Spring fatigue life prediction device, spring fatigue life prediction method, and spring fatigue life prediction program

Abstract

A purpose of an embodiment of the present invention is to provide a spring fatigue life prediction system in which the prediction accuracy is improved. A spring fatigue life prediction device according to one embodiment of the present invention comprises: a database including at least the fatigue life of springs, and the peak residual stress of the springs and the arithmetic average roughness of the surface of the springs associated with the fatigue life of the springs; and a spring fatigue life prediction model generation unit that generates a spring fatigue life prediction model by machine learning using explanatory variables including at least the peak residual stress of a spring and the arithmetic average roughness of the surface of the spring, and the fatigue life of the spring as an objective variable. The spring fatigue life prediction model is used to predict the fatigue life of a spring on the basis of explanatory variables.

Description

Spring fatigue life prediction device, spring fatigue life prediction method, and spring fatigue life prediction program

[0001] This invention relates to a spring fatigue life prediction system; or, a spring fatigue life prediction method; or, a method for detecting irregularly shaped structures contained in an image; or, a spring fatigue life prediction program.

[0002] To manage the design and quality of springs, it is necessary to evaluate the fatigue life of springs. Traditionally, fatigue life was evaluated by conducting durability tests, but in recent years, techniques for predicting fatigue life using machine learning have been investigated. For example, Patent Document 1 discloses a technique for estimating fracture life that provides a highly accurate estimate by estimating an S-N curve using machine learning. This technique comprises a functional unit that reads the type of fatigue limit estimation and the mechanical properties to be used in the machine learning decision tree, a functional unit that reads fatigue data for a specified steel type from a fatigue data sheet, and a machine learning calculation unit that performs supervised machine learning on the read fatigue data for the steel type using the specified mechanical properties. The machine learning calculation unit that performs supervised machine learning outputs a calculation result for the fatigue limit estimation according to the specified type of fatigue limit estimation using the specified mechanical properties.

[0003] Furthermore, Patent Document 2 discloses a technology for estimating the relationship between fatigue life and load stress, which improves the machine learning method to obtain accurate estimations when estimating an S-N curve using machine learning. This technology comprises a functional unit that reads the type of fatigue limit estimation and the mechanical properties to be used in the machine learning decision tree, a functional unit that reads fatigue data of a specified metal structural material from a fatigue data sheet, and a machine learning calculation unit that performs supervised machine learning on the read fatigue data of the metal structural material using the specified mechanical properties to estimate the load stress corresponding to the fatigue life. The machine learning calculation unit that performs supervised machine learning is configured to output a calculation result of load stress estimation corresponding to the fatigue life according to the specified type of fatigue limit estimation, using the specified mechanical properties.

[0004] Japanese Patent Application Laid-Open No. 2023-116283 and Japanese Patent Application Laid-Open No. 2024-034589

[0005] One object of an embodiment of the present invention is to provide a spring durability fatigue life prediction device with improved prediction accuracy. Alternatively, one object of an embodiment of the present invention is to provide a spring durability fatigue life prediction method with improved prediction accuracy. Alternatively, one object of an embodiment of the present invention is to provide a spring durability fatigue life prediction program with improved prediction accuracy.

[0006] A spring durability fatigue life prediction device according to an embodiment of the present invention includes a database including at least the durability fatigue life of a spring, the peak residual stress of the spring associated with the durability fatigue life of the spring, and the arithmetic mean roughness of the surface of the spring, an explanatory variable including at least the peak residual stress of the spring and the arithmetic mean roughness of the surface of the spring, and a spring durability fatigue life prediction model generation unit that generates a spring durability fatigue life prediction model by machine learning using the spring durability fatigue life at a predetermined stress amplitude as an objective variable, and predicts the spring durability fatigue life based on the explanatory variable using the spring durability fatigue life prediction model.

[0007] The database may further include the stress amplitude applied to the spring during the durability test associated with the durability fatigue life of the spring, and the explanatory variable may further include the stress amplitude applied to the spring during the durability test.

[0008] The database may further include the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the durability test associated with the durability fatigue life of the spring, and the explanatory variable may further include the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the durability test.

[0009] The database consists of the durability fatigue life of the spring, the peak residual stress of the spring associated with the durability fatigue life of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude of the spring durability test, the Rockwell hardness, and the mounting height of the spring during the durability test, and the explanatory variable may consist of the peak residual stress of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test.

[0010] A method for predicting the fatigue life of a spring according to one embodiment of the present invention comprises: a computer reading data from a database that includes at least the fatigue life of the spring, the peak residual stress of the spring associated with the fatigue life of the spring, and the arithmetic mean roughness of the spring surface; a computer generating a fatigue life prediction model for the spring by machine learning using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness of the spring surface, and the fatigue life of the spring at a predetermined stress amplitude as the objective variable; and a computer predicting the fatigue life of the spring based on the explanatory variables using the fatigue life prediction model for the spring.

[0011] The computer may further include, from the database, the stress amplitude applied to the spring during endurance testing associated with the spring's endurance fatigue life, and the explanatory variables may further include the stress amplitude applied to the spring during endurance testing.

[0012] The computer reads data from the database, further including the maximum cross-sectional height of the spring, Rockwell stiffness, and the mounting height of the spring during the endurance test, which are associated with the endurance fatigue life of the spring. The explanatory variables may further include the maximum cross-sectional height of the spring, Rockwell stiffness, and the mounting height of the spring during the endurance test.

[0013] The computer reads data from a database consisting of the spring's endurance fatigue life, the spring's peak residual stress associated with the spring's endurance fatigue life, the spring's maximum cross-sectional height, the arithmetic mean roughness of the spring's surface, the stress amplitude applied to the spring during the endurance test, the Rockwell hardness, and the spring's mounting height during the endurance test. The explanatory variables may consist of the spring's peak residual stress, the spring's maximum cross-sectional height, the spring's arithmetic mean roughness of the spring's surface, the stress amplitude applied to the spring during the endurance test, the Rockwell hardness, and the spring's mounting height during the endurance test.

[0014] A spring fatigue life prediction program according to one embodiment of the present invention includes: loading data from a database into a computer that includes at least the fatigue life of a spring, the peak residual stress of the spring associated with the fatigue life of the spring, and the arithmetic mean roughness of the spring surface; generating a spring fatigue life prediction model for the computer by machine learning using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness of the spring surface, and the fatigue life of the spring at a predetermined stress amplitude as the objective variable; and using the spring fatigue life prediction model, causing the computer to predict the fatigue life of the spring based on the explanatory variables.

[0015] The computer may be fed data from the database, which further includes the stress amplitude applied to the spring during durability testing, associated with the spring's fatigue life, and the explanatory variables may further include the stress amplitude applied to the spring during durability testing.

[0016] The computer is fed data from the database, further including the maximum cross-sectional height of the spring, Rockwell hardness, and the mounting height of the spring during the endurance test, which are associated with the endurance fatigue life of the spring. The explanatory variables may further include the maximum cross-sectional height of the spring, Rockwell hardness, and the mounting height of the spring during the endurance test.

[0017] The database is used to input data into a computer, which includes the spring's fatigue life, the peak residual stress associated with the spring's fatigue life, the spring's maximum cross-sectional height, the arithmetic mean roughness of the spring's surface, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the spring's mounting height during the durability test. The explanatory variables may consist of the spring's peak residual stress, the spring's maximum cross-sectional height, the spring's arithmetic mean roughness of the spring's surface, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the spring's mounting height during the durability test.

[0018] One embodiment of the present invention provides a spring fatigue life prediction device with improved prediction accuracy. Alternatively, one embodiment of the present invention provides a spring fatigue life prediction method with improved prediction accuracy. Alternatively, one embodiment of the present invention provides a spring fatigue life prediction program with improved prediction accuracy.

[0019] This is a block diagram showing a spring fatigue life prediction device 100 according to one embodiment of the present invention. This is a flowchart showing a spring fatigue prediction method according to one embodiment of the present invention. This figure shows the results of generating a spring fatigue life prediction model and predicting the spring fatigue life. This is a table showing the results of generating a spring fatigue life prediction model and predicting the spring fatigue life by combining 17 explanatory variables that are highly related to the spring fatigue life and using the spring fatigue life (number of fractures) as the objective variable. This figure shows the results of predicting the spring fatigue life of Comparative Example 1. This figure shows the results of predicting the spring fatigue life of one embodiment of the present invention. This figure shows the results of generating a spring fatigue life prediction model and predicting the spring fatigue life by combining explanatory variables one by one in order of their contribution to the peak residual stress of the spring, which has the highest contribution to the prediction accuracy of the spring fatigue life. This figure shows the results of predicting the fatigue life of a spring by generating a model for predicting the fatigue life of a spring by combining four explanatory variables: stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), spring mounting height during the durability test, and decarburization, and then adding explanatory variables one by one in order of their contribution. This figure shows the results of predicting the fatigue life of a spring by changing the machine learning algorithm from an artificial neural network (ANN) to a random forest, and generating a model for predicting the fatigue life of a spring using the spring's peak residual stress, maximum cross-sectional height (Pt), arithmetic mean surface roughness (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test as explanatory variables. This figure shows the results of predicting the fatigue life of a spring by using a random forest as the machine learning algorithm, and generating a model for predicting the fatigue life of a spring using the stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test as explanatory variables.This figure shows the results of predicting the fatigue life of a spring using a spring fatigue life prediction model generated by changing the machine learning algorithm from ANN to Light Gradient Boosting Machine (LightGBM) and using the peak residual stress of the spring, the maximum cross-sectional height of the spring (Pt), the arithmetic mean roughness of the spring surface (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables. This figure shows the results of predicting the fatigue life of a spring by generating a spring fatigue life prediction model using LightGBM as the machine learning algorithm and using the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables. This figure shows the results of predicting the fatigue life of a spring using a spring fatigue life prediction model generated by changing the machine learning algorithm from ANN to CatBoost, and using the peak residual stress of the spring, the maximum cross-sectional height of the spring (Pt), the arithmetic mean roughness of the spring surface (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables. This figure shows the results of predicting the fatigue life of a spring using a spring fatigue life prediction model generated by using CatBoost as the machine learning algorithm, and using the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables. This figure shows the results of predicting the fatigue life of a spring using a spring fatigue life prediction model generated by changing the machine learning algorithm from ANN to Extreme Gradient Boosting (XGboost), with the peak residual stress of the spring, the maximum cross-sectional height of the spring (Pt), the arithmetic mean roughness of the spring surface (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables. This figure shows the results of predicting the fatigue life of a spring using a spring fatigue life prediction model generated by using XGboost as the machine learning algorithm, with the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables.

[0020] Hereinafter, one embodiment of the present invention will be described with reference to the drawings. The embodiments shown below are examples of embodiments of the present invention, and the present invention is not limited to these embodiments.

[0021] [Spring fatigue life prediction device] Figure 1 is a block diagram showing a spring fatigue life prediction device 100 according to one embodiment of the present invention. The spring fatigue life prediction device 100 includes, for example, a control device 110, an input device 120, an output device 130, a storage device 140, a communication device 150, and a power supply device 160. In one embodiment, the spring fatigue life prediction device 100 further includes, for example, a spring fatigue life prediction model generation unit 111, a spring fatigue life prediction model 113, and a spring fatigue life prediction unit 115. In one embodiment, the storage device 140 further includes a database 141.

[0022] The control device 110 consists of a known central processing unit (CPU), an operating system (OS), and a control program or module for controlling the spring fatigue life prediction device 100. Alternatively, the control device 110 may be provided as a single program including the OS and the control program or module. The control program or module constituting the control device 110 is stored in the storage device 140 and executed by the CPU.

[0023] In Figure 1, as one embodiment, the control device 110 is shown to include a durable fatigue life prediction model generation unit 111, a spring durable fatigue life prediction model 113, and a spring durable fatigue life prediction unit 115. However, the durable fatigue life prediction model generation unit 111, the spring durable fatigue life prediction model 113, and the spring durable fatigue life prediction unit 115 may not be included in the control device 110, but may be provided separately from the control device 110.

[0024] The fatigue life prediction model generation unit 111 consists of a program or module for generating a fatigue life prediction model 113. It is stored in the storage device 140 that constitutes the fatigue life prediction model generation unit 111 and executed by the CPU. In one embodiment, the fatigue life prediction model generation unit 111 performs machine learning using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, and the fatigue life of the spring at a predetermined stress amplitude as the objective variable. The fatigue life prediction model generation unit 111 generates a fatigue life prediction model for the spring by this machine learning. The peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the fatigue life of the spring may be data stored in the database 141, data input from the input device 120, or data read from an external measuring device (not shown) or server (not shown) via the communication device 150.

[0025] In one embodiment, it is preferable that the fatigue life prediction model generation unit 111 performs machine learning using the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, in addition to the stress amplitude applied to the spring during the durability test, as explanatory variables. Furthermore, in one embodiment, it is even more preferable that the fatigue life prediction model generation unit 111 performs machine learning using the maximum cross-sectional height (Pt), Rockwell hardness (HRC), and the mounting height of the spring during the durability test, in addition to the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as explanatory variables. In one embodiment, it is most preferable that the fatigue life prediction model generation unit 111 performs machine learning using explanatory variables consisting of the peak residual stress of the spring, the maximum cross-sectional height (Pt), the arithmetic mean roughness (Ra) of the spring surface, the stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and the mounting height of the spring during the durability test.

[0026] In this specification, the peak residual stress of a spring is determined by electropolishing the spring surface in 10 μm increments, measuring the residual stress at each depth using X-ray diffraction. This measurement is performed at four locations in the effective part of the spring: the inner diameter, outer diameter, and between wires A and B. The maximum compressive residual stress at each measurement location is calculated. Between wires A refers to the position directly below the spring when it is installed, and between wires B refers to the position directly above the spring when it is installed. When predicting the fatigue life, the average value of the maximum compressive residual stress at each measurement location is used.

[0027] In this specification, the maximum cross-sectional height (Pt) of the spring is measured using a roughness meter at four locations: the inner diameter, outer diameter, and between wires A and B in the effective portion of the spring. When predicting the fatigue life, the average value of the four measurements is used.

[0028] In this specification, the arithmetic mean surface roughness (Ra) of the spring is measured using a roughness meter at four locations: the inner diameter, outer diameter, and between wires A and B, within the effective portion of the spring, similar to the maximum cross-sectional height (Pt). When predicting the fatigue life, the average value of the four measurements is used.

[0029] In this specification, Rockwell hardness (HRC) is measured at eight points at 45-degree intervals using a Rockwell hardness tester at the d / 4 portion of the cross-section of the effective spring portion. The average value of the eight points is used when predicting the fatigue life.

[0030] In this specification, the mounting height of the spring during the durability test is adjusted so as to apply a predetermined stress to the spring.

[0031] Patent documents 1 and 2 describe that, in order to estimate the relationship between fracture life or fatigue life and applied stress, the mechanical properties used in a machine learning decision tree include at least one of the following: Vickers hardness, tensile strength, elongation at fracture, reduction of area at fracture, stress ratio, or fatigue test method for the steel type or metal structural material. However, as a result of the inventors' investigation, it has become clear that the accuracy of predicting the fatigue life of a spring can be improved by using a spring fatigue life prediction model 113 that has been subjected to machine learning with the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface as explanatory variables.

[0032] Furthermore, in this embodiment, it was found that a spring fatigue life prediction model 113, which was machine-trained using the spring's peak residual stress and the arithmetic mean roughness (Ra) of the spring's surface as explanatory variables, as well as the stress amplitude applied to the spring during the durability test, reduces the error (or error rate) between the model and the measured value of the spring's fatigue life. Moreover, in this embodiment, a spring fatigue life prediction model 113, which was machine-trained using the spring's peak residual stress, the spring's maximum cross-sectional height (Pt), the spring's arithmetic mean roughness (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the spring's mounting height during the durability test as explanatory variables, exhibits the smallest error (or error rate) between the model and the measured value of the spring's fatigue life. The findings regarding the combination of explanatory variables used to generate the spring fatigue life prediction model 113 by the present inventors differ from the mechanical properties described in Patent Documents 1 and 2, and represent entirely new findings that cannot be predicted from the descriptions in Patent Documents 1 and 2.

[0033] In this embodiment, the machine learning algorithm used by the fatigue life prediction model generation unit 111 is not particularly limited. Examples of machine learning algorithms include, but are not limited to, artificial neural networks (ANN), random forests, Light Gradient Boosting Machine (LightGBM), CatBoost, and Extreme Gradient Boosting (XGboost).

[0034] The spring fatigue life prediction model 113 is a trained machine learning model generated by the fatigue life prediction model generation unit 111. The generated spring fatigue life prediction model 113 is stored in the memory device 140 and used by the control device 110, specifically by the spring fatigue life prediction unit 115.

[0035] The spring fatigue life prediction unit 115 consists of a program or module for predicting the fatigue life of a spring based on the explanatory variables described above, using the spring fatigue life prediction model 113. The spring fatigue life prediction unit 115 is stored in the storage device 140 and executed by the CPU. In one embodiment, the spring fatigue life prediction unit 115 uses a spring fatigue life prediction model 113, which has been trained using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, and the fatigue life of the spring as the objective variable, to predict the fatigue life of the spring based on the explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface.

[0036] In one embodiment, the spring fatigue life prediction unit 115 preferably uses a spring fatigue life prediction model 113, which has been trained using machine learning, with explanatory variables including the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during a durability test, and the fatigue life of the spring as the objective variable, to predict the fatigue life of the spring based on explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface.

[0037] In one embodiment, the spring fatigue life prediction unit 115 more preferably uses a spring fatigue life prediction model 113, which has been trained using machine learning, to predict the spring fatigue life based on the spring fatigue life, the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, the stress amplitude applied to the spring during the durability test, as well as the maximum cross-sectional height of the spring, Rockwell hardness (HRC), and the mounting height of the spring during the durability test, with the spring fatigue life as the objective variable.

[0038] The spring fatigue life prediction unit 115 most preferably uses a spring fatigue life prediction model 113, which has been trained using machine learning, to predict the spring fatigue life based on the spring fatigue life, using the spring fatigue life as the objective variable, and explanatory variables consisting of the spring's peak residual stress, the spring's maximum cross-sectional height (Pt), the spring's arithmetic mean roughness (Ra) of the spring's surface, the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the spring's mounting height during the durability test.

[0039] In this embodiment, in order to predict the fatigue life of a spring, the fatigue life prediction unit 115 can improve the accuracy of predicting the fatigue life of a spring by using a spring fatigue life prediction model 113 that has been trained using machine learning with the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface as explanatory variables.

[0040] Furthermore, in this embodiment, by using a spring fatigue life prediction model 113 that has undergone machine learning with the stress amplitude applied to the spring during the durability test as an explanatory variable, in addition to the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, the spring fatigue life prediction unit 115 can reduce the error (or error rate) between the predicted value and the measured value of the spring fatigue life.

[0041] Furthermore, in this embodiment, by using a spring fatigue life prediction model 113 that has been machine-learned with the peak residual stress of the spring, the maximum cross-sectional height of the spring (Pt), the arithmetic mean roughness of the spring surface (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables, the spring fatigue life prediction unit 115 can minimize the error (or error rate) between the predicted value and the measured value of the spring fatigue life.

[0042] The input device 120 is a device for operating the spring durability fatigue life prediction device 100, and known input devices such as a keyboard, a mouse, and a touch panel arranged on a display device (for example, a liquid crystal display or an organic EL display) can be used. In one embodiment, the input device 120 may be used to input the above-described explanatory variables and objective variables.

[0043] The output device 130 includes a display device 131 that displays the prediction result generated by the spring durability fatigue life prediction device 100. As the display device 131, for example, a liquid crystal display, an organic EL display, or the like can be used, but it is not limited thereto. Further, the output device 130 may include a printer that prints the image displayed by the display device 131.

[0044] The storage device 140 is a device that stores an operating system (OS) and a control program or module that constitute the control device 110, a program or module that constitutes the durability fatigue life prediction model generation unit 111, a program or module that constitutes the spring durability fatigue life prediction model 113, and a program or module that constitutes the spring durability fatigue life prediction unit 115, and a database 141 including the above-described explanatory variables and objective variables. The storage device 140 is composed of, for example, a known main storage device such as a random access memory (RAM), and a known auxiliary storage device such as a read only memory (ROM), a hard disk, a solid state drive (SSD), or a memory card. The auxiliary storage device may be arranged outside the spring durability fatigue life prediction device 100 and arranged in a server or a network drive that can communicate via the communication device 150.

[0045] The communication device 150 is a known wired or wireless communication device controllable by the control device 110. The communication device 150 can be connected to communication networks such as local area networks (LANs), wide area networks (WANs), and the Internet. The communication device 150 may be a communication device conforming to wireless communication standards such as Wi-Fi® (a communication means using the IEEE 802.11 standard) or Bluetooth®. The communication device 150 may perform data communication with a server or network drive located outside the spring fatigue life prediction device 100. In one embodiment, the communication device 150 may include serial buses such as Universal Serial Bus (USB), PCI Express, and Serial ATA (SATA), and parallel buses such as Small Computer System Interface (SCSI) and Peripheral Component Interconnect (PCI).

[0046] The power supply unit 160 is a device that supplies external power to each component of the spring fatigue life prediction device 100, and is not particularly limited.

[0047] In one embodiment, if the explanatory variables and target variables described above are read from an external measuring device (not shown) or a server (not shown) via a communication device 150, a spring fatigue life prediction system may be provided that includes these externally located devices and a spring fatigue life prediction device 100.

[0048] [Method for Predicting Spring Durability Fatigue Life] FIG. 2 is a flowchart showing a method for predicting the durability fatigue of a spring according to an embodiment of the present invention. A computer that executes the method for predicting the durability fatigue of a spring according to the present embodiment, for example, the above-described spring durability fatigue life prediction device 100, reads data including at least the durability fatigue life of the spring, the peak residual stress of the spring associated with the durability fatigue life of the spring, and the arithmetic mean roughness (Ra) of the surface of the spring from the database 141 (step S110). The data including at least the durability fatigue life of the spring read by the spring durability fatigue life prediction device 100, the peak residual stress of the spring associated with the durability fatigue life of the spring, and the arithmetic mean roughness (Ra) of the surface of the spring may be data input from the input device 120, or may be data read from an external measuring device (not shown) or a server (not shown) via the communication device 150.

[0049] In a computer, for example, in the spring durability fatigue life prediction device 100, the durability fatigue life prediction model generation unit 111 performs machine learning using explanatory variables including at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the surface of the spring, and the durability fatigue life of the spring as the target variable (step S120). As described above, the machine learning algorithm used in the machine learning of the present embodiment is not particularly limited. Examples of the machine learning algorithm include, but are not limited to, artificial neural network (ANN), random forest, Light Gradient Boosting Machine (LightGBM), CatBoost, and Extreme Gradient Boosting (XGBoost).

[0050] In one embodiment, it is preferable that the fatigue life prediction model generation unit 111 performs machine learning using the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, in addition to the stress amplitude applied to the spring during the durability test, as explanatory variables. In another embodiment, it is preferable that the fatigue life prediction model generation unit 111 performs machine learning using explanatory variables including the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the stress amplitude applied to the spring during the durability test, in addition to the maximum cross-sectional height of the spring, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test. In yet another embodiment, it is more preferable that the fatigue life prediction model generation unit 111 performs machine learning using explanatory variables consisting of the peak residual stress of the spring, the maximum cross-sectional height (Pt) of the spring, the arithmetic mean roughness (Ra) of the spring surface, the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test.

[0051] The fatigue life prediction model generation unit 111 generates a fatigue life prediction model 113 for the spring using machine learning (step S130).

[0052] In the computer, for example, in the spring fatigue life prediction device 100 described above, the spring fatigue life prediction unit 115 reads prediction data corresponding to the spring fatigue life prediction model 113 (step S140). For example, if the spring fatigue life prediction model 113 is a trained model that has been machine-trained using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, and the spring fatigue life at a predetermined stress amplitude as the objective variable, then the spring fatigue life prediction unit 115 reads prediction data that includes at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface. If the spring fatigue life prediction model 113 is a trained model that has been machine-trained using explanatory variables including the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during the durability test, and the durability fatigue life of the spring as the objective variable, then the spring fatigue life prediction unit 115 reads prediction data including the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during the durability test. Furthermore, if the spring fatigue life prediction model 113 is a trained model that has been machine-trained using explanatory variables including the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the stress amplitude applied to the spring during the durability test, as well as the maximum cross-sectional height of the spring, Rockwell hardness (HRC), and the mounting height of the spring during the durability test, and the spring fatigue life as the objective variable, then the spring fatigue life prediction unit 115 reads prediction data including the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the stress amplitude applied to the spring during the durability test, as well as the maximum cross-sectional height of the spring (Pt), Rockwell hardness (HRC), and the mounting height of the spring during the durability test.

[0053] The spring fatigue life prediction unit 115 predicts the spring fatigue life based on the spring fatigue life prediction model 113 and explanatory variables corresponding to the spring fatigue life prediction model 113 (step S150).

[0054] The computer, for example, the spring fatigue life prediction device 100 described above, outputs the spring fatigue life predicted by the spring fatigue life prediction unit 115 (step S160). The spring fatigue life may be displayed on, for example, the display device 131, or printed by the output device 130. The spring fatigue life may also be stored in the storage device 140, or transmitted via the communication device 150 to a server, network drive, or external terminal located outside the spring fatigue life prediction device 100 for storage.

[0055] [Spring Fatigue Life Prediction Program] Referring to Figure 2, a spring fatigue prediction program according to one embodiment of the present invention will be described. The spring fatigue prediction program causes a computer, for example, the spring fatigue life prediction device 100 described above, to read data from the database 141 that includes at least the spring fatigue life, the peak residual stress of the spring associated with the spring fatigue life, and the arithmetic mean roughness (Ra) of the spring surface (step S110). The data that the spring fatigue life prediction device 100 reads, which includes at least the spring fatigue life, the peak residual stress of the spring associated with the spring fatigue life, and the arithmetic mean roughness (Ra) of the spring surface, may be data input from the input device 120, or data read from an external measuring device (not shown) or a server (not shown) via the communication device 150.

[0056] The spring fatigue endurance prediction program causes a computer, for example, in the spring fatigue endurance life prediction device 100, to perform machine learning using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, and the spring fatigue endurance life at a predetermined stress amplitude as the objective variable (step S120). As described above, the machine learning algorithm used in the machine learning of this embodiment is not particularly limited. Examples of machine learning algorithms include, but are not limited to, artificial neural networks (ANN), random forests, Light Gradient Boosting Machine (LightGBM), CatBoost, and Extreme Gradient Boosting (XGboost).

[0057] In one embodiment, it is preferable that the spring fatigue prediction program causes the fatigue life prediction model generation unit 111 to perform machine learning using explanatory variables including the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during the durability test. In another embodiment, it is preferable that the spring fatigue prediction program causes the fatigue life prediction model generation unit 111 to perform machine learning using explanatory variables including the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the stress amplitude applied to the spring during the durability test, as well as the maximum cross-sectional height (Pt) of the spring, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test. In one embodiment, the spring fatigue prediction program more preferably uses machine learning to have the fatigue life prediction model generation unit 111 perform machine learning using explanatory variables consisting of the peak residual stress of the spring, the maximum cross-sectional height of the spring (Pt), the arithmetic mean roughness of the spring surface (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test.

[0058] The spring fatigue prediction program causes the fatigue life prediction model generation unit 111 to generate a spring fatigue life prediction model 113 using machine learning (step S130).

[0059] The spring fatigue prediction program causes a computer, for example, the spring fatigue life prediction unit 115 in the spring fatigue life prediction device 100 described above, to read prediction data corresponding to the spring fatigue life prediction model 113 (step S140). For example, if the spring fatigue life prediction model 113 is a trained model that has been machine-trained using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, and the spring fatigue life at a predetermined stress amplitude as the objective variable, the spring fatigue prediction program causes the spring fatigue life prediction unit 115 to read prediction data that includes at least the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface. If the spring fatigue life prediction model 113 is a trained model that has been machine-trained using explanatory variables including the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during the durability test, and the spring fatigue life as the objective variable, then the spring fatigue prediction program will be loaded with prediction data including the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during the durability test. Furthermore, in one embodiment, if the spring fatigue life prediction model 113 is a trained model that has been machine-trained using explanatory variables including the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the stress amplitude applied to the spring during the durability test, as well as the maximum cross-sectional height (Pt) of the spring, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test, and the spring fatigue life as the objective variable, then the spring fatigue prediction program causes the spring fatigue life prediction unit 115 to read prediction data including the peak residual stress of the spring, the maximum cross-sectional height (Pt) of the spring, and the arithmetic mean roughness (Ra) of the spring surface, as well as the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test.

[0060] The spring fatigue prediction program causes the spring fatigue life prediction unit 115 to predict the spring fatigue life based on the spring fatigue life prediction model 113 and explanatory variables corresponding to the spring fatigue life prediction model 113 (step S150).

[0061] The spring fatigue prediction program causes a computer, for example, the spring fatigue life prediction device 100 described above, to output the spring fatigue life predicted by the spring fatigue life prediction unit 115 (step S160). The spring fatigue prediction program may, for example, display the spring fatigue life on a display device 131, or print the fatigue life on an output device 130. The spring fatigue prediction program may also store the spring fatigue life in a storage device 140, or transmit the spring fatigue life via a communication device 150 to a server, network drive, or external terminal located outside the prediction device 100 for storage.

[0062] In one embodiment, a spring fatigue prediction program can be provided stored on a computer-readable recording medium. The computer-readable recording medium may be a flexible disk, compact disk, Blu-ray disk, USB memory, memory card, hard disk, or solid-state drive (SSD), etc.

[0063] To generate a spring fatigue life prediction model 113 for predicting the fatigue life of a spring, the following explanatory variables related to the fatigue life of the spring were used: the spring temperature during shot peening, the projection velocity and amount of shot material used for shot peening, and the amount of stress applied to the spring during shot peening, all derived from the shot peening conditions (SP conditions). In addition, the following were used from the test conditions of the spring fatigue life test: the amplitude of the spring during the durability test, the mounting height of the spring during the durability test, and the amplitude of the stress applied to the spring during the durability test. Furthermore, from the analysis values ​​of the springs used in the spring fatigue life test, the following parameters were used: Vickers hardness at 0.05 mm from the surface of the inner diameter of the spring, Vickers hardness at 0.05 mm from the surface of the outer diameter of the spring, Rockwell hardness (HRC) of the spring, grain size of the spring surface, grain size of the spring d / 4 portion, grain size of the spring center, decarburization depth of the spring, arithmetic mean roughness (Ra) of the spring surface, maximum cross-sectional height (Pt) of the spring surface, surface residual stress of the spring, and peak residual stress of the spring. The Vickers hardness at 0.05 mm from the surface of the inner diameter of the spring and the Vickers hardness at 0.05 mm from the surface of the outer diameter of the spring were measured using a micro-Vickers hardness tester at a point 0.05 mm from the surface of the inner and outer diameters of the spring. The grain size of the spring surface, the d / 4 portion of the spring, and the center of the spring were measured at the surface, d / 4 portion, and center of the effective spring portion according to the comparative method of JIS G0551. The decarburization depth was measured at four locations on the surface of the effective spring portion: the inner diameter portion, the outer diameter portion, the A side between wires, and the B side between wires, according to JIS G0558. The average value of the four measurements was used to predict the lifespan. The residual stress on the surface of the spring was measured using X-ray diffraction. This measurement was performed at four locations on the effective spring portion: the inner diameter portion, the outer diameter portion, the A side between wires, and the B side between wires. The average value of the four measurements was used to predict the lifespan.

[0064] Using eleven types of springs that had undergone shot peening, a fatigue and durability testing machine (servopulsar) was used to vibrate the springs under predetermined stresses, and the fatigue life of the springs was tested. The fatigue life (number of fractures) of the springs was measured. In addition, a database was created recording the explanatory variables of the springs whose fatigue life was measured.

[0065] Using the spring fatigue life prediction device described above, a spring fatigue life prediction model was generated using each explanatory variable and the spring fatigue life (number of fractures) as the objective variable, and the spring fatigue life was predicted. As a machine learning algorithm, an artificial neural network (ANN) was used to perform machine learning and generate the spring fatigue life prediction model.

[0066] A spring fatigue life prediction model was generated, and the fatigue life of the spring was predicted. The prediction accuracy was then examined. The prediction results are shown in Figure 3. In the prediction results in Figure 3, the error rate with the measured number of fractures was 56.8%, and none of the prediction results had practical prediction accuracy.

[0067] Based on the results above, it was considered that the number of explanatory variables was too large relative to the number of data points, so we considered improving the prediction accuracy by removing unnecessary explanatory variables. Figure 4 is a table showing the results of generating a spring fatigue life prediction model and predicting the spring fatigue life by combining 17 explanatory variables that were highly related to the spring's fatigue life from among the explanatory variables mentioned above, and using the spring's fatigue life (number of fractures) as the objective variable. An artificial neural network (ANN) was used as the machine learning algorithm. In Figure 4, the spring fatigue life prediction model was generated by removing one explanatory variable at a time that had the lowest contribution to the prediction accuracy, and the spring's fatigue life was predicted.

[0068] In this specification, "contribution" is the average of the contributions calculated using six algorithms: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Lasso regression, artificial neural networks, random forests, and Shapley Additional exPlanations (SHAP). In Figure 4, reducing the number of explanatory variables by removing the explanatory variable with the lowest contribution improved the accuracy of predicting the fatigue life of the spring when generating a spring fatigue life prediction model.

[0069] However, it was found that reducing the number of explanatory variables sometimes resulted in a decrease in the accuracy of predicting the fatigue life of the spring compared to before the reduction, indicating that reducing the number of explanatory variables does not necessarily improve the accuracy of predicting the fatigue life of the spring. Furthermore, it was found that the highest accuracy in predicting the fatigue life of a spring was achieved when using a spring fatigue life prediction model generated with the spring's peak residual stress, maximum cross-sectional height (Pt), arithmetic mean surface roughness (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test as explanatory variables.

[0070] [Comparative Example 1] In Patent Documents 1 and 2, it is stated that the test includes at least one of the following mechanical properties: Vickers hardness, tensile strength, elongation at break, reduction of area at break, stress ratio, or fatigue test method for steel grade or metal structural material. The explanatory variables corresponding to these mechanical properties are the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test. Using these three explanatory variables, a model for predicting the durability fatigue life of a spring was generated, and the durability fatigue life of the spring was predicted. Figure 5A shows the results of generating a model for predicting the durability fatigue life of a spring and predicting the durability fatigue life of the spring using the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test as explanatory variables. In the prediction results in Figure 5A, the error rate with the measured value of the number of fractures was 59.5%.

[0071] [Example 1] Figure 5B shows the results of predicting the fatigue life of a spring using a spring fatigue life prediction model generated with the spring's peak residual stress, maximum cross-sectional height (Pt), arithmetic mean surface roughness (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test as explanatory variables. Compared with Figure 5A, Figure 5B shows a high correlation between the measured number of spring fractures and the predicted number of spring fractures. In the prediction results in Figure 5B, the error rate between the measured number of fractures and the predicted number was 17.3%. In other words, in this example, it became clear that by using six explanatory variables—the spring's peak residual stress, maximum cross-sectional height (Pt), arithmetic mean surface roughness (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test—it is possible to predict with dramatically higher accuracy than when using the explanatory variables used in Patent Documents 1 and 2.

[0072] [Example 2] In order to arrive at the combination of six explanatory variables of Example 1, we examined how to remove unnecessary explanatory variables one by one. In Example 2, we examined the degree to which each explanatory variable contributes to the accuracy of predicting the fatigue life of the spring, and examined combinations of explanatory variables with high contributions. Using each explanatory variable, we generated a fatigue life prediction model for the spring and predicted the fatigue life of the spring. The prediction results are shown in Table 1. From the results in Table 1, it became clear that the peak residual stress of the spring, the maximum cross-sectional height of the spring (Pt), and the arithmetic mean roughness of the spring surface (Ra) contribute more to the accuracy of predicting the fatigue life of the spring than the stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and the mounting height of the spring during the durability test used in Patent Documents 1 and 2.

[0073]

[0074] Next, a spring fatigue life prediction model was generated by combining the spring's peak residual stress, which has the highest contribution to the prediction accuracy of the spring's fatigue life, with explanatory variables in descending order of their contribution, and the spring's fatigue life was predicted. The prediction results are shown in Figure 6. From the results in Figure 6, it was found that the spring's peak residual stress, the spring's maximum cross-sectional height (Pt), and the spring's arithmetic mean roughness (Ra) of the spring surface are essential combinations of explanatory variables for predicting the spring's fatigue life. It was also found that the combination of the spring's peak residual stress, the spring's maximum cross-sectional height (Pt), the spring's arithmetic mean roughness (Ra), the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the spring's mounting height during the durability test yields the highest prediction accuracy for the spring's fatigue life.

[0075] [Example 3] In Figures 4 and 6, the combination of the three explanatory variables is consistent for the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the maximum cross-sectional height (Pt) of the spring. We investigated whether this combination is appropriate for predicting the fatigue life of the spring. In Example 3, the maximum cross-sectional height (Pt) of the spring was changed to the stress amplitude applied to the spring during the durability test as an explanatory variable, and this was combined with the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface.

[0076] Table 2 shows the predicted fatigue life of the spring in Example 3. Table 2 also shows the predicted fatigue life for combinations of explanatory variables: the peak residual stress of the spring, the arithmetic mean roughness (Ra) of the spring surface, and the maximum cross-sectional height (Pt), as well as combinations of explanatory variables: the peak residual stress of the spring, the stress amplitude applied to the spring during the durability test, the arithmetic mean roughness (Ra) of the spring surface, and the maximum cross-sectional height (Pt). From the results in Table 2, it is clear that Example 3, which selected the peak residual stress of the spring, the stress amplitude applied to the spring during the durability test, and the arithmetic mean roughness (Ra) of the spring surface as explanatory variables, exhibits a lower error rate and is more appropriate as a combination of explanatory variables for predicting the fatigue life of the spring.

[0077]

[0078] [Example 4] In Example 4, the fatigue life of a spring was predicted using a fatigue life prediction model generated with the stress amplitude applied to the spring during the durability test fixed as the objective variable, and the peak residual stress of the spring and the arithmetic mean roughness (Ra) of the spring surface as the explanatory variables. The results of the fatigue life prediction for the spring in Example 4 are shown in Table 3. Comparing the results in Table 2 and Table 3, it became clear that the error rates were equivalent.

[0079]

[0080] [Comparative Example 2] Here, we examined whether the combination of the three explanatory variables used in Patent Documents 1 and 2—stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test—is an essential combination. A spring fatigue life prediction model was generated by combining the three explanatory variables—stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test—with one other explanatory variable, and the fatigue life of the spring was predicted. The prediction results are shown in Table 4. From the results in Table 2, it was shown that when the three explanatory variables—stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test—are set as the basic combination, decarburization provides the highest accuracy in predicting the fatigue life of the spring.

[0081]

[0082] A spring fatigue life prediction model was generated by combining four explanatory variables—stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), spring mounting height during the durability test, and decarburization—with other explanatory variables in order of their contribution, in descending order of importance, and the spring's fatigue life was predicted. The prediction results are shown in Figure 7. From the results in Figure 7, it was shown that the combination of stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), spring mounting height during the durability test, and decarburization, along with temperature, peak residual stress, surface residual stress, maximum cross-sectional height of the spring (Pt), amplitude, stress amount, and arithmetic mean roughness (Ra) of the spring surface, yielded the highest accuracy in predicting the spring's fatigue life. However, this combination yielded lower prediction accuracy for the spring's fatigue life than the combination of the spring's peak residual stress, maximum cross-sectional height (Pt), arithmetic mean surface roughness (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test. It became clear that studies using the stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and spring mounting height during the durability test as the basic combination could not arrive at the combination of explanatory variables that yielded the highest prediction accuracy for the spring's fatigue life.

[0083] [Example 5] In predicting the fatigue life of a spring, it was verified that the combination of explanatory variables according to the present invention does not depend on the machine learning algorithm. The machine learning algorithm was changed from an artificial neural network (ANN) to a random forest, and the fatigue life of the spring was predicted using a spring fatigue life prediction model generated with the same spring peak residual stress, maximum cross-sectional height of the spring (Pt), arithmetic mean roughness of the spring surface (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and mounting height of the spring during the durability test as in Example 1. The prediction results are shown in Figure 8A. Figure 8A shows a high correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 8A was 25.1%.

[0084] [Comparative Example 3] Similar to Example 3, a random forest machine learning algorithm was used, and similar to Comparative Example 1, the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test were used as explanatory variables to generate a spring fatigue life prediction model and predict the spring fatigue life. The prediction results are shown in Figure 8B. Figure 8B shows that there is a low correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 8B was 62.3%.

[0085] [Example 6] The machine learning algorithm was changed from ANN to Light Gradient Boosting Machine (LightGBM), and the fatigue life of the spring was predicted using a spring fatigue life prediction model generated with the same spring peak residual stress, maximum cross-sectional height of the spring (Pt), arithmetic mean surface roughness of the spring (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and mounting height of the spring during the durability test as in Example 1 as explanatory variables. The prediction results are shown in Figure 9A. Figure 9A shows a high correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 9A was 13.7%.

[0086] [Comparative Example 4] Similar to Example 4, LightGBM was used as the machine learning algorithm, and similar to Comparative Example 1, the stress amplitude applied to the spring during the durability test, the Rockwell hardness (HRC), and the mounting height of the spring during the durability test were used as explanatory variables to generate a spring fatigue life prediction model and predict the spring fatigue life. The prediction results are shown in Figure 9B. Figure 9B shows that there is a low correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 9B was 62.7%.

[0087] [Example 7] The machine learning algorithm was changed from ANN to CatBoost, and the fatigue life of the spring was predicted using a spring fatigue life prediction model generated with the same spring peak residual stress, maximum cross-sectional height of the spring (Pt), arithmetic mean surface roughness of the spring (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and mounting height of the spring during the durability test as in Example 1 as explanatory variables. The prediction results are shown in Figure 10A. Figure 10A shows a high correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 10A was 18.4%.

[0088] [Comparative Example 5] Similar to Example 5, CatBoost was used as the machine learning algorithm, and similar to Comparative Example 1, the stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and the mounting height of the spring during the durability test were used as explanatory variables to generate a spring fatigue life prediction model and predict the spring fatigue life. The prediction results are shown in Figure 10B. Figure 10B shows that there is a low correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 10B was 58.6%.

[0089] [Example 8] The machine learning algorithm was changed from ANN to Extreme Gradient Boosting (XGboost), and the fatigue life of the spring was predicted using a spring fatigue life prediction model generated with the same spring peak residual stress, maximum cross-sectional height of the spring (Pt), arithmetic mean roughness of the spring surface (Ra), stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and mounting height of the spring during the durability test as in Example 1 as explanatory variables. The prediction results are shown in Figure 11A. Figure 11A shows a high correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 11A was 13.8%.

[0090] [Comparative Example 6] Similar to Example 6, XGboost was used as the machine learning algorithm, and similar to Comparative Example 1, the stress amplitude applied to the spring during the durability test, Rockwell hardness (HRC), and the mounting height of the spring during the durability test were used as explanatory variables to generate a spring fatigue life prediction model and predict the spring fatigue life. The prediction results are shown in Figure 11B. Figure 11B shows that there is a low correlation between the measured number of spring breaks and the predicted number of spring breaks. In addition, the error rate between the predicted result and the measured number of breaks in Figure 11B was 55.1%.

[0091] The results from Examples 1 and 5-8 clearly demonstrate that the combination of explanatory variables according to the present invention makes it possible to generate a machine learning model that improves the accuracy of predicting spring fatigue life, without depending on the machine learning algorithm.

[0092] Although one embodiment of the present invention has been described above with reference to the drawings, the present invention is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit of the invention. For example, a spring fatigue life prediction device, spring fatigue life prediction method, and spring fatigue life prediction program of this embodiment, with additions, deletions, or design changes made by a person skilled in the art, is also included in the scope of the present invention as long as it retains the gist of the present invention. Furthermore, the above-described embodiments can be combined as appropriate as long as they do not contradict each other, and technical matters common to each embodiment are included in each embodiment even without explicit description.

[0093] Any effects or benefits other than those brought about by the embodiments described above, if they are clear from the description herein or easily predictable to a person skilled in the art, are naturally considered to be brought about by the present invention.

[0094] 100 Spring fatigue life prediction device, 110 Control device, 111 Spring fatigue life prediction model generation unit, 113 Spring fatigue life prediction model, 115 Spring fatigue life prediction unit, 120 Input device, 130 Output device, 140 Storage device, 141 Database, 150 Communication device, 160 Power supply device

Claims

1. A spring fatigue life prediction device comprising: a database including at least the fatigue life of a spring, the peak residual stress of the spring associated with the fatigue life of the spring, and the arithmetic mean roughness of the spring surface; an explanatory variable including at least the peak residual stress of the spring and the arithmetic mean roughness of the spring surface; and a unit for generating a spring fatigue life prediction model by machine learning using the fatigue life of the spring at a predetermined stress amplitude as the objective variable; and a unit for predicting the fatigue life of a spring using the spring fatigue life prediction model and predicting the fatigue life of the spring based on the explanatory variable.

2. The spring fatigue life prediction device according to claim 1, wherein the database further includes the stress amplitude applied to the spring during a durability test associated with the fatigue life of the spring, and the explanatory variable further includes the stress amplitude applied to the spring during the durability test.

3. The spring fatigue life prediction device according to claim 2, wherein the database further includes the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the durability test, and the explanatory variables further include the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the durability test.

4. The spring fatigue life prediction device according to claim 3, wherein the database comprises the fatigue life of the spring, the peak residual stress of the spring associated with the fatigue life of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test, and the explanatory variables consist of the peak residual stress of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test.

5. A method for predicting the fatigue life of a spring, comprising: a computer reading data from a database that includes at least the fatigue life of a spring, the peak residual stress of the spring associated with the fatigue life of the spring, and the arithmetic mean roughness of the surface of the spring; the computer generating a fatigue life prediction model for a spring by machine learning using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness of the surface of the spring, and the fatigue life of the spring at a predetermined stress amplitude as the objective variable; and the computer predicting the fatigue life of the spring based on the explanatory variables using the fatigue life prediction model for a spring.

6. The method for predicting the endurance fatigue life of a spring according to claim 5, wherein the computer further includes from the database the stress amplitude applied to the spring during an endurance test associated with the endurance fatigue life of the spring, and the explanatory variable further includes the stress amplitude applied to the spring during the endurance test.

7. The method for predicting the endurance fatigue life of a spring according to claim 6, wherein the computer reads from the database data further including the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the endurance test, and the explanatory variables further include the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the endurance test.

8. A method for predicting the fatigue life of a spring according to claim 7, wherein the computer reads data from the database comprising the fatigue life of the spring, the peak residual stress of the spring associated with the fatigue life of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test, and the explanatory variables consist of the peak residual stress of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test.

9. A spring fatigue life prediction program comprising: loading data from a database into a computer that includes at least the fatigue life of a spring, the peak residual stress of the spring associated with the fatigue life of the spring, and the arithmetic mean roughness of the spring surface; generating a spring fatigue life prediction model in the computer by machine learning using explanatory variables that include at least the peak residual stress of the spring and the arithmetic mean roughness of the spring surface, and the fatigue life of the spring at a predetermined stress amplitude as the objective variable; and using the spring fatigue life prediction model, causing the computer to predict the fatigue life of the spring based on the explanatory variables.

10. A spring fatigue life prediction program according to claim 9, wherein the computer is loaded from the database with data further including the stress amplitude applied to the spring during a durability test associated with the fatigue life of the spring, and the explanatory variable further includes the stress amplitude applied to the spring during the durability test.

11. A spring fatigue life prediction program according to claim 10, wherein the computer is loaded from the database with data further including the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the durability test, and the explanatory variables further include the maximum cross-sectional height of the spring, the Rockwell hardness, and the mounting height of the spring during the durability test.

12. A spring fatigue life prediction program according to claim 11, wherein the computer is loaded with data from the database, comprising the fatigue life of the spring, the peak residual stress of the spring associated with the fatigue life of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test, and the explanatory variables consist of the peak residual stress of the spring, the maximum cross-sectional height of the spring, the arithmetic mean roughness of the surface of the spring, the stress amplitude applied to the spring during the durability test, the Rockwell hardness, and the mounting height of the spring during the durability test.

Images