Estimation method, learning method, estimation device, estimation system, learning device, and computer program

Abstract

To estimate the properties of a metallic material having a specified composition. [Solution] An estimation method comprising: a first learning process that acquires first trained model information by a learning process using training data that includes condition information about metal materials and characteristic information indicating the pitting corrosion resistance of metal materials according to the condition information; a second learning process that extracts only those characteristic information from the training data in which it is judged that pitting corrosion can occur based on predetermined criteria, and acquires second trained model information by a learning process using the extracted training data; a first estimation process that uses the first trained model information and the condition information of the metal material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process that extracts only those estimated to be capable of pitting corrosion as a result of the first estimation process, and estimates information indicating the pitting corrosion resistance of metal materials according to the condition information of the estimated target using the condition information of the extracted estimated target and the second trained model information.

Description

Technical Field

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

Background Art

[0002] Conventionally, research has been underway to improve the material properties of metal materials. For example, in stainless steel, the pitting corrosion resistance changes depending on the alloying elements contained. For example, it has been said that the higher the pitting resistance equivalent number (PREN), which is one of the values related to the ability of stainless steel to resist pitting corrosion, the higher the pitting corrosion resistance of the stainless steel. PREN may be expressed as, for example, Cr + 3.3Mo + 16N (mass%).

[0003] However, simply improving the material properties of metal materials may cause other problems. For example, the alloying elements contained in stainless steel are raw materials of metals refined from various minerals. The price of minerals is at risk of rising due to speculative or political influences. The burden of rising mineral prices will be borne by material manufacturers, distributors, and consumers. Therefore, it is required to minimize raw material costs while ensuring the necessary material properties. For this reason, it is important to clarify the influence of alloying elements contained in metal materials such as stainless steel on material properties. For example, as a document regarding the influence of alloying elements on the material properties of metal materials, there is Patent Document 1 shown below.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the technology disclosed in Patent Document 1 can only linearly express the influence of alloying elements on the material properties of metallic materials. Therefore, the technology disclosed in Patent Document 1 cannot quantify the synergistic effect when multiple types of alloying elements are present. Furthermore, experiments to investigate the influence of alloying elements contained in metallic materials on the material properties of metallic materials need to be conducted for each metallic material with a vast number of chemical compositions. This may place a heavy burden on the experimenter. For example, if it were possible to estimate the properties of metallic materials with predetermined components, this burden could be reduced.

[0006] This invention has been made in view of the above circumstances and provides a technology that makes it possible to estimate the properties of a metallic material having a predetermined composition. [Means for solving the problem]

[0007] One aspect of the present invention is an estimation method comprising: a first learning process that performs a learning process using training data including condition information which is information indicating conditions relating to a metallic material, and characteristic information which indicates the pitting corrosion resistance of the metallic material according to the conditions indicated by the condition information, and obtains first trained model information; a second learning process that extracts only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, performs a learning process using the extracted training data, and obtains second trained model information; a first estimation process that uses the first trained model information and the condition information of the metallic material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process that extracts only the estimated targets which are estimated to be capable of causing pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metallic material according to the condition information of the estimated targets.

[0008] One aspect of the present invention is the estimation method described above, wherein the information indicating pitting corrosion resistance is information regarding the pitting potential of the metal material, and the second learning process is a learning process that uses only the training data from which the value of the pitting potential indicates a value that can cause pitting corrosion.

[0009] One aspect of the present invention is the estimation method described above, wherein the metallic material is stainless steel, and the value at which pitting corrosion can occur based on the predetermined criteria is a pitting potential value of less than 1.0V.

[0010] One aspect of the present invention is the estimation method described above, wherein the condition information includes information on the content of a plurality of elements that affect the pitting corrosion resistance contained in the metal material to be estimated.

[0011] One aspect of the present invention is the estimation method described above, wherein the condition information includes information on the content of silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, and nitrogen as the content of the plurality of elements.

[0012] One aspect of the present invention is the estimation method described above, wherein the condition information includes information relating to the state of the metal structure of the metallic material.

[0013] One aspect of the present invention is the estimation method described above, wherein the information relating to the state of the metal structure includes information indicating whether it is a single-phase structure or a multi-phase structure.

[0014] One aspect of the present invention is the estimation method described above, wherein the metallic material is stainless steel, and the information relating to the state of the metal structure includes information indicating whether it is a duplex stainless steel or a non-duplex stainless steel.

[0015] One aspect of the present invention is the estimation method described above, wherein the condition information includes information regarding the usage environment of the metal material to be estimated.

[0016] One aspect of the present invention is the estimation method described above, wherein the information relating to the usage environment includes information relating to temperature and chloride ion concentration.

[0017] One aspect of the present invention is the estimation method described above, wherein in the second estimation process, the logarithm is taken with respect to the value included in the condition information, and the pitting corrosion resistance of the metal material is estimated using the obtained value.

[0018] One aspect of the present invention is the estimation method described above, wherein in the second estimation process, the logarithm is taken of the value included in the information indicating pitting corrosion resistance, and the pitting corrosion resistance of the metal material is estimated using the obtained value.

[0019] One aspect of the present invention is a learning method comprising: a first learning process that performs a learning process using training data which includes condition information which is information indicating conditions relating to a metallic material and characteristic information which indicates the pitting corrosion resistance of the metallic material according to the conditions indicated by the condition information, and obtains first trained model information; and a second learning process that extracts only the training data from which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, performs a learning process using the extracted training data, and obtains second trained model information.

[0020] One aspect of the present invention is an estimation method comprising: a first estimation process that uses first trained model information and condition information of a metal material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process that extracts only those estimated to be capable of pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets, wherein the first trained model information is information obtained by performing a training process using training data that includes condition information which is information indicating conditions related to the metal material and characteristic information which indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information; and the second trained model information is information obtained by performing a training process using training data obtained by extracting only the training data from which the characteristic information is judged to be capable of pitting corrosion based on predetermined criteria.

[0021] One aspect of the present invention is an estimation device comprising: a storage unit that stores first trained model information obtained by performing a learning process using training data which includes condition information which is information indicating conditions relating to a metal material and characteristic information which indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information; second trained model information obtained by extracting only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, and performing a learning process using the extracted training data; a control unit that performs a first estimation process which uses the first trained model information and the condition information of the metal material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process which extracts only the estimated targets which are estimated to be capable of causing pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets.

[0022] One aspect of the present invention is an estimation system comprising: a learning device comprising a control unit that performs a learning process using training data including condition information which is information indicating conditions relating to a metal material and characteristic information indicating the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information, and acquires first trained model information; a second learning process which extracts only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, performs a learning process using the extracted training data, and acquires second trained model information; and an estimation device comprising a control unit that performs a first estimation process which uses the first trained model information and the condition information of the metal material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process which extracts only the estimated targets that are estimated to be capable of causing pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated target.

[0023] One aspect of the present invention is a computer program for causing a computer to function as an estimation device, comprising: a storage unit that stores first trained model information obtained by performing a learning process using training data which includes condition information that indicates conditions relating to a metal material and characteristic information that indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information; second trained model information obtained by extracting only the training data from which the characteristic information is determined to be capable of causing pitting corrosion based on predetermined criteria, and performing a learning process using the extracted training data; and a control unit that performs a first estimation process which uses the first trained model information and the condition information of the metal material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process which extracts only the estimated targets that are estimated to be capable of causing pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets.

[0024] One aspect of the present invention performs learning processing using teacher data including condition information indicating conditions related to a metal material and characteristic information indicating the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information, to obtain first learned model information, and first learning processing; only extracts teacher data determined to be likely to cause pitting corrosion based on a predetermined criterion from the teacher data, performs learning processing using the extracted teacher data, and second learning processing for obtaining second learned model information, and is a learning device provided with a control unit that performs the above.

[0025] One aspect of the present invention is a computer program for causing a computer to function as a learning device provided with a control unit that performs first learning processing for obtaining first learned model information by performing learning processing using teacher data including condition information indicating conditions related to a metal material and characteristic information indicating the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information, and second learning processing for extracting only teacher data determined to be likely to cause pitting corrosion based on a predetermined criterion from the teacher data, performing learning processing using the extracted teacher data, and obtaining second learned model information.

Advantages of the Invention

[0026] According to the present invention, it becomes possible to estimate the characteristics of a metal material containing a predetermined component. Therefore, it becomes possible to reduce the burden that occurs in determining the chemical composition of a metal material. More specifically, it becomes possible to reduce some or all of the burden that has heretofore been required for actual experiments.

Brief Description of the Drawings

[0027] [Figure 1] It is a schematic block diagram showing the system configuration of the estimation system 100 of the present invention. [Figure 2] It is a schematic block diagram showing a specific example of the functional configuration of the terminal device 10. [Figure 3] It is a schematic block diagram showing a specific example of the functional configuration of the learning device 20. [Figure 4]This figure shows information regarding specific examples of training data used in the learning device 20. [Figure 5] This figure shows a more specific example of the training data used in the learning device 20. [Figure 6] This shows the values ​​of the training data after preprocessing to add the pitting index. [Figure 7] This shows the values ​​of the training data after preprocessing, in which a predetermined value is added to the pitting potential (Vpit) value. [Figure 8] This shows the values ​​of the training data after logarithmic transformation processing. [Figure 9] This figure shows a concrete example of second-stage preprocessed training data. [Figure 10] This flowchart shows a specific example of the processing performed by the learning device 20. [Figure 11] This is a schematic block diagram showing a specific example of the functional configuration of the estimation device 30. [Figure 12] This figure shows a concrete example of conditional information. [Figure 13] This figure shows a specific example of the results of the first estimation process. [Figure 14] This figure shows a specific example of condition information after the second pretreatment has been performed. [Figure 15] This figure shows a specific example of the results after the second estimation process has been performed. [Figure 16] This figure shows a concrete example of the results obtained by estimating characteristic information. [Figure 17] This flowchart shows a specific example of the processing performed by the estimation device 30. [Figure 18] This figure shows a specific example of training data in the second embodiment. [Figure 19] This figure shows an example of the relationship between experimentally determined pitting potential (Vpit) and estimated pitting potential (Vpit) obtained by conventional estimation using trained model information obtained through conventional learning processes. [Figure 20]This figure shows an example of the relationship between experimentally determined pitting potential (Vpit) and estimated pitting potential (Vpit) obtained by performing learning and estimation processing using the pitting occurrence index. [Figure 21] This figure shows an example of the relationship between experimentally determined pitting potential (Vpit) and estimated pitting potential (Vpit) obtained by performing learning and estimation processing using the pitting index and phase. [Figure 22] This figure shows an example of the relationship between the estimated pitting index and the measured pitting potential. [Figure 23] This graph shows the relationship between chromium (Cr) content and pitting potential. [Figure 24] This figure shows a schematic example of the hardware configuration of the information processing device 90 applied to this embodiment. [Figure 25] This figure shows a modified example of the estimation device 30. [Figure 26] This figure shows a modified example of the estimation device 30. [Modes for carrying out the invention]

[0028] [Summary] The following describes specific examples of the present invention with reference to the drawings. Figure 1 is a schematic block diagram showing the system configuration of the estimation system 100 of the present invention. First, the outline of the estimation system 100 will be explained. The estimation system 100 is used when estimating the properties of a metallic material having predetermined components. Hereinafter, the person who performs the operations to make such an estimation will be referred to as the user. The user inputs information indicating conditions related to the metallic material to be estimated (hereinafter referred to as "condition information") into the terminal device 10. Specific examples of such condition information include information on the chemical composition of the metallic material to be estimated and information on factors that affect the properties of the metallic material to be estimated.

[0029] The terminal device 10 transmits the input condition information to the estimation device 30. The estimation device 30 estimates the properties of the metallic material according to the conditions indicated by the condition information. The type of properties to be estimated may be one or multiple types. In this embodiment, a specific example of a property to be estimated is pitting corrosion resistance. A specific example of information indicating pitting corrosion resistance is the pitting potential (Vpit). The unit of the value of the pitting potential may be, for example, V (volt) (based on a silver-silver chloride reference electrode (SSE)). The estimation device 30 transmits information indicating the estimated properties (property information) to the terminal device 10. When the terminal device 10 receives the property information, it outputs the received property information. In this way, the user can easily obtain estimation results for the properties of a metallic material having a composition to be estimated without having to conduct experiments using metallic materials with various actual chemical compositions.

[0030] [System Details: First Embodiment] Next, the details of the first embodiment of the estimation system 100 will be described. The estimation system 100 includes a terminal device 10, a learning device 20, and an estimation device 30. The terminal device 10 and the estimation device 30 are connected communicatively via a network 70. The learning device 20 and the estimation device 30 may also be connected communicatively via the network 70. The network 70 may be a wireless communication network or a wired communication network. The network 70 may be configured using, for example, the Internet or a local area network (LAN). The network 70 may be configured by combining multiple networks.

[0031] Figure 2 is a schematic block diagram showing a specific example of the functional configuration of the terminal device 10. The terminal device 10 is configured using information devices such as a smartphone, tablet, personal computer, or dedicated device. The terminal device 10 includes a communication unit 11, an operation unit 12, an output unit 13, a storage unit 14, and a control unit 15.

[0032] The communication unit 11 is a communication device. The communication unit 11 may be configured, for example, as a network interface. The communication unit 11 communicates data with other devices via the network 70 in accordance with the control of the control unit 15. The communication unit 11 may be a wireless communication device or a wired communication device.

[0033] The operation unit 12 is configured using existing input devices such as a keyboard, pointing device (mouse, tablet, etc.), buttons, or touch panel. The operation unit 12 is operated by the user when inputting user instructions to the terminal device 10. The operation unit 12 may also be an interface for connecting the input device to the terminal device 10. In this case, the operation unit 12 inputs the input signal generated in response to the user's input in the input device to the terminal device 10. The operation unit 12 may also be configured using a microphone and a speech recognition device. In this case, the operation unit 12 performs speech recognition of the words spoken by the user and inputs the recognized string information to the terminal device 10. In this case, the operation unit 12 may only perform voice input, and speech recognition may be performed by the control unit 15. The operation unit 12 may be configured in any way that allows user instructions to be input to the terminal device 10.

[0034] The output unit 13 outputs information in a format that the user can recognize. The output unit 13 may be an image display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display. The output unit 13 may also be an interface for connecting an image display device to the terminal device 10. In this case, the output unit 13 generates a video signal for displaying image data and outputs the video signal to the image display device connected to it. The output unit 13 may also be a device that outputs sound, such as a speaker. The output unit 13 may also be an interface for connecting an audio output device such as a speaker or headphones to the terminal device 10. In this case, the output unit 13 generates an audio signal for playing audio data and outputs the audio signal to the audio output device connected to it.

[0035] The storage unit 14 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 14 stores data used by the control unit 15. The storage unit 14 stores data necessary when the control unit 15 performs processing.

[0036] The control unit 15 is composed of a processor such as a CPU (Central Processing Unit) and memory (main memory). The control unit 15 functions when the processor executes a program. Note that all or part of the functions of the control unit 15 may be implemented using hardware such as an ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), or FPGA (Field Programmable Gate Array). The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor memory devices (e.g., SSDs: Solid State Drives), as well as storage devices such as hard disks and semiconductor memory devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0037] The control unit 15 may, for example, execute an application that has been pre-installed on the terminal device 10. A specific example of such an application is an application provided to the terminal device 10 as a dedicated application for the estimation system 100. Another specific example of such an application is a web browser application. The control unit 15 operates according to the program of the application being executed.

[0038] The control unit 15 controls the terminal device 10 according to user operations and information received from the estimation device 30. For example, the control unit 15 transmits information input by the user through operation of the operation unit 12 to the estimation device 30 using the communication unit 11. For example, when the control unit 15 receives information transmitted from the estimation device 30 via the network 70 to the communication unit 11, it generates screen data based on the received information and displays the screen data on the output unit 13. Such screen data includes images and characters transmitted from the estimation device 30. For example, when the control unit 15 receives information transmitted from the estimation device 30 via the network 70 to the communication unit 11, it generates audio data based on the received information and outputs the audio data from the output unit 13.

[0039] The operation of the control unit 15 will be described below in detail. In the following example, an image display device is used as a specific example of the output unit 13. However, as mentioned above, the output unit 13 does not need to be configured using an image display device; it may be configured using an audio output device, or it may be configured using both an image display device and an audio output device.

[0040] The control unit 15 generates screen data containing characters and images that instruct the user to input condition information. The control unit 15 displays the generated screen data on the output unit 13. The control unit 15 may instruct the input of information regarding silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, and nitrogen as condition information. The control unit 15 may further instruct the input of information regarding the temperature of the operating environment and the chloride ion concentration, in addition to the information regarding silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, and nitrogen as condition information. The control unit 15 may also instruct the input of information regarding the state of the metal structure as condition information. For example, if the metal material is stainless steel, the control unit 15 may instruct the input of information regarding whether it is duplex stainless steel or non-duplex stainless steel. The control unit 15 may also instruct the input of information regarding precipitates and inclusions (for example, the type, size (particle size, etc.), and quantity (number density, etc.) of precipitates and inclusions) as condition information.

[0041] The user inputs condition information into the terminal device 10 by operating the operation unit 12. The control unit 15 transmits the input condition information to the estimation device 30 using the communication unit 11. The control unit 15 receives estimation results from the estimation device 30 according to the condition information. The control unit 15 generates screen data showing information indicating the estimation results. The control unit 15 displays the generated screen data on the output unit 13.

[0042] Figure 3 is a schematic block diagram showing a specific example of the functional configuration of the learning device 20. The learning device 20 is configured using information processing equipment such as a personal computer or a server device. The learning device 20 includes a communication unit 21, a storage unit 22, and a control unit 23.

[0043] The communication unit 21 is a communication device. The communication unit 21 may be configured, for example, as a network interface. The communication unit 21 communicates data with other devices via the network 70 in accordance with the control of the control unit 23. The communication unit 21 may be a wireless communication device or a wired communication device.

[0044] The storage unit 22 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 22 stores data used by the control unit 23. The storage unit 22 may function, for example, as a training data storage unit 221, a pre-processed training data storage unit 222, and a trained model information storage unit 223.

[0045] The training data storage unit 221 stores training data used in the supervised learning process executed in the learning device 20. The training data stored in the training data storage unit 221 includes conditional information (explanatory variables) and characteristic information of the metal material (target variable) corresponding to that conditional information. The conditional information and characteristic information will be explained below.

[0046] Conditional information is information that indicates conditions related to a metallic material. Here, the metallic material is, for example, iron or steel such as stainless steel, but the metallic material is not limited to iron or steel and may be any metal. For example, it may be a nickel alloy. Conditional information may also be, for example, information regarding the chemical composition of the metallic material used to estimate pitting corrosion resistance. Specific examples of information regarding the chemical composition of a metallic material include information on the elements contained in the metallic material being estimated (for example, if the metallic material is stainless steel, information on elements such as silicon (Si), manganese (Mn), nickel (Ni), chromium (Cr), molybdenum (Mo), copper (Cu), titanium (Ti), niobium (Nb), aluminum (Al), nitrogen (N), and tin (Sn) contained therein). Information on only some of these elements may be used. For example, information on the content of multiple elements that affect pitting corrosion resistance contained in the metallic material being estimated may be used. For example, information on the content of chromium and molybdenum may be used. Furthermore, for example, in addition to chromium and molybdenum, information on the content of nickel and copper may also be used. Furthermore, in addition to chromium, molybdenum, nickel, and copper, information regarding the content of silicon, manganese, niobium, titanium, aluminum, and nitrogen may also be used. The unit of content here is mass percent.

[0047] Furthermore, the conditional information may include, in addition to information regarding the chemical composition of the metal material to be estimated, information regarding the assumed usage environment for the metal material whose pitting corrosion resistance is being estimated. Specific examples of information indicating the usage environment include information regarding temperature and chloride ion concentration. For example, in addition to information regarding the chemical composition of the metal material, information regarding temperature and / or chloride ion concentration may be used.

[0048] Furthermore, in addition to information regarding the chemical composition of the metal material (content of alloying components) and information regarding the usage environment, the condition information may also include information regarding the state of the metal structure. Specific examples of information regarding the state of the metal structure include, for example, information indicating whether it is a single-phase structure composed of a single phase or a multi-phase structure composed of multiple phases. If the metal material is stainless steel, information regarding the state of the metal structure may include, for example, information indicating whether it is a single-phase structure or a two-phase mixed structure, or more specifically, information indicating whether it is a duplex stainless steel or a non-duplex stainless steel. Also, for example, if the metal material is stainless steel, information indicating whether it is ferritic stainless steel, martensitic stainless steel, austenitic stainless steel, or austenitic-ferritic duplex stainless steel may be used.

[0049] Furthermore, conditional information may include information on other factors affecting pitting corrosion resistance, such as information on precipitates and inclusions contained in the metal material (type, size, quantity, etc.). Specific examples of precipitate and inclusion types include sulfide inclusions, oxide inclusions, and nitride inclusions. Specific examples of precipitate and inclusion size include average particle size (μm). Specific examples of precipitate and inclusion quantity include number density (pieces / mm²). 2 ) exists.

[0050] Characteristic information refers to information indicating pitting corrosion resistance. A specific example of such characteristic information is the pitting potential (Vpit). More specifically, there is the pitting potential specified in JIS G0577:2014.

[0051] Such training data containing conditional and characteristic information may be obtained, for example, by conducting experiments to measure characteristic information using metal materials corresponding to the conditional information. Specifically, it may be obtained by manufacturing a metal material corresponding to the conditional information as a test material and measuring the characteristic information of the manufactured test material, or by cutting a sample from an actual product material and measuring its characteristic information in place of or in addition to the test material. Furthermore, it is preferable that the training data is created using experimental results related to metal materials having a chemical composition relatively close to that of the metal material whose characteristic information is to be estimated.

[0052] The pre-processed training data storage unit 222 stores pre-processed training data obtained by performing pre-processing on the training data by the pre-processing control unit 232. The trained model information storage unit 223 stores trained model information obtained by performing training processing by the training control unit 233.

[0053] The control unit 23 is composed of a processor such as a CPU and memory. The control unit 23 functions as an information control unit 231, a preprocessing control unit 232, and a learning control unit 233 when the processor executes a program. Note that all or part of the functions of the control unit 23 may be implemented using hardware such as an ASIC, PLD, or FPGA. The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor storage devices (e.g., SSDs), as well as storage devices such as hard disks and semiconductor storage devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0054] The information control unit 231 controls the input and output of information. For example, the information control unit 231 acquires training data from other devices (information processing devices and storage media) and records it in the training data storage unit 221. For example, the information control unit 231 transmits the trained model information stored in the trained model information storage unit 223 to other devices (for example, the estimation device 30).

[0055] The preprocessing control unit 232 generates preprocessed training data by performing predetermined preprocessing on the training data. The preprocessing control unit 232 may also add information to the training data regarding whether or not pitting corrosion can occur, based on characteristic information (information indicating pitting corrosion resistance) included in the training data. For example, if the training data includes pitting potential (Vpit), the preprocessing control unit 232 may add information to the training data regarding whether or not pitting corrosion can occur, based on the value of the pitting potential (Vpit). Specifically, the preprocessing control unit 232 may add information to the training data regarding whether or not pitting corrosion can occur, based on whether or not the pitting potential (Vpit) is above a predetermined threshold. For example, if the metal material to be estimated is stainless steel, it is known that pitting corrosion hardly occurs if the pitting potential (Vpit) is 1.0V or higher. Conversely, it can be said that pitting corrosion can occur in stainless steel with a pitting potential (Vpit) of less than 1.0V. In this case, the predetermined threshold will be set to 1.0V. Furthermore, if no pitting corrosion is observed on the surface of a test piece cut from the stainless steel using an optical microscope (e.g., 100x magnification), it may be evaluated as having almost no pitting corrosion. In addition, a pitting corrosion index may be used as information regarding whether or not pitting corrosion is likely to occur. The pitting corrosion index is an index that, based on its value, can be used to determine whether or not pitting corrosion is likely to occur (in other words, whether or not pitting corrosion is unlikely to occur). For example, if the metal material to be estimated is stainless steel, as mentioned above, if the pitting potential (Vpit) is 1.0V or higher, pitting corrosion will hardly occur. Therefore, 1.0V may be used as a threshold, and a value indicating whether or not the pitting potential (Vpit) is above the threshold (pitting corrosion index) may be added to the training data. More specifically, for example, the pitting corrosion index may be represented as αP, and if the pitting potential (Vpit) is above the threshold (1.0V or higher), αP=1 may be added to the training data, and if the pitting potential (Vpit) is below the threshold (less than 1.0V), αP=0 may be added. In the following explanation, a Pitting Induction Index (αP) value of 0 indicates that the metal material is susceptible to pitting corrosion, while a Pitting Induction Index (αP) value of 1 indicates that the metal material is highly resistant to pitting corrosion. However, the values ​​and criteria for the Pitting Induction Index (αP) are not limited to these values.

[0056] In the following explanation, we describe a process of adding a predetermined value to the pitting potential (Vpit) value. However, in the process of determining the pitting index, it is preferable to use the pitting potential (Vpit) value before the process of adding the predetermined value is performed. If the pitting potential (Vpit) value after the process of adding the predetermined value is used, it is preferable to determine the pitting index by using a new threshold value obtained by adding the added predetermined value to a predetermined threshold.

[0057] The preprocessing control unit 232 may, for example, perform logarithmic transformation on numerical values ​​included in the training data. More specifically, it may perform logarithmic transformation on values ​​included in conditional information (e.g., information indicating the chemical composition of a metal material). Even more specifically, it may perform logarithmic transformation on the content (mass percentage) of multiple (e.g., all) elements included in the conditional information. Furthermore, for example, if the conditional information also includes information indicating the usage environment, it may perform logarithmic transformation on values ​​included in the information indicating the usage environment (e.g., temperature and chloride ion concentration). Furthermore, for example, it may perform logarithmic transformation on values ​​included in characteristic information indicating pitting corrosion resistance (e.g., pitting potential). By performing such preprocessing, even when the content of each element differs significantly (e.g., the number of digits differs by one, several, or around 10 digits), it becomes possible to perform estimation processing with higher accuracy by bringing the numerical values ​​representing the content of each element closer together.

[0058] The preprocessing control unit 232 may, for example, perform a process of adding a predetermined value to the pitting potential (Vpit) before performing logarithmic transformation on the numerical values ​​included in the training data. A predetermined fixed value is used for this predetermined value. Preferably, the predetermined value is one that can convert negative values ​​that the metal material being estimated can generally take as pitting potential (Vpit) into positive values ​​(values ​​greater than 0). For example, if the metal material being estimated is stainless steel, a value of about 0.2 may be used as the predetermined value. By performing such preprocessing, all pitting potential (Vpit) values ​​can be made positive before the logarithmic transformation process, and as a result, the logarithmic transformation process can be executed correctly.

[0059] The preprocessing control unit 232 may perform scaling on the explanatory variables included in the training data. Specific examples of such scaling include standardization and normalization. Specific examples of such standardization include dividing the deviation from the mean of the data by the standard deviation.

[0060] Of the preprocessing steps performed by the preprocessing control unit 232, the steps described so far are referred to as the "first preprocessing step." Next, the "second preprocessing" performed by the preprocessing control unit 232 will be explained. "Second preprocessing" refers to the process of excluding data in which pitting corrosion rarely occurs, or in other words, the process of extracting only data in which pitting corrosion may occur. After performing the first preprocessing, the preprocessing control unit 232 may perform second preprocessing on the training data for which the first preprocessing has been completed (hereinafter referred to as "first preprocessed training data") (hereinafter, training data for which second preprocessing has been completed will be referred to as "second preprocessed training data"). In the second preprocessing of the first preprocessed training data, the preprocessing control unit 232 may exclude training data in which the pitting potential (Vpit) value is above a predetermined threshold, and extract only training data in which the pitting potential (Vpit) value is below a predetermined threshold. More specifically, for example, if the metal material to be estimated is stainless steel, training data in which the pitting potential (Vpit) is 1.0V or higher may be excluded from the first preprocessed training data. In other words, the preprocessing control unit 232 may, in the second preprocessing, select and extract only the training data from the first preprocessed training data in which the pitting potential (Vpit) is less than 1.0V. Also, as mentioned above, if the pitting occurrence index (αP) is set to "0" for training data in which pitting may occur and to "1" for training data in which pitting will hardly occur, then only the training data with a pitting occurrence index (αP) value of 0 may be extracted from the first preprocessed training data.

[0061] When the first preprocessing is completed, the preprocessing control unit 232 records the training data converted by the first preprocessing as first preprocessed training data in the preprocessed training data storage unit 222. When the second preprocessing is completed, the preprocessing control unit 232 records the training data extracted by the second preprocessing as second preprocessed training data in the preprocessed training data storage unit 222.

[0062] The learning control unit 233 executes supervised learning using the first pre-processed training data and the second pre-processed training data stored in the pre-processed training data storage unit 222. Specific examples of such learning processes include, for instance, support vector regression (SVR), or other learning techniques such as neural networks or deep learning. While support vector regression is a complex model with low model interpretability, it has the advantage of enabling both linear and nonlinear regression and achieving high estimation accuracy with less training data than deep learning. The learning control unit 233 records the trained model information obtained through the execution of the learning process in the trained model information storage unit 223. This trained model information obtained by the learning control unit 233 may be transmitted to the estimation device 30 and recorded in the trained model information storage unit 321 of the estimation device 30.

[0063] Specifically, the learning control unit 233 obtains first trained model information by performing a learning process using first pre-processed training data. Furthermore, the learning control unit 233 obtains second trained model information by performing a learning process using second pre-processed training data.

[0064] The first trained model information indicates a trained model that can obtain an estimated value of the pitting corrosion index as an output (dependent variable) by providing conditional information as input (explanatory variables). The estimated value of the pitting corrosion index may be used as a value indicating the possibility that the pitting potential is 1.0V or higher. The second trained model information indicates a trained model that can obtain an estimated result of characteristic information (information indicating pitting corrosion resistance) (for example, an estimated value of the pitting potential) as an output (dependent variable) by providing conditional information as input (explanatory variables).

[0065] Figure 4 shows information regarding specific examples of training data used in the learning device 20. Figure 4 shows specific examples of conditional and characteristic information included in the training data, such as information on the content of multiple elements contained in the target metal material and information on the pitting corrosion resistance of the metal material. In Figure 4, the information on the content of multiple elements is the content (mass percentage) of silicon (Si), manganese (Mn), nickel (Ni), chromium (Cr), molybdenum (Mo), copper (Cu), titanium (Ti), niobium (Nb), aluminum (Al), and nitrogen (N). The information on pitting corrosion resistance is the pitting potential (Vpit), with the unit V (based on a silver-silver chloride reference electrode (SSE)).

[0066] The training data was created by manufacturing metallic materials according to the condition information shown in Figure 4, specifically metallic materials containing the alloying elements shown in Figure 4, and measuring the pitting potential of the manufactured metallic materials. The remainder of the metallic material manufactured to obtain the training data shown in Figure 4 consisted of iron (Fe) and impurities. Impurities are components that are mixed in during the industrial manufacturing of metallic materials due to various factors in the raw materials such as ore and scrap, and the manufacturing process. The metallic material used to obtain the training data may contain trace amounts of elements other than the alloying elements (silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, nitrogen) and iron (Fe) and impurities included in the training data. Specifically, metallic materials containing trace amounts of alloying elements such as tin (Sn) and carbon (C) may be used. The measurement conditions for the pitting potential were, for example, in accordance with JIS G0577:2014 (Method B), and an aqueous NaCl solution was used as the test solution.

[0067] Figure 4 shows the minimum and maximum values ​​of silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, and nitrogen content, as well as pitting potential. The values ​​of the training data used in the learning device 20 may be distributed to satisfy, for example, the statistics shown in Figure 4.

[0068] Figure 5 is a diagram that shows a more specific example of training data used in the learning device 20. In Figure 5, information on stainless steels, including ferritic stainless steel, austenitic stainless steel, and duplex stainless steel, is shown as a specific example of training data. In Figure 5, specific examples of condition information included in the training data are information on the elements contained in the stainless steel and their content, and information on the environment in which the stainless steel is used, such as temperature and chloride ion concentration. In addition, specific examples of characteristic information included in the training data are shown, such as information on pitting potential, which indicates the pitting corrosion resistance of the stainless steel.

[0069] Figure 5 shows more specific information regarding the alloying elements (silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, and nitrogen) and their respective content (in mass percent) and pitting potential (Vpit, in V) (based on a silver-silver chloride reference electrode (SSE)) as shown in Figure 4. In Figure 5, the measurement conditions for the pitting potential are as described above, and the steel material used is 800 μm thick. Figure 5 also shows information regarding the usage environment as a specific example of training data, where "Temp" represents temperature and "Chloride ion concentration" represents chloride ion concentration. The temperature (Temp) shown in Figure 5 is the temperature of the test solution (NaCl aqueous solution) used to measure the pitting potential, and the unit is °C. The chloride ion concentration (Cl) shown in Figure 5 - ) is the concentration of chloride ions dissolved in the test solution (NaCl aqueous solution) used to measure the pitting potential, and its unit is ppm. Figure 5 shows examples of measuring the pitting potential at temperatures of 30°C, 50°C, and 80°C for four types of test solutions with chloride ion concentrations of 20 ppm, 200 ppm, 2000 ppm, and 20000 ppm.

[0070] Figure 6 shows the values ​​of the training data shown in Figure 5 after preprocessing (first preprocessing) that adds a pitting potential index (αP). In Figure 6, data with a pitting potential value of less than 1.0V are assigned a pitting potential index of 0, and data with a pitting potential value of 1.0V or greater are assigned a pitting potential index of 1. The example in Figure 6 is data obtained by performing the first preprocessing on the training data shown in Figure 5.

[0071] Figure 7 shows the values ​​of the training data shown in Figure 6 after a preprocessing step (first preprocessing) has been performed, in which a predetermined value is added to the pitting potential (Vpit) value. In Figure 7, 0.2 is added to the pitting potential (Vpit) as a specific example of the predetermined value, and the value after the addition is shown as Vpit+.

[0072] Figure 8 shows the values ​​of the training data after logarithmic transformation (first preprocessing) has been performed on the training data (excluding the pitting potential index) shown in Figure 7. In the logarithmic transformation, each numerical value included in the training data (excluding the pitting potential index) is subjected to logarithmic transformation (a process to calculate the common logarithm). In this way, by performing one or more first preprocessing steps on the originally obtained training data (Figure 5), the first preprocessed training data shown in Figures 6 to 8 is obtained.

[0073] Figure 9 shows a specific example of second-preprocessed training data after a second preprocessing step has been performed on the training data shown in Figure 8. The example in Figure 9 is data obtained by extracting only the data where the pitting index (αP) is 0 from the first-preprocessed training data shown in Figure 8 (second-preprocessed training data).

[0074] Figure 10 is a flowchart illustrating a specific example of the processing performed by the learning device 20. First, the information control unit 231 acquires training data (step S101). The training data may be input by a user, acquired by communication from another information device, or acquired from a recording medium connected to the learning device 20. The preprocessing control unit 232 performs a predetermined first preprocessing on the training data (step S102). When the first preprocessing is completed, the preprocessing control unit 232 records the training data converted by the first preprocessing as first preprocessed training data in the preprocessed training data storage unit 222. The learning control unit 233 performs a learning process using the first preprocessed training data stored in the preprocessed training data storage unit 222 and records the first trained model information in the trained model information storage unit 223 (step S103). The preprocessing control unit 232 performs a predetermined second preprocessing on the first preprocessed training data (step S104). When the second preprocessing is completed, the preprocessing control unit 232 records the training data extracted by the second preprocessing as second preprocessed training data in the preprocessed training data storage unit 222. The learning control unit 233 executes the learning process using the second preprocessed training data stored in the preprocessed training data storage unit 222 and records the second trained model information in the trained model information storage unit 223 (step S105).

[0075] Figure 11 is a schematic block diagram showing a specific example of the functional configuration of the estimation device 30. The estimation device 30 is configured using an information processing device such as a personal computer or a server device. The estimation device 30 includes a communication unit 31, a storage unit 32, and a control unit 33.

[0076] The communication unit 31 is a communication device. The communication unit 31 may be configured, for example, as a network interface. The communication unit 31 communicates data with other devices via the network 70 in accordance with the control of the control unit 33. The communication unit 31 may be a device that performs wireless communication or a device that performs wired communication.

[0077] The storage unit 32 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 32 stores data used by the control unit 33. The storage unit 32 may also function, for example, as a trained model information storage unit 321.

[0078] The trained model information storage unit 321 stores information on the first trained model and the second trained model, which are generated in advance by a training process. Such a training process may be performed by another device (e.g., a training device 20) or by the device itself (estimation device 30).

[0079] The control unit 33 is configured using a processor such as a CPU and memory. The control unit 33 functions as an information control unit 331, an estimation unit 332, and a preprocessing control unit 333 when the processor executes a program. Note that all or part of the functions of the control unit 33 may be implemented using hardware such as an ASIC, PLD, or FPGA. The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor storage devices (e.g., SSDs), as well as storage devices such as hard disks and semiconductor storage devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0080] The information control unit 331 acquires information from other devices such as the terminal device 10. A specific example of the information acquired is condition information entered into the terminal device 10. The information control unit 331 records the acquired information in a storage device such as memory. The information control unit 331 transmits information to other devices such as the terminal device 10. A specific example of the information transmitted is characteristic information estimated by the estimation unit 332. Such information exchange between the information control unit 331 and other devices may be carried out, for example, by communication via the communication unit 31.

[0081] The estimation unit 332 estimates characteristic information (information indicating pitting corrosion resistance) according to the condition information obtained from the user. This estimation process is performed using the first trained model information and the second trained model information stored in the trained model information storage unit 321. The estimation unit 332 uses the first trained model and the condition information of the metal material to be estimated to perform a first estimation process to estimate information regarding whether or not the metal material can undergo pitting corrosion according to the condition information. In the first estimation process, for example, the pitting corrosion index (αP) may be estimated as information regarding whether or not pitting corrosion can occur. Furthermore, the estimation unit 332 extracts only the estimated targets that are estimated to be capable of undergoing pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to perform a second estimation process to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets. In the second estimation process, for example, the pitting potential (Vpit) may be estimated as information indicating pitting corrosion resistance.

[0082] The preprocessing control unit 333 performs a process (second preprocessing) on ​​the test data to which the results of the first estimation process have been added (hereinafter referred to as "first estimation processed test data"), by excluding data in which pitting corrosion hardly occurs. Next, the details of the processing of the estimation unit 332 and the preprocessing control unit 333 will be explained using a specific example of condition information.

[0083] Figure 12 shows information about the metal material to be estimated (hereinafter referred to as "test data"). As specific examples of condition information included in the test data, Figure 12 shows information about the content of multiple elements contained in the metal material, and information about the environment in which the metal material is used, namely temperature and chloride ion concentration. In Figure 12, as specific examples of information about the content of multiple elements, the content (mass percent) of silicon (Si), manganese (Mn), nickel (Ni), chromium (Cr), molybdenum (Mo), copper (Cu), titanium (Ti), niobium (Nb), aluminum (Al), and nitrogen (N) is shown. Also in Figure 12, as specific examples of information about the environment in which the metal material is used, temperature (°C) and chloride ion concentration (ppm) are shown as "Temp" and "Chloride Ion Concentration," respectively. Naturally, the property information (information indicating pitting corrosion resistance), which is the value to be estimated (dependent variable), is not given in Figure 12.

[0084] Figure 13 shows a specific example of test data (first-estimation processed test data) to which the results of the first estimation process have been added to the test data shown in Figure 12. In the example in Figure 13, the estimation unit 332, as the first estimation process, uses the first-trained model information and the condition information regarding the metal material to be estimated to estimate the pitting corrosion index (αP) under the conditions indicated by each condition information. In Figure 13, the estimated value of the pitting corrosion index (αP) is represented as Predicted αP. In the example of this embodiment, the closer the estimated value of the pitting corrosion index (αP) is to 0, the higher the probability that the pitting potential value is less than 1.0V, and the closer the estimated value of the pitting corrosion index (αP) is to 1, the higher the probability that the pitting potential value is 1.0V or higher. The estimation unit 332 may also add a value to the condition information of the metal material to be estimated that indicates whether the estimated value of the pitting corrosion index (αP) is above a predetermined threshold. In the example shown in Figure 13, condition information where the estimated value of the pitting corrosion index (αP) is above a predetermined threshold is assigned the value "1" as the pitting corrosion index (αP), and condition information where the estimated value of the pitting corrosion index (αP) is below the predetermined threshold is assigned the value "0" as the pitting corrosion index (αP). The predetermined threshold is determined according to the conditions under which pitting corrosion hardly occurs. For example, if the metal material to be estimated is stainless steel, pitting corrosion hardly occurs if the pitting potential is 1.0V or higher. Therefore, in the case of stainless steel, as will be described later, the correlation between the pitting corrosion index and the pitting potential should be determined experimentally in advance, and the threshold of the pitting corrosion index should be set so that the pitting potential is 1.0V or higher. In the example shown in Figure 13, 0.5 is used as the predetermined threshold. In other words, in Figure 13, a value of 1 is assigned to the pitting index (αP) for items where the estimated value of the pitting index (Predicted αP) is 0.5 or higher, and a value of 0 is assigned to items where the estimated value of the pitting index (Predicted αP) is less than 0.5.

[0085] Figure 14 shows a specific example of test data after the second preprocessing has been performed on the first estimated processed test data shown in Figure 13. The preprocessing control unit 333 performs the second preprocessing on the first estimated processed test data. In the second preprocessing of the first estimated processed test data, condition information in which the estimated value of the pitting corrosion index (Predicted αP), which indicates a high probability that the pitting corrosion potential will be 1.0V or higher, is above a predetermined threshold is excluded from the first estimated processed test data. In other words, in the second preprocessing, the preprocessing control unit 333 selects and extracts only the condition information in which the estimated value of the pitting corrosion index (Predicted αP), which indicates a high probability that the pitting corrosion potential will be less than 1.0V, is below a predetermined threshold from the first estimated processed test data. In the example shown in Figure 14, only condition information in which the estimated value of the pitting corrosion index (Predicted αP) is below a predetermined threshold (0.5) and the pitting corrosion index (αP) is assigned a value of 0 is extracted.

[0086] Figure 15 shows a specific example of the results of a second estimation process performed on the test data extracted by the second pretreatment shown in Figure 14 (hereinafter referred to as "second pretreatment test data"). As the second estimation process, the estimation unit 332 uses the condition information contained in the second pretreatment test data and the second trained model information to estimate the characteristic information of the metallic material (information indicating pitting corrosion resistance) under the conditions indicated by each condition information. In Figure 15, the pitting potential (Vpit) is estimated as characteristic information, and its value (in units of V) is shown as Predicted Vpit.

[0087] Figure 16 is a diagram showing a specific example of the results of estimating characteristic information (information indicating pitting corrosion resistance) for the test data shown in Figure 12. Specifically, Figure 16 is a diagram showing a specific example of the results of estimating characteristic information (pitting potential (Vpit)) after performing the first estimation process, the second pre-processing, and the second estimation process on the test data shown in Figure 12. In this embodiment, the estimation unit 332, as a result of performing the first estimation process, does not perform the estimation of the pitting potential value by the second estimation process for test data where the estimated value of the pitting corrosion index is above a predetermined threshold, and provides an estimation result indicating that the value of the pitting potential is sufficiently high to the point that pitting corrosion hardly occurs. For example, in the example shown in Figure 16, the estimated value of the pitting corrosion index (Predicted αP) is 0.5 or higher, and for test data assigned a value of 1 as the pitting corrosion index (αP), 1.0V is assigned as the estimated value of characteristic information (pitting potential). However, this is not limited to this, and a larger value, such as 1.5V or 2.0V, may be assigned as the estimated value of the pitting potential. To clearly distinguish between cases where the estimated pitting potential obtained by performing the second estimation process is around 1.0V and cases where it is not 1.0V, it is better to assign a larger value such as 1.5V or 2.0V instead of 1.0V. Alternatively, a value indicating that the pitting potential is sufficiently high (e.g., "H") may be assigned. In Figure 16, for test data assigned a pitting index (αP) of 1, the estimated pitting potential (Predicted Vpit) of 1.0V is output, while for test data assigned a pitting index (αP) of 0, the estimated pitting potential (Predicted Vpit) obtained in the second estimation process is output.

[0088] Figure 17 is a flowchart illustrating a specific example of the processing performed by the estimation device 30. First, the information control unit 331 acquires condition information regarding the metal material to be estimated (step S201). The condition information may be, for example, information input by a user to the terminal device 10 and acquired via the network 70. The estimation unit 332 performs a first estimation process using the condition information regarding the metal material to be estimated and the first trained model information (step S202). By performing the first estimation process, the estimation unit 332 acquires, for example, an estimated value (Predicted αP) of the pitting corrosion index for each condition information. Next, the preprocessing control unit 333 performs a predetermined second preprocessing on the condition information (step S203). For example, the preprocessing control unit 333 selects only the condition information in which the estimated value (Predicted αP) of the pitting corrosion index is less than a predetermined threshold. The estimation unit 332 performs a second estimation process using the second trained model information on the selected condition information (step S204). The estimation unit 332 obtains characteristic information (information indicating pitting corrosion resistance), such as an estimated value of the pitting potential (Predicted Vpit), as an estimation result by executing a second estimation process. The information control unit 331 outputs the estimation result of the obtained characteristic information (information indicating pitting corrosion resistance), such as an estimated value of the pitting potential (step S205). For example, the information control unit 331 transmits the obtained estimated value of the pitting potential (Predicted Vpit) to the terminal device 10.

[0089] The estimation system 100 of this embodiment, configured in this way, makes it possible to estimate the properties of a metallic material having predetermined components. Therefore, it is possible to reduce the burden that arises in determining the chemical composition of metallic materials. More specifically, it is possible to reduce some or all of the burden that was previously required for actual experiments.

[0090] Furthermore, in the estimation system 100 of this embodiment, the learning process (second learning process) with pitting potential (Vpit) as the target variable is performed using only training data with values ​​smaller than a threshold defined based on a predetermined criterion as the potential value at which pitting erosion is almost completely eliminated. Therefore, compared to the case where the learning process is performed including training data with values ​​greater than or equal to the potential value at which pitting erosion is almost completely eliminated based on a predetermined criterion, it is possible to estimate the pitting potential with higher accuracy.

[0091] [System Details: Second Embodiment] Next, the details of a second embodiment of the estimation system 100 will be described. In the first embodiment, an example was described in which the condition information does not include information about the metal structure of the metallic material, but the condition information may further include information about the metal structure of the metallic material. The second embodiment is an example in which the condition information further includes information about the metal structure of the metallic material. As a specific example of information about the state of the metallic structure, for example, there is information indicating whether it is a single-phase structure composed of a single phase or a multi-phase structure composed of multiple phases. If the metallic material is stainless steel, for example, information indicating whether it is a single-phase structure or a two-phase mixed structure may be used as information about the state of the metallic structure, more specifically, information indicating whether it is a duplex stainless steel or a non-duplex stainless steel. In addition, for example, information indicating whether it is a ferritic stainless steel, atensitic stainless steel, austenitic stainless steel, or austenitic-ferritic duplex stainless steel may be used. When the metallic material to be estimated is stainless steel, as condition information, for example, information to distinguish whether it is a duplex system or not may be expressed as a phase, and "1" is assigned if it is a duplex system and "0" if it is not a duplex system.

[0092] Figure 18 shows a specific example of training data in the second embodiment. In the example of training data shown in Figure 18, conditional information includes information about the state of the metal structure. Here, as a specific example of information about the state of the metal structure, information indicating whether it is a duplex system or not is shown and is represented as phase. In Figure 18, duplex stainless steel (austenitic-ferrite duplex system) is given a phase of 1, and non-duplex stainless steel (e.g., ferritic stainless steel and austenitic stainless steel) is given a phase of 0. That is, a phase value of 0 indicates that it is not a duplex system, and a phase value of 1 indicates that it is a duplex system.

[0093] Information regarding the state of the metal microstructure (e.g., the phase value) may be provided in advance as training data and conditional information. Furthermore, if the state of the metal microstructure (e.g., whether it is a two-phase system or not) can be determined from the information included in the conditional information, such as the chemical composition of the metal material (content of alloying components), the first pretreatment may determine this from the chemical composition of the metal material and add the obtained information regarding the state of the metal microstructure (e.g., the phase value) to the training data.

[0094] In the second embodiment, the learning control unit 233 uses information about the state of the metal structure (e.g., the phase value) as one of the conditional information (explanatory variables), and executes a first learning process and a second learning process, similar to the first embodiment, to obtain first trained model information and second trained model information.

[0095] In the second embodiment, the estimation unit 332 performs estimation processing by using information about the state of the metal structure (e.g., the phase value) as one of the conditional information (explanatory variables) and obtains the estimation result of characteristic information (information indicating pitting corrosion resistance), for example, the estimated value of the pitting potential. In the second embodiment, the learning control unit 233 may perform estimation processing by using the phase value as one of the explanatory variables and obtain the estimated value of the pitting corrosion index.

[0096] In the estimation system 100 of the second embodiment configured in this way, learning and estimation processes are performed using information about the state of the metal structure as one of the explanatory variables. More specifically, for example, when the metal material is stainless steel, information about whether it is a duplex stainless steel or a non-duplex stainless steel may be used as one of the explanatory variables. For example, whether or not it is a duplex system may be represented as an explanatory variable using a value called phase, and learning and estimation processes may be performed using that explanatory variable. In austenitic-ferrite duplex stainless steel, in which austenite and ferrite phases are mixed, the component distribution in each phase is different and does not result in a uniform component distribution like that of a single-layer structure consisting of a single phase, so it is desirable to distinguish whether or not it is a duplex system. For example, when the metal material is stainless steel and includes both duplex and non-duplex systems, it is possible to estimate the pitting potential with higher accuracy by using information about whether or not it is a duplex system or a non-duplex system (e.g., the value of phase) as an explanatory variable.

[0097] Figure 19 shows an example of the relationship between experimentally determined pitting potential (Vpit) and the estimated pitting potential (Vpit) obtained by performing estimation using trained model information that has undergone the first preprocessing steps described above, which include adding 0.2 to the value of the pitting potential (Vpit) and logarithmic transformation. In this trained model, the training process is performed using training data with pitting potential values ​​of 1.0V or higher, without performing the processing using the pitting index (αP) as described above. Furthermore, the training process is performed without using the phase value. In other words, estimation of the pitting index is not performed, and the estimation process is performed without using the phase value. The horizontal axis of Figure 19 shows the experimentally determined pitting potential (V) (based on the silver-silver chloride reference electrode (SSE)). The vertical axis of Figure 19 shows the pitting potential (V) (based on the silver-silver chloride reference electrode (SSE)) obtained by the estimation process. The training data shown in Figures 4 to 9 is shown as white circles, and the black circles (hatched circles) are test data different from the training data described above.

[0098] Thus, when the estimated values ​​obtained without using the pitting potential index (αP) described above are low (specifically, less than 0.5V), the estimation is performed with a reasonable degree of accuracy. Similarly, when the pitting potential is moderate (specifically, between 0.5V and 1.0V), the estimation is performed with a reasonable degree of accuracy, although slightly lower than when the pitting potential is low. On the other hand, when the pitting potential is high (specifically, 1.0V or higher), the estimation accuracy is significantly reduced.

[0099] Figure 20 shows an example of the relationship between experimentally determined pitting potential (Vpit) and estimated pitting potential (Vpit) obtained by learning and estimation processing using the pitting potential index. Specifically, Figure 20 shows a concrete example of the results of estimating pitting potential using the first trained model information and the second trained model information obtained by the learning device 20 of the first embodiment. The estimation accuracy is improved compared to the case in Figure 19. Note that in Figure 20, there are data where the estimated pitting potential (Vpit) obtained by the estimation processing is "1.0V" corresponding to multiple experimentally determined pitting potentials (Vpit). Therefore, in Figure 20, points with a value of "1.0" on the vertical axis are arranged linearly horizontally. This situation occurs because the threshold for the estimated value of αP is set to 0.5, and all cases where the estimated value of αP is 0.5 or higher as a result of the first estimation processing (those with αP=1) have an estimated pitting potential of 1.0V.

[0100] Figure 21 shows an example of the relationship between experimentally determined pitting potential (Vpit) and estimated pitting potential (Vpit) obtained by performing learning and estimation processing using the pitting index and phase. Specifically, Figure 21 shows a concrete example of the results of estimating pitting potential using the first and second trained model information obtained by the learning device 20 of the second embodiment. The estimation accuracy is improved compared to the case in Figure 19. As is clear from Figures 20 and 21, according to the first or second embodiment, it is possible to output pitting resistance data showing a highly reliable pitting potential (Vpit). In other words, it can be seen that characteristic information can be obtained with high accuracy from condition information by using the trained model information obtained by the learning device 20. Thus, the experimental results shown in Figures 20 and 21 show that there is a correlation between condition information and characteristic information. In addition, in Figure 21, for the same reasons as in Figure 20, data where the estimated value of pitting potential (Vpit) is "1.0V" exists corresponding to multiple experimentally determined pitting potentials (Vpit) and is arranged linearly.

[0101] Figure 22 shows an example of the relationship between the estimated pitting index and the measured pitting potential. In Figure 21, the horizontal axis shows the measured pitting potential of stainless steel under certain conditions, and the vertical axis shows the value obtained by estimating the pitting index (αP) for stainless steel under the same conditions using the first trained model information. From Figure 21, it can be seen that when the estimated value of the pitting index is 0.5 or higher, the measured pitting potential is 1.0V or higher for stainless steel under most conditions, with some exceptions. Based on such experiments, it can be seen that the estimation accuracy can be improved by using a value of around 0.5 as a predetermined threshold for the pitting index used by the estimation unit 332. Based on this, 0.5 was adopted as the threshold in the embodiment described above. However, as can be seen from such experimental results, any specific value can be adopted as the threshold if the pitting index used is such that the measured pitting potential is 1.0V or higher under most conditions (e.g., 80% or more, or 90% or more).

[0102] Figure 23 is a graph showing the relationship between chromium (Cr) content and pitting potential. Figure 22 shows the relationship between the chromium (Cr) content in the metal material and the pitting potential estimated using a trained model obtained by the learning device 20 for three types of stainless steel materials with different microstructures (ferritic, austenitic, and duplex). F, A, and D in the graph represent ferritic, austenitic, and duplex, respectively. The components other than chromium in each metal material (elements that affect pitting resistance and their content) are as follows. F: X%Cr-0.1%Ni-0.01%Mo-0.1%Cu-0.3%Si-0.2%Mn-0.1%Ti-0.1%Nb-0.1%Al-0.01%N A: X%Cr-8%Ni-0.01%Mo-0.1%Cu-0.3%Si-0.2%Mn-0.1%Ti-0.1%Nb-0.1%Al-0.01%N D: X%Cr-4%Ni-0.01%Mo-0.1%Cu-0.3%Si-1.5%Mn-0.1%Ti-0.1%Nb-0.1%Al-0.15%N

[0103] Figure 23 shows the trend of changes in pitting potential associated with changes in chromium content in ferritic, austenitic, and duplex stainless steel materials. By using this embodiment, information on such characteristic trends can be obtained at a lower cost and time compared to actually producing metal materials and obtaining experimental results.

[0104] Metal materials whose pitting corrosion resistance is estimated by the estimation device 30 are offered to the market as industrial products. Furthermore, structures formed at least partially from metal materials whose pitting corrosion resistance is estimated by the estimation device 30 are also offered to the market as industrial products.

[0105] Figure 24 is a schematic diagram of an example hardware configuration of an information processing device 90 applied to this embodiment. The information processing device 90 comprises a processor 91, main memory 92, communication interface 93, auxiliary storage device 94, input / output interface 95, and internal bus 96. The processor 91, main memory 92, communication interface 93, auxiliary storage device 94, and input / output interface 95 are connected to each other via the internal bus 96 so as to be able to communicate with each other. The information processing device 90 may be applied to, for example, a learning device 20 and an estimation device 30. In this case, for example, the communication unit 21 and the communication unit 31 may be configured using the communication interface 93. For example, the storage unit 22 and the storage unit 32 may be configured using the auxiliary storage device 94. Also, the control unit 23 and the control unit 33 may be configured using the processor 91 and the main memory 92.

[0106] The estimation system 100 of this embodiment makes it possible to estimate the properties of a metallic material having predetermined components. Therefore, it is possible to reduce the burden that arises in determining the chemical composition of metallic materials. More specifically, it is possible to reduce some or all of the burden that was previously required for actual experiments.

[0107] (modified version) In this embodiment, the terminal device 10 and the estimation device 30 are configured as separate devices, but they may be configured as a single integrated device. Figure 25 shows a modified example of the estimation device 30 configured in this way. The estimation device 30 shown in Figure 25 includes an operation unit 34 and an output unit 35. The operation unit 34 and output unit 35 of the estimation device 30 shown in Figure 25 function similarly to the operation unit 12 and output unit 13 of the terminal device 10, respectively. The control unit 33 operates in response to operations on the operation unit 34 and outputs information using the output unit 35.

[0108] In this embodiment, the learning device 20 and the estimation device 30 are configured as separate devices, but they may be configured as a single integrated device. Figure 26 shows a modified example of the estimation device 30 configured in this way. The storage unit 32 of the estimation device 30 shown in Figure 26 also functions as a teacher data storage unit 322 and a pre-processed teacher data storage unit 323. The control unit 33 of the estimation device 30 shown in Figure 26 also functions as a pre-processing control unit 333 and a learning control unit 334. The teacher data storage unit 322 and the pre-processed teacher data storage unit 323 function similarly to the teacher data storage unit 221 and the pre-processed teacher data storage unit 222 of the learning device 20, respectively. The pre-processing control unit 333 and the learning control unit 334 function similarly to the pre-processing control unit 232 and the learning control unit 233 of the learning device 20, respectively.

[0109] The learning device 20 may be implemented using multiple information processing devices. For example, the learning device 20 may be implemented using a cloud or other device. For example, in the learning device 20, the memory unit 22 and the control unit 23 may be implemented on different information processing devices. For example, the memory unit 22 of the learning device 20 may be distributed and implemented across multiple information processing devices. The estimation device 30 may be implemented using multiple information processing devices. For example, the estimation device 30 may be implemented using a cloud or other device. For example, in the estimation device 30, the memory unit 32 and the control unit 33 may be implemented on different information processing devices. For example, the memory unit 32 of the estimation device 30 may be distributed and implemented across multiple information processing devices.

[0110] While embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention. [Explanation of Symbols]

[0111] 100…Estimation System, 10…Terminal Device, 11…Communication Unit, 12…Operation Unit, 13…Output Unit, 14…Storage Unit, 15…Control Unit, 20…Learning Device, 21…Communication Unit, 22…Storage Unit, 221…Training Data Storage Unit, 222…Preprocessed Training Data Storage Unit, 223…Trained Model Information Storage Unit, 23…Control Unit, 231…Information Control Unit, 232…Preprocessing Control Unit, 233…Learning Control Unit, 30…Estimation Device, 31…Communication Unit, 32…Storage Unit, 321…Trained Model Information Storage Unit, 33…Control Unit, 331…Information Control Unit, 332…Estimation Unit

Claims

1. A first learning process involves performing a learning process using training data that includes condition information, which is information indicating conditions related to a metallic material, and characteristic information indicating the pitting corrosion resistance of the metallic material according to the conditions indicated by the condition information, in order to obtain first trained model information. A second learning process involves extracting only the training data from the aforementioned training data in which the characteristic information is judged to be capable of causing pitting based on predetermined criteria, performing a learning process using the extracted training data, and obtaining second-trained model information. A first estimation process that uses the first trained model information and the condition information of the metal material to be estimated to estimate information regarding whether or not pitting corrosion can occur according to the condition information, As a result of the first estimation process, only the estimated targets that are estimated to be capable of developing pitting corrosion are extracted, and a second estimation process is performed to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets, using the condition information of the extracted estimated targets and the second trained model information. An estimation method for performing this task.

2. The information indicating pitting corrosion resistance is information regarding the pitting potential of the metal material. The estimation method according to claim 1, wherein the second learning process is a learning process that uses only the training data from which the pitting potential value indicates a value that can cause pitting.

3. The aforementioned metal material is stainless steel. Based on the aforementioned predetermined criteria, the value at which pitting corrosion can occur is when the pitting potential is less than 1.0 V. The estimation method according to claim 2.

4. The estimation method according to any one of claims 1 to 3, wherein the condition information includes information on the content of multiple elements that affect the pitting corrosion resistance contained in the metal material to be estimated.

5. The estimation method according to claim 4, wherein the condition information includes information regarding the content of each of the plurality of elements: silicon, manganese, nickel, chromium, molybdenum, copper, titanium, niobium, aluminum, and nitrogen.

6. The estimation method according to claim 5, wherein the condition information includes information regarding the state of the metal structure of the metallic material.

7. The estimation method according to claim 6, wherein the information relating to the state of the metal structure includes information indicating whether it is a single-phase structure or a multi-phase structure.

8. The estimation method according to claim 6, wherein the metallic material is stainless steel, and the information relating to the state of the metallic structure includes information indicating whether it is a duplex stainless steel or a non-duplex stainless steel.

9. The estimation method according to claim 4, wherein the condition information includes information regarding the usage environment of the metal material to be estimated.

10. The estimation method according to claim 9, wherein the information relating to the usage environment includes information relating to temperature and chloride ion concentration.

11. The estimation method according to claim 4, wherein in the second estimation process, the logarithm is taken with respect to the value included in the condition information, and the pitting corrosion resistance of the metal material is estimated using the obtained value.

12. The estimation method according to claim 11, wherein in the second estimation process, the logarithm is taken of the value included in the information indicating pitting corrosion resistance, and the pitting corrosion resistance of the metal material is estimated using the obtained value.

13. A first learning process involves performing a learning process using training data that includes condition information, which is information indicating conditions related to a metallic material, and characteristic information indicating the pitting corrosion resistance of the metallic material according to the conditions indicated by the condition information, in order to obtain first trained model information. A learning method comprising: extracting from the aforementioned training data only the training data in which the characteristic information is judged to be capable of causing pitting based on predetermined criteria; performing a learning process using the extracted training data; and performing a second learning process to obtain second trained model information.

14. A first estimation process that uses first trained model information and condition information of the metal material to be estimated to estimate information regarding whether or not pitting corrosion can occur according to the said condition information, As a result of the first estimation process, only the estimated targets that are estimated to be capable of developing pitting corrosion are extracted, and a second estimation process is performed to estimate information indicating the pitting corrosion resistance of the metal material according to the estimated target's condition information, using the condition information of the extracted estimated targets and the second trained model information. The first trained model information is information obtained by performing a training process using training data that includes condition information, which is information indicating conditions related to the metallic material, and characteristic information, which indicates the pitting corrosion resistance of the metallic material according to the conditions indicated by the condition information. The estimation method provides that the second trained model information is information obtained by performing a training process using training data obtained by extracting only the training data from the training data in which the characteristic information is judged to be capable of causing pitting based on predetermined criteria.

15. A storage unit that stores: first trained model information obtained by performing a learning process using training data that includes condition information, which is information indicating conditions related to a metal material, and characteristic information, which indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information; and second trained model information obtained by extracting only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, and performing a learning process using the extracted training data; A control unit that performs the following: first estimation process, which uses the first trained model information and the condition information of the metal material to be estimated to estimate information regarding whether or not pitting corrosion can occur according to the condition information; and second estimation process, which extracts only the estimated targets that are estimated to be capable of pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets. An estimation device equipped with the following features.

16. A learning device comprising a control unit that performs a first learning process to acquire first trained model information by performing a learning process using training data which includes condition information which is information indicating conditions related to a metal material and characteristic information which indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information, and a second learning process which extracts only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, performs a learning process using the extracted training data, and acquires second trained model information, An estimation system comprising an estimation device having a control unit that performs: a first estimation process that uses the first trained model information and condition information of the metal material to be estimated to estimate to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process that extracts only the estimated targets that are estimated to be capable of pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets.

17. A storage unit that stores: first trained model information obtained by performing a learning process using training data that includes condition information, which is information indicating conditions related to a metal material, and characteristic information, which indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information; and second trained model information obtained by extracting only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, and performing a learning process using the extracted training data; A computer program for causing a computer to function as an estimation device, comprising: a control unit that performs: a first estimation process that uses the first trained model information and condition information of the metal material to be estimated to estimate information on whether or not pitting corrosion can occur according to the condition information; and a second estimation process that extracts only the estimated targets that are estimated to be capable of pitting corrosion as a result of the first estimation process, and uses the condition information of the extracted estimated targets and the second trained model information to estimate information indicating the pitting corrosion resistance of the metal material according to the condition information of the estimated targets.

18. A learning device comprising a control unit that performs a first learning process to acquire first trained model information by performing a learning process using training data which includes condition information which is information indicating conditions related to a metal material and characteristic information which indicates the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information, and a second learning process which extracts only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, performs a learning process using the extracted training data, and acquires second trained model information.

19. A computer program for causing a computer to function as a learning device, comprising a control unit that performs a first learning process to acquire first trained model information by using training data that includes condition information, which is information indicating conditions related to a metal material, and characteristic information indicating the pitting corrosion resistance of the metal material according to the conditions indicated by the condition information, and a second learning process that extracts only the training data from the training data in which the characteristic information is judged to be capable of causing pitting corrosion based on predetermined criteria, performs a learning process using the extracted training data, and acquires second trained model information.

Images