prediction program 21. Next, the designer 130 checks the performance prediction result, which is the predicted value of the performance index calculated by the performance prediction program 21 and output from the prediction result output unit 3 (S3).

[0067] Next, the designer 130 compares the performance prediction results with the design specifications and determines whether or not it is necessary to reconsider the performance improvement measures (S4). If the designer 130 determines that it is not necessary to consider the performance improvement measures (NO in S4), the design is terminated. Note that the configuration diagram shown in FIG. 5 represents the NO determination in step S4.

[0068] On the other hand, if the designer 130 judges that it is necessary to consider performance improvement measures (YES in S4), he or she presses the performance improvement measures presentation button 4 (S5) and executes the interpretability calculation program 22 and the performance improvement measures calculation program 23.

[0069] Next, the designer 130 checks the interpretability and the performance improvement measures presented by the performance improvement measure output unit 5 (S6). Then, the designer 130 judges whether or not to adopt the presented performance improvement measures (S7). If the designer 130 judges that it is not necessary to adopt the presented performance improvement measures (NO in S7), the designer 130 ends the design. Note that the configuration diagram shown in FIG. 6 represents the NO judgment in step S7. The NO judgment in step S7 also includes the case where the performance improvement measures have been adopted once or more before reaching step S7 (YES in S7), and the processes of steps S1 to S6 have been performed again, so that it is judged that it is not necessary to adopt any more performance improvement measures.

[0070] On the other hand, if the designer 130 determines that it is necessary to adopt performance improvement measures (YES in S7), the process returns to step S1, where the designer 130 modifies the structural design and inputs the modified design change information 1 again to the program unit 120. In this way, the designer 130 modifies the structural design and inputs the design change information 1, thereby making it possible to narrow down the performance improvement measures. This eliminates the need to request a structural analysis from an analyst other than the designer 130 every time there is a design change or minor design modification, which is expected to reduce the number of work steps for the designer 130 and reduce backtracking.

[0071] Next, an example of the processing of each program executed by the program unit 120 will be described with reference to Fig. 8 to Fig. 13. In the following explanation, the processing of each program will be described separately for a learning stage and a prediction stage.

[0072] <Performance prediction program processing> First, the learning stage and prediction stage of the performance prediction program 21 will be described with reference to FIGS. FIG. 8 is a flowchart showing an example of the learning process of the performance prediction program 21.

[0073] In the learning stage, the performance prediction program 21 reads (S11) learning information from the design and analysis results DB 11. The learning information read by the performance prediction program 21 is any data included in the design and analysis results DB 11 shown in FIG.

[0074] Next, the performance prediction program 21 performs data standardization processing as preprocessing (S12). The data standardization processing is, for example, a normalization method that converts the average of the original data to "0" and the standard deviation to "1". The data of various types of products is normalized by the data standardization processing. Next, the performance prediction program 21 performs machine learning using Xgboost based on the relationship between the explanatory variables and the objective variables described in FIG. 3 (S13).

[0075] Then, the performance prediction program 21 saves the trained model created in the machine learning process of step S13 (S14), and ends this process. The trained model created in the machine learning process of the performance prediction program 21 is a first prediction model capable of predicting the performance of the product. The first prediction model is saved, for example, in a predetermined area of ​​the storage device 57 that is readable and writable by the program unit 120.

[0076] FIG. 9 is a flowchart showing an example of the prediction process of the performance prediction program 21.

[0077] In the prediction stage, the performance prediction program 21 reads input information input by the designer 130 (S21). This input information includes design change information 1, which is a design parameter, and an input instruction for the performance prediction button 2. Next, the performance prediction program 21 reads the first prediction model created in the learning stage from the storage device 57 (S22).

[0078] Next, the performance prediction program 21 inputs the input information into the first prediction model, and calculates the performance index based on the first prediction model (S23). As shown in Fig. 3, the performance index includes the natural frequency in the left-right direction and the natural frequency in the front-rear direction that are predefined for the product.

[0079] Then, the performance prediction program 21 outputs the calculation results, which are the performance prediction results, to the prediction result output unit 3 (S24). The calculation results are displayed on the display device 55 through the prediction result output unit 3, so that the designer 130 can check the calculation results.

[0080] <Processing of interpretability calculation program> Next, the processing in the learning stage and the prediction stage of the interpretability calculation program 22 will be described with reference to FIGS. FIG. 10 is a flowchart showing an example of the learning process of the interpretability calculation program 22.

[0081] In the learning stage, the interpretability calculation program 22 reads learning information from the design and analysis results DB 11 (S31). The learning information read by the interpretability calculation program 22 is any data included in the design and analysis results DB 11 shown in FIG.

[0082] Next, the interpretability calculation program 22 performs a data standardization process as preprocessing (S32). The data standardization process ensures that data for various types of products falls within a certain range. Next, the interpretability calculation program 22 performs machine learning using SHAP on the relationship between the explanatory variables and the objective variables described in FIG. 3 (S33).

[0083] Then, the interpretability calculation program 22 saves the trained model created in the machine learning process of step S13 (S34) and ends this process. The trained model created in the machine learning process of the interpretability calculation program 22 is a second prediction model capable of calculating interpretability. The second prediction model is saved, for example, in a predetermined area of ​​the storage device 57 that is readable and writable by the program unit 120.

[0084] FIG. 11 is a flowchart showing an example of the prediction process of the interpretability calculation program 22.

[0085] In the prediction stage, the interpretability calculation program 22 reads input information input by the designer 130 (S41). This input information includes design change information 1, which is a design parameter, and an input instruction for the performance improvement measure presentation button 4. Next, the interpretability calculation program 22 reads the second prediction model created in the learning stage from the storage device 57 (S42).

[0086] Next, the interpretability calculation program 22 inputs the input information into the second prediction model and calculates the interpretability based on the second prediction model (S43). This calculation determines the importance of the feature that contributes to the performance index predicted by the first prediction model. Next, the interpretability calculation program 22 stores the calculation result of the interpretability in a predetermined area of ​​the storage device 57 that is readable and writable by the program unit 120 (S44).

[0087] Then, the interpretability calculation program 22 generates a diagram that visualizes the interpretability, and stores this diagram in a predetermined area of ​​the storage device 57 (S45), and then ends this process.

[0088] <Processing of performance improvement calculation program> Next, the processing in the learning stage and prediction stage of the performance improvement measure calculation program 23 will be described with reference to FIG. 12 and FIG. FIG. 12 is a flowchart showing an example of the learning process of the performance improvement measure calculation program 23.

[0089] In the learning stage, the performance improvement measure calculation program 23 reads learning information from the performance improvement measure learning information DB 12 (S51). The learning information read by the performance improvement measure calculation program 23 is any data included in the performance improvement measure learning information DB 12 shown in FIG.

[0090] Next, the performance improvement measure calculation program 23 creates deviation data, which is the deviation between the design parameters and performance index before and after the change for the explanatory variables explained in Fig. 4 (S52). For the product with the label "DDD" in the performance improvement measure learning information DB 12 shown in Fig. 4, the mass "390" before the change becomes "400" after the change, so the deviation between the design parameters before and after the change is mass "10". Also, the natural frequency, which is the performance index before the change, changes from left and right "5.5" and front and rear "10" to left and right "6.0" and front and rear "11" after the change, so the deviation of the performance index before and after the change is left and right "0.5" and front and rear "1". Deviation data summarizing these deviations is created in the processing of step S52.

[0091] Next, the performance improvement measure calculation program 23 performs machine learning using random forest on the relationship between the explanatory variables and the objective variables described with reference to FIG. 3 (S53).

[0092] Then, the performance improvement measure calculation program 23 saves the trained model created in the machine learning process of step S53 (S54), and ends this process. The trained model created in the machine learning process of the performance improvement measure calculation program 23 is a third prediction model capable of predicting the performance improvement measures of the product. The third prediction model is saved, for example, in a predetermined area of ​​the storage device 57 that is readable and writable by the program unit 120.

[0093] FIG. 13 is a flowchart showing an example of the prediction process of the performance improvement measure calculation program 23.

[0094] In the prediction stage, the performance improvement measure calculation program 23 reads pre-change data stored in the major item before the change from the performance improvement measure learning information DB 12 (S61). Next, the performance improvement measure calculation program 23 reads the change input value of the design parameter input by the designer 130 (S62). The change input value of the design parameter is the post-change data stored in the major item after the change stored in the performance improvement measure learning information DB 12. Next, the performance improvement measure calculation program 23 reads the third prediction model created in the learning process shown in Fig. 12 from the storage device 57 (S63).

[0095] Next, the performance improvement measure calculation program 23 uses the third prediction model to predict the performance of the product based on the changed design parameters (S64). Next, the performance improvement measure calculation program 23 calculates the deviation of the design parameters and the deviation of the predicted value of the performance index (S65).

[0096] Next, the performance improvement measure calculation program 23 inputs each deviation calculated in step S65 into the third prediction model, and calculates performance improvement measures (performance indexes) based on the third prediction model (S66). Then, the performance improvement measure calculation program 23 displays the calculation results of the performance improvement measures and the calculation results of the interpretability stored by the interpretability calculation program 22 on the display device 55 through the performance improvement measure output unit 5, and displays a diagram visualizing the calculation results of the interpretability as necessary (S67). The designer 130 can select information to be displayed on the display device 55 as necessary.

[0097] In the performance improvement measure presentation system 100 according to the embodiment described above, not only is performance predicted in relation to changes in the structural design of a product, but also interpretability and performance improvement measures are calculated. At this time, the program unit 120 performs machine learning on structural performance improvement measures previously applied to the product, and the basis of the performance prediction is presented to the designer 130. Therefore, after the designer 130 performs the structural design of the product, the program unit 120 can be instructed to predict performance. As a result, it becomes possible for the designer 130 to check the performance prediction result after the design change without requesting a structural analysis from an analyst, shortening the time required to check the performance prediction result after the design change and reducing the number of work steps. In addition, the number of times of work required for structural analysis and various procedures for quality assurance of the product after the design change can be reduced.

[0098] Furthermore, when the designer 130 instructs the program unit 120 to present a performance improvement measure, the interpretability that is the basis of the performance prediction and the performance improvement measure are presented to the designer 130. Therefore, the designer 130 can determine whether or not to apply the performance improvement measure after checking the interpretability.

[0099] Conventionally, when changing the design of a product, an analyst analyzes the product after the structural design, and if there is a problem with the analysis result, the designer 130 considers modifying the design. Then, the designer 130 finds and modifies improvements based on his / her tacit knowledge and experience, and this requires trial and error by the designer 130. On the other hand, in the performance improvement measure presentation system 100 according to the present embodiment, the designer 130 only needs to make design changes by adding data (e.g., design change information 1) that has been modified in accordance with his / her desire to, for example, increase earthquake resistance or change the mounting structure. In addition, the designer 130 can have performance improvement measures presented every time a design change is made by simply pressing the performance improvement measure presentation button 4. This allows the designer 130 to confirm the performance improvement measures and modify the design, making it easier to modify the design.

[0100] In the above-described embodiment, the performance improvement measure presentation system 100 presents the performance improvement measures of the product based on the natural frequency for evaluating the strength (performance) of the product. Here, the strength of the product may be, for example, the earthquake resistance of the product. Alternatively, the performance improvement measure presentation system 100 may present the performance improvement measures of the product based on the weight for evaluating the weight reduction (performance) and the density of mounted components for evaluating the mountable space (performance). If the designer 130 finds that the performance of the control panel after the design change is excessive, the weight of the control panel can be reduced by, for example, making a design change to remove unnecessary members such as the frame of the control panel. In addition, by evaluating the mountable space, the designer 130 can determine that no more components can be mounted in the mountable space of the control panel, or that smaller components can be placed, and can make a new design change.

[0101] In the above-described embodiment, the target object is a control panel, and the control panel is the target of structural design, but the performance improvement measure presentation system 100 may be used for a structure similar to a control panel as the target of structural design and performance evaluation. For example, the performance improvement measure presentation system 100 may be used for a switchboard, distribution board, server, PC, explosive detection device, or medical device as the target of structural design. In addition, the performance improvement measure presentation system 100 may be used for a civil engineering building (e.g., a house, a bridge, or a plant facility) as the target object, when carrying out earthquake-resistant design of the civil engineering building.

[0102] The present invention is not limited to the above-described embodiment, and it goes without saying that various other applications and modifications are possible without departing from the gist of the present invention as set forth in the claims. For example, the above-mentioned embodiment describes the system configuration in detail and specifically in order to explain the present invention in an easily understandable manner, and is not necessarily limited to a system having all of the described configurations. In addition, it is also possible to add, delete, or replace part of the configuration of the present embodiment with other configurations. In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0103] 1...design change information, 2...performance prediction button, 3...prediction result output section, 4...performance improvement measure presentation button, 5...performance improvement measure output section, 6, 7...design completion, 11...design and analysis results DB, 21...performance prediction program, 22...interpretability calculation program, 23...performance improvement measure calculation program, 100...performance improvement measure presentation system, 110...learning data section, 120...program section, 130...designer, 140...designer PC, 11...design and analysis results DB, 12...performance improvement measure learning information DB

Claims

1. A learning data section holds, as learning data for machine learning, design parameters including performance values ​​of structural designs performed in the past on an object that is the subject of structural design, and analysis values ​​that are indexes of the structural performance of the object that has been subjected to structural analysis based on the design parameters; a calculation unit that predicts an index of structural performance of the object from design parameters input due to a design change using a prediction model obtained by machine learning of the design parameters and the analysis values ​​read from the learning data unit, calculates an interpretability that indicates an explainability of a result that contributes to the index of structural performance predicted by the prediction model, and presents a performance improvement measure for improving the performance of the object obtained from the interpretability together with the index of structural performance predicted by the prediction model. A system that suggests performance improvement measures.

2. The learning data section includes: a design and analysis results database that stores the design parameters and the analysis values, which are results of structural designs performed on objects in the past; a performance improvement measure learning information database that stores the design parameters and the analysis values ​​before a design change extracted for each of the objects from the design and analysis results database, and performance improvement measures for the objects after a design change. The performance improvement proposal system according to claim 1 .

3. The calculation unit is a performance prediction unit that performs machine learning based on the design parameters and the analysis values ​​before the design change read from the design and analysis results database to create a first prediction model, and predicts an index of structural performance using the first prediction model; a prediction result output unit that outputs the structural performance index predicted by the performance prediction unit as a prediction result; The performance improvement measure presentation system according to claim 2 .

4. The calculation unit is and an interpretability calculation unit that performs machine learning based on the design parameters and the analysis values ​​before the design change read from the design and analysis results database to create a second prediction model, and calculates the importance of the feature amount of the first prediction model, the explainability of the first prediction model that influences the prediction result, and the interpretability using the second prediction model. The performance improvement measure presentation system according to claim 3 .

5. The calculation unit is a performance improvement measure calculation unit that performs machine learning based on the design parameters before the design change, the design parameters after the design change, and the performance improvement measures read from the performance improvement measure learning information database to create a third prediction model, and calculates the performance improvement measure using the third prediction model based on the design parameters before the design change, the design parameters after the design change, and the deviation of the structural performance index predicted by the first prediction model after the design change from the analysis value before the design change; a performance improvement measure output unit that outputs the interpretability and the performance improvement measure. The performance improvement measure presentation system according to claim 3 .

6. The object is a control panel or a structure similar to the control panel. The performance improvement proposal system according to claim 5 .

7. The index of structural performance for which the performance prediction unit predicts the structural performance is any one of a natural frequency of the object, a weight of the object, and a density of components mounted on the object. The performance improvement proposal system according to claim 5 .

8. A step of obtaining a prediction model by machine learning design parameters and analytical values ​​that are indices of structural performance of an object that is the subject of structural design and that are obtained by structural analysis based on the design parameters, the design parameters and the analytical values ​​read from a learning data section that has the design parameters and the analytical values ​​as learning data for machine learning; predicting an index of structural performance of the object from design parameters input due to a design change using the prediction model; The method includes: calculating an interpretability that indicates the explainability of the results that contribute to the structural performance index predicted by the prediction model; and presenting a performance improvement measure for improving the performance of the object, obtained from the interpretability, together with the structural performance index predicted by the prediction model. How to present performance improvement measures.

Citation Information

Patent Citations

  • Airplane sensibility weighing optimization method based on support vector machine

    CN108509722A

  • Robust design method of angular contact ball bearing

    CN108830005A

  • Design supporting system using neural net

    JP1992367965A

  • Analysis device, analysis method, and program

    JP2016006587A

  • Design support device and design support method

    JP2020095377A