Method and device for generating an estimation model of product performance, device and method for estimating product performance, and program

Abstract

To provide a generating method of a product performance estimating model which can improve the estimation precision for a performance evaluation on a product.SOLUTION: Plural pieces of individual product data each containing specification data on a product and a performance evaluation value of the product are obtained. The individual product data which has, as the performance evaluation value, a value different from a pass reference value that is a reference value for pass or fail determination on the performance evaluation value is extracted from the plural pieces of individual product data as regression teacher data. A regression model for estimating the performance evaluation value of an estimation object product is generated using the regression teacher data.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present disclosure relates to a method and device for generating an estimation model of product performance, a device and method for estimating product performance, and a program. [Background technology]

[0002] In the design and development of tires, tire specifications such as tire dimensions, types of components, and materials are designed, and tire performance (e.g., durability, strength, etc.) obtained from the designed specifications is predicted. Whether the predicted performance satisfies required levels is then evaluated. Patent Document 1 listed below discloses the use of a machine learning model to estimate tire performance (physical values) from multiple rubber materials, fillers, etc. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-149423 Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, some tire performance tests end measurement of the performance evaluation value when the performance evaluation value reaches a pass criterion. An example of such a test is a durability test. In one example of a durability test, a load is applied to a tire, and the time until damage such as cracks or separation occurs in the tire is measured as a performance evaluation value representing durability. If damage is found in the tire (e.g., at 2500 minutes) before the measurement time reaches the pass criterion (e.g., 3000 minutes), 2500 minutes is recorded as the performance evaluation value of the tire. On the other hand, if no damage occurs until the pass criterion of 3000 minutes, the measurement ends when 3000 minutes has elapsed, and 3000 minutes is recorded as the performance evaluation value of the tire. The performance evaluation value measured in this manner is recorded in association with the tire specification data.

[0005] If a regression model for estimating tire performance evaluation values ​​is generated using such data, the estimated performance evaluation value will be influenced by the pass standard value, resulting in insufficient estimation accuracy. For example, even if a tire specification should output an estimated performance evaluation value lower than the pass standard of 3000 minutes, it is likely to output a performance evaluation value that meets the pass standard of 3000 minutes. This type of issue can also occur in the design of products other than tires. [Means for solving the problem]

[0006] (1) The method for generating an estimation model of product performance proposed in the present disclosure includes a product data acquisition step of acquiring a plurality of individual product data, each of which includes specification data representing at least one of the product's structure, materials, and manufacturing method, and a product performance evaluation value; a regression training data acquisition step of extracting, from the plurality of individual product data, individual product data having a performance evaluation value that is different from an acceptance reference value that is a reference value for determining whether the performance evaluation value is pass or fail, as regression training data; and a regression model generation step of generating a regression model for estimating the performance evaluation value of a product to be estimated using the regression training data. This method can generate a regression model using only individual product data having a performance evaluation value that is different from the acceptance reference value, thereby achieving a regression model with high estimation accuracy.

[0007] (2) The generation method of (1) may further include a judgment training data acquisition step of acquiring, from the plurality of individual product data, judgment training data including the specification data and a judgment result indicating whether the performance evaluation value achieves the pass standard value, and a judgment model generation step of generating, using the judgment training data, a judgment model for estimating the judgment result indicating whether the performance evaluation value of the estimation target product achieves the pass standard value. By using this judgment model, it becomes possible to estimate the performance evaluation value using a regression model for products whose performance evaluation value is estimated not to achieve the pass standard value when estimating product performance.

[0008] (3) In the step of acquiring training data for regression in (1), individual product data having a performance evaluation value that does not reach the pass standard value may be extracted as the training data for regression.

[0009] (4) The method for generating an estimation model in (1) may further include a consistency determination step for determining whether the judgment result estimated by the judgment model is consistent with the performance evaluation value estimated by the regression model, thereby further improving the accuracy of product performance estimation.

[0010] (5) In the process of acquiring the training data for judgment in (1), it may be determined whether the performance evaluation value contained in the individual product data satisfies the pass standard value, and the result may be generated as the judgment result data, and the judgment result data and the specification data may be used as the training data for judgment.

[0011] (6) In the method for generating an estimation model according to (1), the pass criteria value is determined according to the specification data. In the step of acquiring training data for judgment, it may be determined whether the performance evaluation value satisfies the pass criteria value corresponding to the specification data, and the result may be generated as the judgment result data.

[0012] (7) The device for generating an estimation model of product performance proposed in the present disclosure includes: a product data acquisition means for acquiring a plurality of individual product data, each of which includes specification data representing at least one of the product structure, material, and manufacturing method, and a performance evaluation value of the product; a regression training data acquisition means for extracting, from the plurality of individual product data, individual product data having a performance evaluation value that is different from an acceptance reference value that is a reference value for determining whether the performance evaluation value is pass or fail, as regression training data; and a regression model generation means for generating a regression model for estimating the performance evaluation value of a product to be estimated using the regression training data.

[0013] (8) The program proposed in this disclosure causes a computer to function as: a product data acquisition means for acquiring a plurality of individual product data, each of which includes specification data representing at least one of the product structure, material, and manufacturing method, and a performance evaluation value of the product; a regression training data acquisition means for extracting, from the plurality of individual product data, individual product data having a performance evaluation value that is different from an acceptance reference value that is a reference value for determining whether the performance evaluation value is pass or fail, as regression training data; and a regression model generation means for generating a regression model for estimating the performance evaluation value of a product to be estimated using the regression training data.

[0014] (9) A product performance estimation device proposed in this disclosure includes a storage means storing a judgment model that estimates a judgment result of whether a performance evaluation value of a product to be estimated will achieve a pass standard value based on specification data representing at least one of the structure, materials, and manufacturing method of the product to be estimated, and a regression model that estimates the performance evaluation value of the product to be estimated based on the specification data of the product to be estimated. The estimation device also includes a data acquisition means that acquires the specification data of the product to be estimated, a first estimation means that inputs the specification data of the product to the judgment model and estimates whether the performance evaluation value of the product to be estimated will achieve the pass standard value, and a second estimation means that inputs the specification data of the product to the regression model and estimates the performance evaluation value of the product to be estimated if the performance evaluation value of the product to be estimated does not achieve the pass standard value. In this estimation device, a performance evaluation value is estimated using the regression model for specification data for which a judgment result of the performance evaluation value not achieving the pass standard value is estimated. Therefore, in the process of generating the regression model, only product data having a performance evaluation value that does not reach the pass standard value is allowed to be used as training data. In other words, even if only such product data is used in the process of generating the regression model, a high degree of estimation accuracy can be ensured for the performance evaluation value.

[0015] (10) The product performance estimation device of (9) may further include a consistency determination means for determining whether the judgment result estimated by the judgment model and the performance evaluation value estimated by the regression model are consistent with each other, thereby further improving the accuracy of product performance estimation.

[0016] (11) In the product performance estimation device of (9), the regression model may be a model generated using individual product data having a performance evaluation value that is different from the pass standard value as training data.

[0017] (12) The method for estimating product performance proposed in the present disclosure includes a first estimation step of inputting specification data representing at least one of the structure, material, and manufacturing method of the estimated product into a judgment model and estimating a judgment result as to whether or not the performance evaluation value of the estimated product achieves the pass standard value, and a second estimation step of inputting the specification data of the estimated product into a regression model and estimating the performance evaluation value of the estimated product if the performance evaluation value of the estimated product does not achieve the pass standard value.

[0018] (13) The program proposed in the present disclosure causes a computer to function as a first estimation means that inputs specification data representing at least one of the structure, material, and manufacturing method of the estimated product into a judgment model and estimates a judgment result as to whether or not the performance evaluation value of the estimated product will achieve the pass standard value, and as a second estimation means that inputs the specification data of the estimated product into a regression model and estimates the performance evaluation value of the estimated product if the performance evaluation value of the estimated product does not achieve the pass standard value. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a block diagram showing hardware of a tire design support system that functions as a tire performance estimation device and an estimation model generation device proposed in the present disclosure. FIG. [Figure 2] FIG. 2 is a block diagram showing functions of the design support system. [Figure 3]FIG. 4 is a diagram showing an example of individual tire data stored in a storage device. [Figure 4] FIG. 3 is a diagram showing an example of individual tire data (teaching data for determination) used to generate a determination model. [Figure 5A] FIG. 10 is a diagram showing another example of individual tire data stored in the storage device. [Figure 5B] FIG. 5B is a diagram showing an example of data obtained by processing the individual tire data of FIG. 5A. [Figure 6] FIG. 10 is a flowchart illustrating an example of processing executed by a learning unit. [Figure 7] FIG. 10 is a flowchart illustrating an example of processing executed by an estimation unit. [Figure 8A] FIG. 10 is a diagram showing yet another example of individual tire data stored in the storage device. [Figure 8B] 8B is a diagram showing an example of data (teaching data for determination) used to generate a determination model, obtained from the individual tire data shown in FIG. 8A. FIG. [Figure 9] FIG. 10 is a flowchart illustrating an example of processing executed by a preprocessing unit that configures a learning unit. [Figure 10A] FIG. 10 is a diagram showing yet another example of individual tire data stored in the storage device. [Figure 10B] 10B is a diagram showing an example of data (teaching data for determination) used to generate a determination model, obtained from the individual tire data shown in FIG. 10A. FIG. DETAILED DESCRIPTION OF THE INVENTION

[0020] The following describes a method and device for generating an estimation model of product performance, and a device and method for estimating product performance proposed in the present disclosure. In the present disclosure, a tire is mainly used as an example of a product. That is, the following mainly describes a method and device for generating an estimation model of tire performance, and a device and method for estimating tire performance.

[0021] FIG. 1 is a block diagram showing the hardware of a tire design support system 1 that functions as a tire performance estimation device and an estimation model generation device proposed in this disclosure.

[0022] As shown in FIG. 1, the design support system 1 includes a processing device 11, a storage device 12, a display device 13, and an input device .

[0023] The processing device 11 includes, for example, a central processing unit (CPU) and a graphics processing unit (GPU), and operates according to a program stored in the storage device 12. A field programmable gate array (FPGA) may be used as the processing device 11. The storage device 12 includes, for example, a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), a solid state drive (SSD), and stores programs executed by the processing device 11, tire data for generating an estimation model, and the like. One or more personal computers, server computers, or the like may be used as the processing device 11 and the storage device 12.

[0024] The display device 13 is, for example, a liquid crystal display or an organic EL display, and displays the data stored in the storage device 12, the output results of the estimation model, and the like.

[0025] The input device 14 is a user interface such as a keyboard or a mouse, and receives an operation input from the operator and inputs a signal indicating the content of the operation input to the processing device 11.

[0026] Fig. 2 is a block diagram showing the functions of the design support system 1. As shown in Fig. 2, the design support system 1 has a learning unit 20, an estimation unit 30, and an estimation model group 40. The learning unit 20 and the estimation unit 30 are realized by the processing unit 11 executing a program stored in the storage device 12.

[0027] The estimation model group 40 includes a plurality of estimation models for estimating the performance of a tire (a tire whose performance is unknown) from its specifications. Specifically, the estimation model group 40 includes a determination model 41 that estimates whether the tire's performance evaluation value satisfies an acceptance standard value, and a regression model 42 that estimates the tire's performance evaluation value. The performance evaluation value is a numerical value that represents tire performance, such as a numerical value that represents the tire's durability (e.g., distance or time) or a numerical value that represents the tire's strength. The acceptance standard value is a standard value for determining whether the performance evaluation value passes or fails. If the performance evaluation value reaches the acceptance standard value, for example, if the performance evaluation value is equal to or greater than the acceptance standard value, the tire is deemed to pass. Alternatively, depending on the performance evaluation item, the performance evaluation value may be deemed to pass if it is equal to or less than the acceptance standard value. The tire specifications include, for example, the tire's dimensions, the types of components, materials, their properties (e.g., hardness), manufacturing method, etc.

[0028] The learning unit 20 causes a determination model 41 and a regression model 42 constituting an estimation model group 40 to learn a plurality of tire data stored in the storage device 12 .

[0029] The estimation unit 30 inputs specification data of a tire (estimation target tire) whose performance is to be estimated into a determination model 41, and estimates whether the performance evaluation value of the estimation target tire satisfies the pass standard value. The estimation unit 30 also inputs the specification data of the estimation target tire into a regression model 42, and estimates the performance evaluation value of this tire.

[0030] In the following, the learning unit 20 will be described first, and then the estimation unit 30 will be described.

[0031] [Tire data] FIG. 3 is a diagram showing an example of individual tire data used to generate estimation models 41 and 42. In this diagram, each row represents one individual tire data. The individual tire data is data about each manufactured tire. Each individual tire data includes specification data and a performance evaluation value. Each individual tire data may also include measurement conditions for the tire's performance evaluation value.

[0032] 3, the specification data includes items such as tire width, tread width, tire outer diameter, tire cross-sectional height (section height), etc. The specification data items are not limited to these, and may include, for example, rim width, rim diameter, tread type, number of belts, belt hardness, rubber material, manufacturing method (manufacturing conditions such as temperature), etc.

[0033] In the example of Fig. 3, each individual tire data has a durability time as a performance evaluation value. The durability time is, for example, the time until damage such as cracks or separation occurs in the tire under load. Measurement of the time until such damage occurs is terminated, for example, when an acceptance reference time, which is a pass reference value, is reached.

[0034] In the example shown in FIG. 3, the pass criteria time for tires ID: 1 to 3 is 3000. For tires "ID: 1" and "ID: 2," no damage such as cracks was observed when the measurement time reached 3000, so 3000 is recorded as the endurance time in the individual tire data. On the other hand, for tire "ID: 3," damage occurred when the measurement time reached 2520, so 2520 is recorded as the endurance time.

[0035] Such a pass reference time (pass reference value) does not have to be the same for all tires. For example, the pass reference time may differ depending on the type or size of the tire. In the example shown in FIG. 3, the pass reference time for tires ID: 11 to 13 is 4500. In the test for tires ID: 11 and ID: 12, as with tires ID: 1 and ID: 2, no damage such as cracks was observed when the measurement time reached 4500, so the measurement ended at that point and 4500 was recorded as the endurance time (performance evaluation value). On the other hand, for tire ID: 13, damage occurred when the measurement time reached 3850, so 3850 was recorded as the endurance time.

[0036] Each individual tire data may include a measurement condition of the performance evaluation value (endurance time in the example of FIG. 3). As shown in FIG. 3, the measurement condition may be, for example, a load acting on the tire during measurement.

[0037] In the design and development of tires, the inventors of the present application are considering using machine learning models to estimate tire performance based on design specifications such as tire dimensions, component types, materials, and manufacturing conditions. In the example of Figure 3, the inventors are considering estimating endurance time (performance evaluation value) based on specification data such as tire width and tread width. However, as illustrated in Figure 3, among tire performance evaluation items, measurement ends when it is determined that the tire reaches a pass standard value, and the pass standard value is recorded as the performance evaluation value in the individual tire data. Therefore, if a regression model for estimating performance evaluation values ​​is generated using all of the multiple individual tire data stored in the storage device 12, the estimated value obtained from the regression model will be influenced by the pass standard value, resulting in insufficient estimation accuracy. In the example of Figure 3, the estimated endurance time will be influenced by the pass standard time of 3000 or 4500.

[0038] Therefore, the learning unit 20 extracts individual tire data of tires having performance evaluation values ​​that are different from the pass standard value from the individual tire data shown in Fig. 3 as training data (training data for regression) for generating the regression model 42. The learning unit 20 also acquires or generates training data for judgment, including a judgment result as to whether or not the performance evaluation value achieves the pass standard value, and specification data, from the individual tire data shown in Fig. 3. The learning unit 20 generates the regression model 42 and the judgment model 41 by using these two types of training data, respectively.

[0039] [Study Department] As shown in FIG. 2, the learning unit 20 has, as its functions, a preprocessing unit 21, a judgment model generating unit 22, a regression model generating unit 23, and a consistency determining unit 24.

[0040] [Preprocessing section] When the original individual tire data (for example, the data illustrated in Figure 3) stored in the storage device 12 is not suitable as training data, the pre-processing unit 21 processes the original individual tire data or extracts a portion of the individual tire data to obtain training data (training data for judgment and training data for regression).

[0041] The judgment model 41 is a model that estimates whether or not the performance evaluation value of a tire will achieve the pass standard value based on the tire specification data. Therefore, the training data that the judgment model 41 learns requires a judgment result of whether or not the performance evaluation value achieves the pass standard value. If the original individual tire data does not include this judgment result, the pre-processing unit 21 generates individual tire data that includes this judgment result.

[0042] This processing by the pre-processing unit 21 will be described with reference to FIGS. 3 and 4. FIG. 4 is a diagram showing an example of data generated by the pre-processing unit 21 from the individual tire data of FIG. 3. As described above, in the example of FIG. 3, the tires "ID: 1" and "ID: 2" are not damaged until the pass reference time (pass reference value), and therefore the pass reference time (3000) is recorded as the endurance time (performance evaluation value) of these two tires. On the other hand, the endurance time of the tire "ID: 3" has not reached the pass reference time. Therefore, as shown in FIG. 4, the pre-processing unit 21 adds a determination result (specifically, pass / fail) of whether the endurance time has reached the pass reference time to each individual tire data, and uses this as training data for determination. More specifically, the pre-processing unit 21 compares the endurance time with the pass reference time (3000) for each individual tire data, and if the endurance time is equal to or greater than the pass reference time, adds "pass" as the determination result. If the endurance time is less than the pass reference time, adds "fail" as the determination result.

[0043] As described above, the pass reference time (pass reference value) does not have to be uniform for all tires. For example, the pass reference time may differ depending on the tire specifications (dimensions, type of components, material, manufacturing method, etc.). In the example of FIG. 3, the pass reference time is 3000 for tires (IDs: 1 to 3) with a tire section height of less than 120, while the pass reference time is 4500 for tires (IDs: 11 to 13) with a tire section height of 120 or more. In this way, when the pass reference time differs depending on the tire specifications, the pre-processing unit 21 compares the pass reference time (pass reference value) corresponding to the specification data with the tire's endurance time (performance evaluation value) and adds the judgment result (pass / fail) to the individual tire data.

[0044] To perform such processing, a table associating specification data with a pass reference time (pass reference value) may be stored in advance in the storage device 12. The pre-processing unit 21 may refer to the table and read out a reference time (3000 or 4500) corresponding to the specification data of each individual tire data (for example, the tire cross-section height). Then, the pre-processing unit 21 may compare the pass reference time (pass reference value) with the endurance time (performance evaluation value).

[0045] In this table, the number of specification data items to which the pass reference time (pass reference value) is associated is not limited to one, and may be two or more. In the example of Fig. 3, the pass reference time may be associated with the tire section height and tire width. In this case, the pre-processing unit 21 may refer to the table, read out the pass reference time corresponding to the tire section height and tire width included in the individual tire data, and compare the pass reference time with the endurance time of each tire.

[0046] The pass / fail determination result may be included in the individual tire data in advance. For example, the pass / fail determination result illustrated in Fig. 4 may be included in the original individual tire data illustrated in Fig. 3. In this case, the process of the pre-processing unit 21 for adding the pass / fail determination result may be omitted.

[0047] FIG. 5A is a diagram showing another example of individual tire data. As shown in this figure, the individual tire data may include, as performance evaluation values, the number of driving steps (number of tests) and the driving time (or driving distance) of the final step. Such tests may be performed, for example, multiple times at predetermined time intervals or multiple times at predetermined driving distance intervals. In the example of FIG. 5A, steps (driving) are performed multiple times at predetermined time intervals (e.g., 10-minute intervals), and the number of times is recorded as one item of the performance evaluation value. The individual tire data may also include the driving time (or driving distance) of the final step. FIG. 5A shows that the test for tire "ID:1" was performed from the start of the final driving step until 9 minutes, and the test for tires "ID:2" and "ID:3" was performed from the start of the final driving step until 8 minutes and 7 minutes, respectively.

[0048] When multiple numerical values ​​with different units are recorded as performance evaluation values ​​like this, it is difficult to generate a regression model 42 by directly using these numerical values. Therefore, the preprocessing unit 21 may calculate a performance evaluation value (continuous value) that can be used to generate the regression model 42 based on the number of driving steps (number of tests) and the driving time (or driving distance) in the final step. FIG. 5B shows an example of individual tire data generated by this processing of the preprocessing unit 21. In this figure, the preprocessing unit 21 calculates the total driving time (specifically, the number of steps × 10 + the driving time in the final step) based on the number of steps and the driving time in the final step.

[0049] [Decision model generation part] The judgment model generation unit 22 acquires individual tire data (training data for judgment) having a pass / fail judgment result (pass / fail) indicating whether or not the performance evaluation value achieves the pass standard value (training data for judgment acquisition step). When the pre-processing unit 21 adds the judgment result to the original individual tire data, the judgment model generation unit 22 acquires the individual tire data generated by the pre-processing unit 21. When the judgment result is included in the original individual tire data, the judgment model generation unit 22 may acquire the original individual tire data.

[0050] The judgment model generation unit 22 uses the acquired individual tire data as training data to generate a judgment model 41. At this time, the judgment result (pass / fail) is used as a correct label. Furthermore, the specification data is used as an explanatory variable. As the judgment model 41, for example, a machine learning model such as a decision tree, an SVM (support vector machine), or a neural network may be used.

[0051] [Regression model generation part] The regression model generation unit 23 extracts, from the original individual tire data, a plurality of individual tire data (training data for regression) having a value different from the pass standard value as a performance evaluation value (training data for regression acquisition step). For example, the regression model generation unit 23 extracts, as training data for regression, individual tire data having a value that does not reach the pass standard value as a performance evaluation value. In the example of FIG. 3, the regression model generation unit 23 extracts the individual tire data of tires "ID: 3" and "ID: 13" that have endurance times that do not reach the pass standard time.

[0052] The regression model generation unit 23 uses the extracted individual tire data to generate a regression model 42. For example, a machine learning model such as a neural network or a random forest may be used as the regression model 42. In this case, the regression model generation unit 23 inputs the performance evaluation value as a response variable to the machine learning model, and inputs the specification data as an explanatory variable to the machine learning model.

[0053] As shown in FIG. 5A, when multiple numerical values ​​with different units are recorded as performance evaluation values, the regression model generation unit 23 may generate the regression model 42 using the individual tire data including the performance evaluation values ​​calculated by the pre-processing unit 21.

[0054] [Consistency judgment part] The consistency determination unit 24 determines whether the determination result estimated by the determination model 41 and the performance evaluation value estimated by the determination model 42 are consistent with each other. For example, the consistency determination unit 24 inputs specification data of individual tire data that has not been used to generate the determination model 41, among the plurality of individual tire data stored in the storage device 12 (or individual tire data to which a determination result has been added by the processing of the pre-processing unit 21), to the determination model 41. Then, a pass / fail determination result (pass / fail) is obtained as an output from the determination model 41. The consistency determination unit 24 also inputs the same specification data to the regression model 42, and obtains a performance evaluation value as an output from the regression model 42. Then, the consistency determination unit 24 compares the acquired performance evaluation value with a pass reference value and determines whether the comparison result is consistent with the pass / fail determination result that is the output of the determination model 41.

[0055] For example, if the judgment result obtained from the judgment model 41 is "fail" and the performance evaluation value obtained from the regression model 42 does not reach the pass standard value, the consistency judgment unit 24 judges that the estimates made by the two models 41 and 42 are consistent. On the other hand, if the judgment result obtained from the judgment model 41 is "fail" (or "pass") and the performance evaluation value obtained from the regression model 42 reaches (or does not reach) the pass standard value, the consistency judgment unit 24 judges that the estimates made by the two models 41 and 42 are not consistent.

[0056] The consistency determination unit 24 may display the result (consistency / inconsistency) on the display device 13. If the display device 13 displays that the estimation results of the two models 41 and 42 are inconsistent, the operator may set new hyperparameters that define the determination model 41 for the determination model 41. Similarly, the operator may set new hyperparameters that define the regression model 42 for the regression model 42. In other words, the learning unit 20 may accept new hyperparameters via the input device 14.

[0057] As another example, if the estimation results from the two models 41 and 42 are inconsistent, the above-mentioned model generation units 22 and 23 may regenerate the regression model 42 and / or the judgment model 41. For example, the model generation units 22 and 23 may change hyperparameters according to a predetermined algorithm. Then, the judgment model generation unit 22 may generate a new judgment model 41 by using the individual tire data generated by the preprocessing unit 21 as training data, as described above. Furthermore, the regression model generation unit 24 may generate the regression model 41 by using the individual tire data stored in the storage device 12.

[0058] [flow] FIG. 6 is a flowchart showing an example of the flow of processing executed by the learning unit 20.

[0059] The learning unit 20 acquires the individual tire data exemplified in Fig. 3 stored in the storage device 12 (S101). The learning unit 20 accepts designation of performance evaluation items to be estimated by the estimation models (regression model 42 and determination model 41) (S102). The operator may be able to designate the performance evaluation items via the input device 14.

[0060] It is determined whether the individual tire data acquired in S101 includes a pass / fail judgment result for the performance evaluation item specified by the operator (S103). If the individual tire data does not include a judgment result, the pre-processing unit 21 adds the pass / fail judgment result to each individual tire data (S104). Specifically, the pre-processing unit 21 compares the pass standard value associated with the performance evaluation item with the performance evaluation value in the individual tire data. If endurance time is specified as the performance evaluation value as shown in FIG. 3, the pre-processing unit 21 compares the endurance time of each individual tire data with the pass standard time (3000 or 4500). Then, the pre-processing unit 21 adds either "pass" or "fail" to each individual tire data as the pass / fail judgment result according to the comparison result.

[0061] Next, the judgment model generating unit 22 generates (learns) the judgment model 41 using the plurality of individual tire data including the pass / fail judgment results and the specification data as training data (S105).

[0062] The regression model generation unit 23 extracts a plurality of individual tire data (training data for regression) having a value different from the pass reference value as a performance evaluation value from the plurality of individual tire data acquired in S101 (S106). In the example of FIG. 3, the regression model generation unit 23 extracts individual tire data having a value that does not reach the pass reference value (pass reference time) as a performance evaluation value (endurance time). The regression model generation unit 23 generates a regression model 42 using the training data for regression extracted in S106 (S107).

[0063] The consistency determination unit 24 determines whether the determination result (pass / fail) estimated by the determination model 41 generated in S105 is consistent with the performance evaluation value estimated by the determination model 42 generated in S107 (S108). If the estimates by the two models 41 and 42 are inconsistent, the learning unit 20 changes the hyperparameters of at least one of the determination model 41 and the regression model 42 (S109). In S109, the learning unit 20 may change the hyperparameters in accordance with instructions from an operator. That is, the operator may input changes to the hyperparameters to the learning unit 20 via the input device 14. Then, the processing of the learning unit 20 returns to S105, and the determination model generation unit 22 and the regression model generation unit 23 regenerate the determination model 41 and the regression model 42, respectively.

[0064] [Estimation part] The estimation unit 30 inputs the specification data of the estimation target tire into the trained determination model 41, and estimates whether or not the performance evaluation value of the estimation target tire will achieve the pass standard value. Furthermore, when the determination model 41 outputs a determination result that the performance evaluation value of the estimation target tire will not achieve the pass standard value, the estimation unit 30 inputs the specification data of the estimation target tire into the regression model 42, and estimates the performance evaluation value of the estimation target tire.

[0065] As described above, the regression model 42 is generated using, as training data, tire data having a performance evaluation value different from the pass standard value. More specifically, the regression model 42 is generated using, as training data, tire data having a performance evaluation value that does not reach the pass standard value. Meanwhile, the estimation unit 30 estimates a performance evaluation value using the regression model 42 for specification data for which a determination result indicates that the performance evaluation value does not achieve the pass standard value. In other words, both the training of the regression model 42 and the estimation by the regression model 42 use specification data that indicates a performance evaluation that does not reach the pass standard. Therefore, the estimation unit 30 can ensure high estimation accuracy of the performance evaluation value.

[0066] FIG. 7 is a flowchart showing an example of the flow of processing executed by the estimation unit 30.

[0067] The estimation unit 30 acquires specification data of the estimation target tire (S201). The specification data of the estimation target tire is data of the tire specifications designed by an operator, and is stored in, for example, the storage device 12. The estimation unit 30 may acquire this specification data via a network.

[0068] The estimation unit 30 inputs the specification data into the trained judgment model 41, and estimates, as its output, a pass / fail judgment result indicating whether or not a performance evaluation value (e.g., endurance time) achieves a pass standard value (e.g., pass standard time) (S202). The estimation unit 30 judges whether or not the estimated judgment result is "pass" (S203). If the estimated judgment result is "pass," the estimation unit 30 displays the result on the display device 13 (S204).

[0069] On the other hand, if the estimated judgment result is "fail", the estimation unit 30 inputs the specification data into the trained regression model 42 and estimates a performance evaluation value (for example, endurance time) as its output (S205).

[0070] Then, the estimation unit 30 determines whether the judgment result (fail) estimated in S202 is consistent with the performance evaluation value estimated in S205 (S206). Specifically, the estimation unit 30 compares the performance evaluation value acquired in S205 with a pass reference value, and determines whether the comparison result is consistent with the judgment result (specifically, "fail") output from the judgment model 41. If the performance evaluation value obtained from the regression model 42 does not reach the pass reference value, the estimation unit 30 determines that the estimations made by the two models 41 and 42 are consistent. On the other hand, if the judgment result obtained from the judgment model 41 in S202 is "fail" but the performance evaluation value obtained from the regression model 42 reaches the pass reference value, the estimation unit 30 determines that the estimations made by the two models 41 and 42 are inconsistent. As explained with reference to FIG. 3, when multiple pass criteria values ​​are specified according to the specification data (for example, pass criteria times of 3000 and 4500), the estimation unit 30 compares the pass criteria values ​​according to the specification data acquired in S201 with the performance evaluation value estimated in S205.

[0071] Finally, the estimation unit 30 displays on the display device 13 the judgment result (fail) estimated in S202, the performance evaluation value estimated in S205, and the judgment result (match / mismatch) in S206.

[0072] [Another example of tire data] In the example described with reference to the individual tire data in FIGS. 3 and 4, the pre-processing unit 21 compares the performance evaluation value with the pass / fail criteria and adds a judgment result (pass / fail) based on the comparison result to the individual tire data. The processing of the pre-processing unit 21 is not limited to this. For example, the pre-processing unit 21 may add a pass / fail judgment result to the individual tire data based on the test results included in the individual tire data. An example of such a test result is a strength test using a stick. In this strength test, the tip of a stick is pressed against the tire, and the pressing force of the stick is gradually increased until it reaches the pass criteria, testing whether or not damage occurs to the tire.

[0073] In the individual tire data in Figure 8A, the values ​​("OK" and "NG") for the strength test (first), strength test (second), and strength test (third) are the results of such stick penetration tests. "OK" indicates that no damage was observed, and "NG" indicates that damage occurred.

[0074] When the individual tire data includes such test results, the pre-processing unit 21 may assign a pass / fail judgment result to the individual tire data based on the multiple test results. Fig. 8B shows an example of individual tire data to which a pass / fail judgment result has been added by processing by the pre-processing unit 21.

[0075] In the example of FIG. 8A, the tires with "ID:1" and "ID:11" did not experience any damage in any of the first to third strength tests. Therefore, the pre-processing unit 21 marks the data of these tires as "pass" as a pass / fail determination result, as shown in FIG. 8B. On the other hand, the tires with "ID:2" and "ID:12" experienced damage in the third test, so the pre-processing unit 21 marks the data of these tires as "fail," as shown in FIG. 8B.

[0076] The "measured penetration force" in the right column of the strength test (third time) is, for example, the force with which a stick is inserted into a tire. Penetration force is an example of a performance evaluation value. In the individual tire data in FIG. 8A, the measured penetration force of "100" is the pass standard value. For the data of "ID:1" and "ID:11," "100" is recorded as the "measured penetration force." This means that no damage occurred to the tire even when the stick penetration force reached the pass standard value of "100." On the other hand, for the data of "ID:2" and "ID:12," "80" is recorded as the "measured penetration force." This means that damage occurred to the tire in the second strength test when the stick penetration force was "80," before reaching the pass standard value of "100."

[0077] 8A and 8B, if there is even one "NG" among the multiple test results (the first to third test results), the pre-processing unit 21 marks the individual tire data as "fail." Alternatively, the pre-processing unit 21 may mark the individual tire data as "fail" if the ratio of the number of "OK" results to the total number of tests is lower than a predetermined value.

[0078] When the individual tire data exemplified in FIG. 8B is generated by the pre-processing unit 21, the model generating units 22 and 23 execute the following process, for example.

[0079] In this case, the processing executed by the determination model generation unit 22 may be the same as the processing of the determination model generation unit 22 that uses the individual tire data exemplified in Fig. 4. That is, the determination model generation unit 22 may generate the determination model 41 by using the individual tire data to which the determination result (pass / fail) has been added by the preprocessing unit 21 as training data.

[0080] The regression model generation unit 23 extracts, as training data for regression, individual tire data having a performance evaluation value that does not reach the pass / fail standard value. In the example of FIG. 8B , the regression model generation unit 23 extracts tire data (tires with ID: 2 and ID: 12) to which "fail" has been assigned as the pass / fail judgment result from the plurality of individual tire data illustrated in FIG. 8B . Then, the regression model generation unit 23 uses the individual tire data extracted in this manner to generate a regression model 42 for estimating a performance evaluation value (the penetration force at which damage occurs). At this time, the regression model generation unit 23 inputs the performance evaluation value (measured penetration force) into a machine learning model as a dependent variable and inputs the specification data into the machine learning model as an explanatory variable. For example, a neural network or a random forest may be used as the machine learning model.

[0081] [Modification of pre-processing unit] There may be multiple types of performance evaluation items, such as the following three types, as described with reference to Fig. 3 and Fig. 8A. (First type) This is an evaluation item for which the same pass criteria are applied to all tire data. (Second type) This is an item to which multiple pass criteria values ​​are applied according to the tire specification data. In the example shown in Figures 3 and 4, two pass criteria values ​​(pass criteria time) for endurance time are specified: "3000" and "4500." These multiple pass criteria values ​​are specified according to the tire section height, which is one of the specification data. (Third type) This is an item that adds a pass / fail judgment result to individual tire data based on multiple factors such as multiple test results, multiple specification data, etc. In the example shown in Figures 8A and 8B, the pass / fail judgment result is added based on the values ​​("OK" or "NG") of the strength test (first time), strength test (second time), and strength test (third time).

[0082] The pre-processing unit 21 may automatically add a judgment result (pass / fail). Fig. 9 is a flow diagram showing an example of such processing by the pre-processing unit 21. The storage device 12 may store a map that associates performance evaluation items with numbers (1 to 3) that indicate the types of those items.

[0083] 9, the preprocessing unit 21 receives a specification of a performance evaluation item for which an estimation model (a determination model 41 and a regression model 42) is to be generated (S301). For example, the operator can specify, via the input device 14, "endurance time" or "stick insertion force" as the performance evaluation item to be estimated.

[0084] Next, the pre-processing unit 21 determines whether the specified performance evaluation item corresponds to a first type (item for which the same pass standard value can be applied to all data) (S302). If the specified performance evaluation item corresponds to the first type, the pre-processing unit 21 compares the performance evaluation value of each individual tire data with the pass standard value and adds the comparison result to each individual tire data as a pass / fail judgment result (S303).

[0085] If the determination in S302 indicates that the specified performance evaluation item does not fall under the first category, the pre-processing unit 21 determines whether the performance evaluation item falls under the second category (S304). That is, the pre-processing unit 21 determines whether the pass standard value for the performance evaluation item has a plurality of different pass standard values ​​(3000 and 4500 in the example of FIG. 3) depending on the specification data. If the performance evaluation item falls under the second category, the pre-processing unit 21 compares the pass standard value according to the specification data with the performance evaluation value and adds the comparison result to each individual tire data as a pass / fail determination result (S305). In the example of FIG. 3, for tires with a tire section height of less than 120 (tires with IDs 1 to 3), the pass standard value: 3000 is compared with the performance evaluation value, and for tires with a tire section height of 120 or more (tires with IDs 11 to 13), the pass standard value: 4500 is compared with the performance evaluation value.

[0086] If the determination in S306 determines that the specified performance evaluation item does not fall under the second type, the preprocessing unit 21 determines whether the performance evaluation item falls under the third type (S306). That is, as described with reference to FIGS. 8A and 8B, the preprocessing unit 21 determines whether the specified performance evaluation item is an item for which pass / fail is determined based on multiple test results (or multiple specification data). If the determination in S306 determines that the performance evaluation item falls under the third type, the preprocessing unit 21 adds a pass / fail determination result to each individual tire data based on the multiple test results (S307). For example, as described above, if one test result is "NG," the preprocessing unit 21 adds "fail" to the data for that tire. If the specified performance evaluation item does not fall under the third type either, the preprocessing unit 21 may display that fact on the display device 13 and end the processing.

[0087] [Search for tire specifications] The design support system 1 may have, as its functions, a search unit that searches for tire specification data that satisfies the characteristics desired by the operator, in addition to the learning unit 20 and the estimation unit 30. This process of the search unit can be executed, for example, as follows.

[0088] The search unit first provides the estimation unit 30 with specification data that is defined in advance or specification data that is set by an operator. The estimation unit 30 inputs the specification data acquired from the search unit into an estimation model (the judgment model 41 and / or the regression model 42), calculates a pass / fail judgment result and / or a performance evaluation value as the output, and provides them to the search unit. The search unit may generate new specification data so that the calculated pass / fail judgment result and / or performance evaluation value approaches the target set by the operator, and provide this to the estimation unit 30. The design support system 1 searches for specification data that achieves performance close to the target set by the operator by repeatedly executing the processing of the search unit and the processing of the estimation unit 30.

[0089] The estimation unit 30 may estimate multiple performances based on the specification data acquired from the search unit. That is, the estimation unit 30 may estimate multiple performances, such as endurance time and penetration force. The estimation unit 30 may estimate these multiple performances based on the specification data acquired from the search unit and provide the estimated values ​​to the search unit. The search unit may then generate new specification data so that all of these multiple performances approach the targets set by the operator and provide this to the estimation unit 30. The design support system 1 may search for specification data that achieves values ​​close to the targets set by the operator for the multiple performances by repeatedly executing the processing of the search unit and the processing of the estimation unit 30. Such a search may be achieved by a so-called genetic algorithm or a gradient method.

[0090] [summary] In the above-described estimation model generation method, individual tire data having a performance evaluation value that is different from the pass / fail reference value, which is a reference value for determining whether a performance evaluation value passes or fails, is extracted as training data for regression from the plurality of individual tire data. More specifically, individual tire data having a performance evaluation value that does not reach the pass reference value is extracted as training data for regression from the plurality of individual tire data. A regression model for estimating the performance evaluation value of a tire to be estimated is then generated using this training data for regression. In this generation method, the regression model is generated using individual tire data having a performance evaluation value that is different from the pass reference value, thereby achieving a regression model with high estimation accuracy.

[0091] In the above-described method for generating an estimation model, training data for judgment, including a judgment result indicating whether or not the performance evaluation value meets the pass standard value and the specification data, is acquired from a plurality of individual tire data. A judgment model for estimating whether or not the performance evaluation value of a tire to be estimated meets the pass standard value is generated using the training data for judgment. By using this judgment model, it becomes possible to estimate the performance evaluation value using a regression model for tires whose performance evaluation value does not meet the pass standard value when estimating tire performance.

[0092] The estimation unit 30 also includes a means (S202) for inputting the specification data of the tire to be estimated into the judgment model 41 and estimating whether the performance evaluation value of the tire to be estimated will achieve the pass standard value. The estimation unit 30 also includes a means (S205) for inputting the specification data of the tire to be estimated into the regression model 42 and estimating the performance evaluation value of the tire to be estimated if the performance evaluation value of the tire to be estimated does not achieve the pass standard value. In this manner, the estimation unit 30 estimates the performance evaluation value using the regression model 42 for specification data for which a judgment result indicating that the performance evaluation value will not achieve the pass standard value has been estimated. Therefore, in the process of generating the regression model 42, only tire data having a performance evaluation value that does not reach the pass standard value is permitted to be used as training data. In other words, even if only such tire data is used in the process of generating the regression model 42, specification data estimated to have a performance evaluation value that will not achieve the pass standard value is input into the regression model 42 during estimation, ensuring high estimation accuracy of the performance evaluation value.

[0093] [Variations] It should be noted that the method and device for generating an estimation model of product performance, and the device and method for estimating product performance proposed in the present disclosure are not limited to the above-described examples.

[0094] For example, although tires have been used as examples of products in this disclosure, the methods and the like proposed in this disclosure may be applied to products other than tires.

[0095] For example, the method proposed in this disclosure may be applied to a key box for storing valuables. In this case, the individual product data includes specification data for the key box, such as the wall structure, wall material, and assembly method. The key box is subjected to, for example, a strength test. That is, the load applied to the key box is gradually increased up to a predetermined upper limit (e.g., 3 tons). If the key box does not break even when the load reaches the upper limit, an "OK" is assigned as the test result (see FIG. 8A), which is one item of the individual product data, and the upper limit of the load (e.g., 3 tons) is assigned as the performance evaluation value (see FIG. 8A). On the other hand, if the key box breaks before the load reaches the upper limit, an "NG" is assigned as the test result, and the load at which the breakage occurred is assigned as the performance evaluation value.

[0096] As in the example described with reference to FIGS. 8A and 8B , such a test may be performed multiple times. The preprocessing unit 21 assigns "pass" as a pass / fail judgment result to individual product data that has been assigned "OK" as a test result, and assigns "fail" as a pass / fail judgment result to individual product data that has been assigned "NG" as a test result. The discriminant model generation unit 22 generates a discriminant model using "pass" and "fail" as objective variables and specification data such as structure and materials as explanatory variables. The regression model generation unit 23 extracts the individual product data that has been assigned "fail" as a pass / fail judgment result as training data for regression. The regression model generation unit 23 then generates a regression model using the load, which is a performance evaluation value, as the objective variable and specification data such as structure and materials as explanatory variables.

[0097] As another example, the method proposed in this disclosure may be applied to a switching power supply circuit having multiple output terminals. In this case, the individual product data includes specification data of the power supply circuit, such as the material and length of the wires and the number of output terminals. The power supply circuit is subjected to, for example, a current endurance test. For example, a predetermined current (specified load current) is output from the multiple output terminals of the power supply circuit, and it is determined whether a power supply failure occurs over a predetermined pass reference time (pass reference value). If no power supply failure occurs over the pass reference time, the pass reference time is assigned as a performance evaluation value, which is one of the items of the individual product data. On the other hand, if a power supply failure is recognized before the pass reference time is reached, the time during which the power supply failure was recognized is assigned as the performance evaluation value.

[0098] The generation of a discriminant model and a regression model for evaluating the performance of a switching power supply circuit is performed, for example, as follows. The preprocessing unit 21 assigns "pass" as a pass / fail judgment result to individual product data that has been assigned a performance evaluation value that is equal to or greater than the pass reference time. The preprocessing unit 21 also assigns "fail" as a pass / fail judgment result to individual product data that has been assigned a performance evaluation value that is smaller than the pass reference time. The discriminant model generation unit 22 generates a discriminant model using "pass" and "fail" as objective variables and specification data such as the material and length of the electric wire as explanatory variables. The regression model generation unit 23 extracts the individual product data that has been assigned "fail" as a pass / fail judgment result as training data for regression. The regression model generation unit 23 then generates a regression model using the load, which is the performance evaluation value, as the objective variable and specification data such as the material and length of the electric wire as explanatory variables.

[0099] In the above description, the regression model 42 is generated using tire data having a performance evaluation value that does not reach the pass standard value. However, for items for which the measured performance evaluation value exceeds the pass standard value, the regression model may be generated using only tire data having a performance evaluation value that exceeds the pass standard value. Specification data whose performance evaluation value is estimated to reach the pass standard value may be input to the regression model generated in this manner, and the performance evaluation value for this specification data may be estimated.

[0100] In yet another example, the regression model 42 may be generated using tire data having a performance evaluation value that does not reach the pass standard value (a performance evaluation value lower than the pass standard value), and a regression model may be generated separately from this model 42 using only tire data having a performance evaluation value that exceeds the pass standard value.

[0101] An example of such an item is a strength test using a stick. This strength test has not only an upper limit for the stick insertion force, which is the pass criterion, but also an upper limit for the tire height (half the value obtained by subtracting the rim diameter from the tire outer diameter). When a stick insertion test is performed on a tire with a small tire height, the tip of the stick may come into contact with the rim before the penetration force reaches the upper limit (pass criterion). In this case, the point at which the tip of the stick touches the rim is the upper limit of the penetration force. In this strength test, if the stick touches the rim before the penetration force reaches the upper limit (pass criterion), the tire is determined to meet the pass criterion. In other words, a "pass" is assigned to the individual tire data as a pass / fail judgment result. The stick insertion force at the time the stick comes into contact with the rim is used as the performance evaluation value.

[0102] Fig. 10A is an example of individual tire data obtained by such a strength test, and Fig. 10B is a diagram showing an example of data to which pre-processing unit 21 has assigned a pass / fail judgment result by referring to the individual tire data of Fig. 10A.

[0103] In the individual tire data in Figure 10A, the strength test values ​​include "OK," "NG," and "M." As in the example shown in Figure 8A, "OK" indicates no damage occurred, and "NG" indicates damage occurred. In the data in Figure 10A, "M" indicates that the rod contacted the rim before the rod penetration force reached the upper limit (pass criterion value). The testing device may measure the rod position in addition to the rod penetration force. If the rod position does not change even when the rod penetration force is increased, it is determined that the tip of the rod is in contact with the rim, and an "M" is assigned as the test result. In Figure 10A, the "Measured Penetration Force" in the column to the right of the strength test is the force with which the rod penetrates the tire and is used as a performance evaluation value. If the tip of the rod contacts the rim, the penetration force at that time is recorded as the performance evaluation value.

[0104] In the individual tire data of FIG. 10A, the measured penetration force of "100" is the pass standard value. For the data of "ID: 12" in FIG. 10A, "M" is recorded as the test result, and "80" is recorded as the "measured penetration force." This means that the rod contacted the rim when the penetration force of the rod was "80" before reaching the pass standard value. As shown in FIG. 10B, the pre-processing unit 21 assigns "pass" as the pass / fail judgment result to the data of "ID: 12."

[0105] The pre-processing unit 21 may perform the same processing as in the examples of FIGS. 8A and 8B described above for the data of "ID:1," "ID:2," and "ID:11." That is, "OK" is recorded as the test result for the data of "ID:1" and "ID:2." As shown in FIG. 10B, the pre-processing unit 21 assigns "pass" to these data as the pass / fail determination result. Also, in FIG. 10A, "NG" is recorded as the test result for the data of "ID:11." As shown in FIG. 10B, the pre-processing unit 21 assigns "fail" to the data of "ID:11" as the pass / fail determination result.

[0106] When the individual tire data of FIG. 10B is stored in the storage device 12, the discriminant model generation unit 22 generates a discriminant model 41 using the pass / fail judgment results "pass" and "fail" as the objective variable and the specification data as the explanatory variable. The regression model generation unit 23 extracts data marked "fail" from the individual tire data and uses this data as training data for regression to generate a regression model 42. The regression model generation unit 23 also extracts data marked "M" as the test result (the pass judgment result is "pass"). Then, using this data as training data for regression, it generates another regression model. That is, the regression model generation unit 23 generates another regression model using the "measured penetration force" as the objective variable and the specification data as the explanatory variable. By using this regression model, the penetration force required for the tip of the rod to contact the rim can be accurately estimated as a performance evaluation value. [Explanation of symbols]

[0107] 11: Processing device, 12: Storage device, 13: Display device, 14: Input device, 20: Learning unit, 21: Preprocessing unit, 22: Decision model generation unit, 23 Regression model generation unit, 24 Consistency determination unit, 40 Estimation model group, 41 Decision model, 42 Regression model.

Claims

1. a product data acquisition step of acquiring a plurality of individual product data, each of which includes specification data indicating at least one of the structure, material, and manufacturing method of the product and a performance evaluation value of the product; a regression training data acquisition step of extracting, as training data for regression, individual product data having, as the performance evaluation value, a value different from an acceptance reference value which is a reference value for determining whether the performance evaluation value is acceptable or not, from the plurality of individual product data; a regression model generation step of generating a regression model for estimating the performance evaluation value of the estimation target product using the training data for regression; A method for generating an estimated model of product performance, including:

2. a judgment training data acquisition step of acquiring judgment training data including a judgment result indicating whether the performance evaluation value achieves the pass standard value and the specification data from the plurality of individual product data; and a judgment model generation step of generating a judgment model for estimating a judgment result as to whether or not the performance evaluation value of the estimation target product achieves the pass standard value, using the training data for judgment. A method for generating an estimation model of product performance according to claim 1.

3. In the step of acquiring training data for regression, individual product data having a value that does not reach the pass standard value as the performance evaluation value is extracted as the training data for regression. A method for generating an estimation model of product performance according to claim 1.

4. The method further includes a consistency determination step of determining whether or not the determination result estimated by the determination model matches the performance evaluation value estimated by the regression model. A method for generating an estimation model of product performance according to claim 2.

5. In the step of acquiring the training data for judgment, it is judged whether or not the performance evaluation value included in the individual product data satisfies the pass standard value, and the result is generated as the judgment result, and the judgment result and the specification data are used as the training data for judgment. A method for generating an estimation model of product performance according to claim 2.

6. the pass criteria are determined according to the specification data, In the judgment training data acquisition step, it is determined whether the performance evaluation value satisfies a pass standard value corresponding to the specification data, and the result is generated as the judgment result. A method for generating an estimation model of product performance according to claim 2.

7. a product data acquisition means for acquiring a plurality of individual product data, each of which includes specification data indicating at least one of the structure, material, and manufacturing method of the product and a performance evaluation value of the product; a training data acquisition means for extracting, as training data for regression, individual product data having, as the performance evaluation value, a value different from an acceptance reference value which is a reference value for determining whether the performance evaluation value is acceptable or not, from the plurality of individual product data; a regression model generating means for generating a regression model for estimating the performance evaluation value of the estimation target product by using the training data for regression; A device for generating an estimation model of product performance including:

8. a product data acquisition means for acquiring a plurality of individual product data, each of which includes specification data indicating at least one of the structure, material, and manufacturing method of the product and a performance evaluation value of the product; a training data acquisition means for extracting, as training data for regression, individual product data having, as the performance evaluation value, a value different from an acceptance reference value which is a reference value for determining whether the performance evaluation value is acceptable or not, from the plurality of individual product data; a regression model generating means for generating a regression model for estimating the performance evaluation value of the estimation target product by using the training data for regression; A program that makes a computer function as a

9. a storage means for storing a judgment model for estimating a judgment result as to whether or not a performance evaluation value of a product to be estimated will achieve a pass standard value based on specification data representing at least one of a structure, material, and manufacturing method of the product to be estimated, and a regression model for estimating the performance evaluation value of the product to be estimated based on the specification data of the product to be estimated; data acquisition means for acquiring specification data of the estimation target product; a first estimation means for inputting specification data of the estimation target product into the judgment model and estimating whether or not a performance evaluation value of the estimation target product will achieve the pass standard value; a second estimation means for inputting specification data of the estimated product into the regression model and estimating a performance evaluation value of the estimated product when the performance evaluation value of the estimated product does not achieve the pass standard value; A product performance estimation device including:

10. The system further includes a consistency determination means for determining whether or not the determination result estimated by the determination model matches the performance evaluation value estimated by the regression model. The product performance estimation device according to claim 9.

11. The regression model is a model generated by using, as training data, individual product data having a performance evaluation value that is different from the pass standard value. The product performance estimation device according to claim 9.

12. a first estimation step of inputting specification data representing at least one of the structure, material, and manufacturing method of a product to be estimated into a judgment model and estimating a judgment result as to whether or not a performance evaluation value of the product to be estimated achieves an acceptance standard value; a second estimation step of inputting the specification data of the estimated product into a regression model to estimate the performance evaluation value of the estimated product when the performance evaluation value of the estimated product does not achieve the acceptance standard value; A method for estimating product performance, including:

13. a first estimation means for inputting specification data representing at least one of the structure, material, and manufacturing method of the estimation target product into a judgment model and estimating a judgment result as to whether or not a performance evaluation value of the estimation target product achieves an acceptance standard value; a second estimation means for inputting the specification data of the estimated product into a regression model and estimating the performance evaluation value of the estimated product when the performance evaluation value of the estimated product does not achieve the acceptance standard value; A program that makes a computer function as a

Images