Remaining life prediction device, remaining life prediction method, and program

Abstract

To predict the remaining life of a metal material with high precision after considering a difference in a chemical composition during heating.SOLUTION: An output value of a learning mode is acquired by inputting predetermined time, and material condition data of a current metal material as a standard metal material to a learning model allowed to learn by using a creep rupture strength value and creep rupture time of each of multiple metal materials that have the same standard and are manufactured at different heat, and material condition data for showing conditions of each of the metal materials. An acquired output value is substituted into an estimation formula as load stress, and the estimation formula is adjusted such that creep rupture time calculated by the estimation formula matches predetermined time when predetermined temperature is substituted. By substituting actually used temperature and an actual load stress into the adjusted estimation formula, creep rupture time of a current metal material is calculated. By subtracting actually used time from the creep rupture time of the calculated, current metal material, the remaining life of the current metal material is calculated.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a remaining life prediction device, a remaining life prediction method, and a program. [Background technology]

[0002] Pipes such as heat transfer tubes used in plants and the like are used for long periods of time at high temperatures, and therefore undergo periodic inspections to periodically check their integrity. If inspections could be narrowed down to only the pipes that truly require inspection during periodic inspections, the inspections could be performed more efficiently, thereby shortening the inspection period and reducing the manpower required for inspection. For example, Patent Document 1 discloses a technique for evaluating the remaining lifespan of pipes, and by using such a technique, it is possible to narrow down inspections to pipes with a short remaining lifespan. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-174236 Summary of the Invention [Problem to be solved by the invention]

[0004] As shown in FIG. 9, a single pipe 100 is formed by welding together multiple pipe metal materials 100-1 to 100-10. Each of the pipe metal materials 100-1 to 100-10 has the same specifications but is manufactured from different heats. Here, a heat refers to an individual molten charge that is melted in a melting furnace such as a blast furnace. Because the chemical composition changes slightly during melting, differences in chemical composition occur between metal materials from different heats. Therefore, it is known that even if the pipe metal materials 100-1 to 100-10 have the same specifications, differences in chemical composition can result in significant differences in remaining life. However, the technology disclosed in Patent Document 1 has the problem of not taking into account the differences in heats described above when evaluating remaining life.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a remaining life prediction device, a remaining life prediction method, and a program that can predict the remaining life of a metallic material with high accuracy, taking into account differences in chemical composition between heats. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, the remaining life prediction device according to the present disclosure includes an estimation unit that uses creep rupture strength values ​​and creep rupture times of each of a plurality of metallic materials of the same standard that are manufactured by different heats, and material condition data indicating the conditions of each of the metallic materials, the material condition data including at least the shape and chemical composition of the metallic materials, to obtain an output value output by the learning model by providing as input a predetermined time and material condition data of a metallic material of the standard that is currently in use to a learning model trained with teacher data that takes as input the creep rupture times and the material condition data and outputs the creep rupture strength value corresponding to the input; and a prediction unit that uses a predetermined prediction formula to calculate the creep rupture time of each of the metallic materials of the same standard from the load stress applied to the metallic material and the temperature of the metallic material. a prediction formula adjustment unit that adjusts the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time when the output value acquired by the estimation unit is substituted as the load stress and a predetermined temperature is substituted; a creep rupture time calculation unit that calculates the creep rupture time of the currently used metallic material by substituting an actual use temperature, which is the temperature at which the currently used metallic material is actually used, and an actual load stress actually applied to the currently used metallic material, into the prediction formula adjusted by the prediction formula adjustment unit; and a remaining life calculation unit that calculates the remaining life of the currently used metallic material by subtracting the actual use time, which is the time the currently used metallic material is actually used, from the creep rupture time of the currently used metallic material calculated by the creep rupture time calculation unit.

[0007] The remaining life prediction method according to the present disclosure includes an estimation step of using creep rupture strength values ​​and creep rupture times of a plurality of metallic materials manufactured by different heats and having the same standard, and material condition data indicating the conditions of each of the metallic materials, the material condition data including at least the shape and chemical composition of the metallic materials, to obtain an output value output by a learning model trained with teacher data that takes the creep rupture times and the material condition data as inputs and outputs the creep rupture strength values ​​corresponding to the inputs, by providing as inputs a predetermined time and material condition data of metallic materials of the standard that are currently in use; and a predetermined prediction formula that predicts the creep rupture time of the metallic material from the load stress applied to the metallic material of the standard and the temperature of the metallic material. a prediction formula adjustment step of substituting the output value acquired in the estimation step as the load stress and adjusting the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time when a predetermined temperature is substituted; a creep rupture time calculation step of substituting an actual use temperature, which is the temperature at which the currently used metallic material is actually used, and an actual load stress actually applied to the currently used metallic material into the prediction formula adjusted in the prediction formula adjustment step to calculate the creep rupture time of the currently used metallic material; and a remaining life calculation step of calculating the remaining life of the currently used metallic material by subtracting an actual use time, which is the time the currently used metallic material is actually used, from the creep rupture time of the currently used metallic material calculated in the creep rupture time calculation step.

[0008] The program according to the present disclosure includes a computer that uses creep rupture strength values ​​and creep rupture times of a plurality of metallic materials of the same standard that are manufactured by different heats, and material condition data indicating the conditions of each of the metallic materials, the material condition data including at least the shape and chemical composition of the metallic materials, to obtain an output value output by the learning model by providing as input a predetermined time and material condition data of a metallic material of the standard that is currently in use to a learning model trained with teacher data that takes as input the creep rupture times and the material condition data and outputs the creep rupture strength value corresponding to the input; and a predetermined prediction formula that predicts the creep rupture time of the metallic material from the load stress applied to the metallic material of the standard and the temperature of the metallic material. a prediction formula adjusting means for adjusting the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time when the output value acquired by the estimation means is substituted as the load stress and a predetermined temperature is substituted into the prediction formula that measures the creep rupture time; a creep rupture time calculating means for calculating the creep rupture time of the current metallic material by substituting an actual use temperature, which is the temperature at which the current metallic material is actually used, and an actual load stress actually applied to the current metallic material, into the prediction formula adjusted by the prediction formula adjusting means; and a remaining life calculating means for calculating the remaining life of the current metallic material by subtracting the actual use time, which is the time the current metallic material is actually used, from the creep rupture time of the current metallic material calculated by the creep rupture time calculating means. [Effects of the Invention]

[0009] According to the remaining life prediction device, remaining life prediction method, and program disclosed herein, it is possible to predict the remaining life of a metallic material with high accuracy, taking into account differences in chemical composition between heats. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram illustrating an example configuration of a learning device according to an embodiment of the present disclosure. [Figure 2] FIG. 10 is a diagram illustrating a data format of a creep rupture test table according to an embodiment of the present disclosure. [Figure 3] FIG. 2 is a block diagram illustrating an example of the internal configuration of a function approximator according to an embodiment of the present disclosure. [Figure 4] 10 is a flowchart illustrating an example of the operation of the learning device according to an embodiment of the present disclosure. [Figure 5] 1 is a block diagram illustrating an example configuration of a remaining life prediction device according to an embodiment of the present disclosure. [Figure 6] FIG. 1 illustrates an example of a stress creep rupture diagram according to an embodiment of the present disclosure. [Figure 7] 10 is a flowchart illustrating an example of operation of a remaining life prediction device according to an embodiment of the present disclosure. [Figure 8] FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to an embodiment of the present disclosure. [Figure 9] FIG. 1 is a block diagram showing general piping. DETAILED DESCRIPTION OF THE INVENTION

[0011] (Example of learning device configuration) A learning device, a remaining life prediction device, a remaining life prediction method, and a program according to embodiments of the present disclosure will be described below with reference to FIGS. 1 to 8. FIG. 1 is a block diagram showing an example configuration of a learning device 1 according to an embodiment of the present disclosure. FIG. 2 is a diagram showing a data format of a creep rupture test table 12 stored in a creep rupture test data storage unit 11 included in the learning device 1 according to an embodiment of the present disclosure. FIG. 3 is a block diagram showing an example internal configuration of a function approximator 15 according to an embodiment of the present disclosure. FIG. 4 is a flowchart showing an example operation of the learning device 1 according to an embodiment of the present disclosure. FIG. 5 is a block diagram showing an example configuration of a remaining life prediction device 2 according to an embodiment of the present disclosure. FIG. 6 is a diagram showing an example of a stress-creep rupture diagram according to an embodiment of the present disclosure. FIG. 7 is a flowchart showing an example operation of the remaining life prediction device 2 according to an embodiment of the present disclosure. FIG. 8 is a schematic block diagram showing the configuration of a computer according to an embodiment of the present disclosure. Note that the same components in each drawing are designated by the same reference numerals, and description thereof will be omitted as appropriate.

[0012] FIG. 1 is a block diagram showing an example configuration of a learning device 1 according to an embodiment of the present disclosure. The learning device 1 includes a creep rupture test data storage unit 11 and a learning unit 13 as functional components configured as hardware components or a combination of hardware and software. The creep rupture test data storage unit 11 stores a creep rupture test table 12 in the data format shown in FIG. 2. The creep rupture test table 12 has fields for "Test No.", "Material Condition Data," and "Test Result Data," and stores multiple records in rows, each containing data written in the "Test No.", "Material Condition Data," and "Test Result Data" fields. In other words, each time a creep rupture test is performed, one row of records is added to the creep rupture test table 12. Identification information assigned to each of the multiple records is written in the "Test No." field. FIG. 2 shows an example in which consecutive integer numbers, initialized to "1," are assigned as identification information.

[0013] The "Material Condition Data" field contains data indicating the conditions of the metallic material of the test object in the creep rupture test. The test object in the creep rupture test is a plurality of metallic materials of the same specification manufactured by different heats. In this example, the metallic material is assumed to be a metallic material for piping used as a heat transfer tube. Hereinafter, the "metallic material of the piping" will be abbreviated to "piping."

[0014] The "Material Condition Data" item has sub-items of "Pipe Thickness," "Chemical Composition," "Heat Treatment Conditions," and "Test Temperature." In the "Pipe Thickness" item, a numerical value indicating the thickness of the pipe to be tested is written, for example, in units of [mm]. The "Chemical Composition" item further has sub-items corresponding to each of a predetermined number of elements, such as "C (Carbon)," "Si (Silicon)," and "Mn (Manganese)." In each of the multiple element items, a numerical value indicating the mass of each element that is the chemical composition of the pipe to be tested is written, for example, in units of [mass% (mass percent)].

[0015] The "Heat Treatment Conditions" item further has the sub-items of "Temperature" and "Time." In the "Temperature" item, a numerical value indicating the temperature at which the heat treatment was performed when the test pipe was manufactured is written, for example, in units of [°C], and in the "Time" item, a numerical value indicating the time for which the heat treatment was performed is written, for example, in units of [minutes].

[0016] In a creep rupture test, heat of a constant temperature is applied to the pipe being tested, and the temperature of the pipe is maintained constant during the test. In the "Test temperature" field, a numerical value indicating the temperature of the pipe maintained constant during the creep rupture test is entered, for example, in units of °C.

[0017] In the "Material Condition Data" section, the value entered in the "Pipe Thickness" section is a value obtained by actually measuring the thickness of the pipe being tested. The value entered in the "Chemical Composition" section is a value obtained by taking a sample from the pipe being tested and analyzing its chemical composition. The value entered in the "Heat Treatment Conditions" section is a value obtained from the mill certificate that certifies the quality of the metallic material of the pipe being tested. The value entered in the "Test Temperature" section is a value determined in advance when conducting a creep rupture test on each of the pipes being tested.

[0018] The "Test Result Data" item further has the sub-items of "Creep Rupture Time" and "Creep Rupture Strength." The "Creep Rupture Time" item is entered as a numerical value indicating the creep rupture time obtained as a result of the creep rupture test, for example, in units of hours. The "Creep Rupture Strength" item is entered as a numerical value indicating the creep rupture strength obtained as a result of the creep rupture test, i.e., the creep rupture strength value, for example, in units of MPa (Mega Pascal). Here, creep rupture time refers to the time it takes for a test object, such as a pipe, to increase in strain and rupture when subjected to a constant stress at a constant temperature. Creep rupture strength refers to the constant stress applied to the test object when the creep rupture time is measured.

[0019] The learning unit 13 includes a learning processing unit 14, a function approximator 15, a coefficient storage unit 16, and a coefficient reading unit 17. The learning model 20 shown in the learning unit 13 is the function approximator 15 to which the coefficients stored in the coefficient storage unit 16 have been applied. The learning processing unit 14 generates training data to be used in the learning process from data stored in the creep rupture test table 12 of the creep rupture test data storage unit 11, the training data including input element data and output element data. The learning processing unit 14 generates one training data from one row of record stored in the creep rupture test table 12. The learning processing unit 14 obtains an output value output by the learning model 20 by providing the input element data included in the training data to the learning model 20. The learning processing unit 14 calculates a loss value by substituting the obtained output value and the output element data included in the training data into a predetermined loss function. Here, the loss function is, for example, a loss function that calculates a sum-of-squares error. The learning processing unit 14 calculates new coefficients to be applied to the function approximator 15 so as to reduce the loss value, based on the loss function, the acquired output values, the data of the output elements included in the training data, and the coefficients currently being applied to the function approximator 15, for example, by backpropagation.

[0020] The function approximator 15 has, for example, the internal configuration shown in Fig. 3. The function approximator 15 includes an input layer neuron group 30, a hidden layer neuron group 36, and an output layer neuron 37. The input layer neuron group 30 includes a neuron 31, a chemical composition neuron group 32, a heat treatment condition neuron group 33, a neuron 34, and a neuron 35.

[0021] Neuron 31 acquires numerical data indicating pipe thickness included in the training data and outputs the acquired numerical data indicating pipe thickness to the neuron in the hidden layer neuron group 36 to which it is connected. The chemical composition neuron group 32 has neurons 32-1, 32-2, 32-3, ..., the number of which corresponds to the number of element items indicating chemical composition included in the training data. Each of neurons 32-1, 32-2, 32-3, ... is pre-associated with one of the elements indicating the chemical composition in a one-to-one relationship. For example, as shown in FIG. 3 , neuron 32-1 is pre-associated with C (carbon), and neuron 32-2 is pre-associated with Si (silicon). Each of neurons 32-1, 32-2, 32-3, ... acquires numerical data indicating the mass of the element included in the training data and outputs the acquired numerical data indicating the mass of the element to the neuron in the hidden layer neuron group 36 to which it is connected.

[0022] The heat treatment condition neuron group 33 has a neuron 33-1 corresponding to the temperature of the heat treatment condition and a neuron 33-2 corresponding to the time of the heat treatment condition. Neuron 33-1 acquires numerical data indicating the temperature of the heat treatment condition, which is included in the training data, and outputs the acquired numerical data indicating the temperature to the neuron in the hidden layer neuron group 36 to which it is connected. Neuron 33-2 acquires numerical data indicating the time of the heat treatment condition, which is included in the training data, and outputs the acquired numerical data indicating the time to the neuron in the hidden layer neuron group 36 to which it is connected. Neuron 34 acquires numerical data indicating the test temperature, which is included in the training data, and outputs the acquired numerical data indicating the test temperature to the neuron in the hidden layer neuron group 36 to which it is connected. Neuron 35 acquires numerical data indicating the creep rupture time, which is included in the training data, and outputs the acquired numerical data indicating the creep rupture time to the neuron in the hidden layer neuron group 36 to which it is connected.

[0023] As can be seen from the above description of the input layer neuron group 30, the input element data of the training data generated by the learning processing unit 14 is the numerical data of pipe thickness, chemical composition, heat treatment conditions, test temperature, and creep rupture time given to the input layer neuron group 30. The output element data of the training data is the creep rupture strength value data not included in the input element data.

[0024] The hidden layer neuron group 36 has a plurality of neurons, with the number of layers being one or more. The number of layers of the hidden layer neuron group 36 and the number of neurons per layer are predetermined depending on the scale of the training data to be learned. Each of the multiple neurons included in the hidden layer neuron group 36 outputs an output value calculated based on the output value output by the neuron of the input layer neuron group 30 connected to it in a lower layer or by a neuron of the hidden layer neuron group 36, its coefficient applied to itself, and its activation function to the neuron of the hidden layer neuron group 36 connected to it in an upper layer or to the output layer neuron 37. Here, the upper layer refers to a layer located above the layer in which a certain neuron is located, in a layer configuration in which the output layer neuron 37 is the top layer and the input layer neuron group 30 is the bottom layer. Similarly, the lower layer refers to a layer located below the layer in which a certain neuron is located, in the layer configuration.

[0025] Each output layer neuron 37 outputs an output value calculated based on the output value output by the neurons in the hidden layer neuron group 36 connected to it, the coefficients applied to it, and the activation function. That is, the input layer neuron group 30, the hidden layer neuron group 36, and the output layer neuron 37 constitute a so-called multilayer perceptron forward propagation neural network, and when the hidden layer neuron group 36 has two or more layers, a so-called deep neural network is constituted. The coefficients applied to the neurons in the hidden layer neuron group 36 and the output layer neuron 37 are weights and biases. The activation function used in the neurons in the hidden layer neuron group 36 and the output layer neuron 37 is, for example, a ReLU (Rectified Linear Unit) function.

[0026] The coefficient storage unit 16 stores coefficients written by the learning processing unit 14 to be applied to the neurons of the hidden layer neuron group 36 and the output layer neurons 37. In the initial state, the coefficient storage unit 16 stores initial value coefficients in advance.

[0027] When the learning device 1 is started, the coefficient reading unit 17 reads out a plurality of coefficients stored in the coefficient storage unit 16 and applies each of the read coefficients to the corresponding neurons in the hidden layer neuron group 36 and the output layer neuron 37. When new coefficients are written to the coefficient storage unit 16 by the learning processing unit 14, the coefficient reading unit 17 reads the new coefficients written from the coefficient storage unit 16 and applies each of the read coefficients to the corresponding neurons in the hidden layer neuron group 36 and the output layer neuron 37.

[0028] (Example of learning device operation) FIG. 4 is a flowchart showing the processing flow when the learning device 1 performs a learning process. As described above, the coefficient storage unit 16 stores initial coefficient values ​​in advance. When the learning device 1 starts up, the coefficient reading unit 17 reads the initial coefficient values ​​from the coefficient storage unit 16 and applies each of the read coefficients to the corresponding neurons in the hidden layer neuron group 36 and the output layer neuron 37. This results in the construction of a learning model 20 (SL1). The learning processing unit 14 generates training data from each record stored in the creep rupture test table 12 of the creep rupture test data storage unit 11, and writes and stores the generated training data in an internal storage area (SL2).

[0029] The learning processing unit 14 starts a learning process (L-start to L-end) by repeatedly performing the following steps SL3 to SL6 on each piece of teacher data. The learning processing unit 14 reads one piece of teacher data from its internal storage area and provides each of the input element data of the read teacher data to the corresponding input layer neuron group 30 of the learning model 20. The neurons of the input layer neuron group 30 and the neurons of the hidden layer neuron group 36 perform forward propagation calculations in the forward propagation neural network. The output layer neuron 37 outputs an output value calculated based on the output value output by the neuron of the hidden layer neuron group 36 connected to it in the lower layer, the coefficient applied to itself, and the activation function (SL3).

[0030] The learning processing unit 14 acquires the output value output by the output layer neuron 37 and calculates a loss value by substituting the acquired output value and the creep rupture strength value included as an output element in the training data read in the process of SL3 into a predetermined loss function (SL4). The learning processing unit 14 calculates new coefficients based on the loss function, the acquired output value, the creep rupture strength value, and the coefficients being applied to the function approximator 15, i.e., the coefficients stored in the coefficient storage unit 16, so as to reduce the loss value. The learning processing unit 14 updates the coefficients stored in the coefficient storage unit 16 by rewriting them with the calculated new coefficients (SL5).

[0031] When new coefficients are written to the coefficient storage unit 16 by the learning processing unit 14, the coefficient reading unit 17 reads the new coefficients from the coefficient storage unit 16 and applies them to the neurons in the hidden layer neuron group 36 corresponding to each of the read coefficients and to the output layer neuron 37. In this way, a new learning model 20 is constructed (SL6).

[0032] In the second and subsequent SL3 processes, the learning processing unit 14 reads out from its internal storage area the teacher data that has not been processed up to that point. If there is a sufficient amount of teacher data, the number of repetitions of the loop process L-start to L-end may be set to the number of times equal to the number of teacher data, or the number of repetitions may be set appropriately regardless of the number of teacher data. For example, the loop process L-start to L-end may be repeated until the loss value calculated in the SL4 process becomes equal to or less than a predetermined threshold.

[0033] The process shown in FIG. 4 is a so-called online learning process in which the learning processor 14 calculates new coefficients each time it provides data for one input element of training data to the training model 20. Alternatively, mini-batch learning may be performed in which the learning processor 14 repeatedly performs process SL3 on a predetermined number of training data sets, and then repeatedly performs each of processes SL4 to SL6 once. Alternatively, batch learning may be performed in which the learning processor 14 repeatedly performs process SL3 on all training data sets, and then repeatedly performs each of processes SL4 to SL6 once. In the case of mini-batch learning or batch learning, the number of times the processes SL4 to SL6, i.e., the loss value calculation process and the coefficient update process, are repeated is determined as appropriate, as in the case of online learning. Note that when mini-batch learning or batch learning is performed, multiple combinations of output values ​​output by the output layer neurons 37 and the creep rupture strengths corresponding to the output values ​​are obtained in process SL4. Therefore, a loss function that calculates the squared error of the multiple combinations is applied.

[0034] (Example of remaining life prediction device configuration) 5 is a block diagram showing an example configuration of a remaining life prediction device 2 according to an embodiment of the present disclosure. The remaining life prediction device 2 includes, as functional components configured as hardware components or a combination of hardware and software, a learned coefficient storage unit 16a, a coefficient reading unit 17, an estimation unit 41, a prediction formula adjustment unit 43, a creep rupture time calculation unit 44, and a remaining life calculation unit 45. The learned coefficient storage unit 16a stores in advance the learned coefficients stored in the coefficient storage unit 16 when the learning process by the learning processing unit 14 of the learning device 1 is completed.

[0035] The estimation unit 41 includes a data acquisition unit 42 and a function approximator 15. The learning model 20a shown in the estimation unit 41 is the function approximator 15 to which the learned coefficients stored in the learned coefficient storage unit 16a have been applied. The data acquisition unit 42 acquires data on the current pipe thickness, current chemical composition, current heat treatment conditions, adjustment temperature, and adjustment time provided from the outside, and provides each of the acquired data to a corresponding neuron in the input layer neuron group 30 of the function approximator 15.

[0036] Here, the term "currently used" in the terms "current pipe thickness," "current chemical composition," and "current heat treatment conditions" refers to the piping currently in use in the plant that is the subject of the periodic inspection. In other words, the current pipe thickness refers to the pipe thickness of the piping currently in use. The current pipe thickness data is numerical data expressed in the same units as the "pipe thickness" item in the creep rupture test table 12 of Figure 2. The current chemical composition refers to the chemical composition of the piping currently in use. The current chemical composition data is numerical data indicating the mass of each of the multiple elements specified in the sub-items of the "chemical composition" item in the creep rupture test table 12 of Figure 2, and includes multiple numerical values ​​expressed in the same units as the "chemical composition" item in the creep rupture test table 12.

[0037] The current heat treatment conditions are the heat treatment conditions under which pipes currently in use were manufactured. The data on the current heat treatment conditions includes a numerical value indicating temperature expressed in the same units as the "Temperature" item in the sub-item of the "Heat Treatment Conditions" item in the creep rupture test table 12 in Figure 2, and a numerical value indicating time expressed in the same units as the "Time" item in the sub-item of the "Heat Treatment Conditions" item.

[0038] The data for the adjustment temperature is a predetermined temperature selected in advance from the range of design temperatures that were assumed to be used when the current pipe was designed, and is numerical data expressed in the same units as the "Test Temperature" item in the creep rupture test table 12 in Figure 2. The data for the adjustment temperature is expressed as a fixed numerical value, for example, "600" °C.

[0039] The adjustment time data is a predetermined time, and is numerical data indicating a time expressed in the same units as the "creep rupture time" item in the creep rupture test table 12 of FIG. 2. For example, when the multiple records stored in the creep rupture test table 12 shown in FIG. 2 are classified into creep rupture time categories separated by appropriately determined time intervals, the time indicating the representative value of the creep rupture time category with the largest number of records is predetermined as the adjustment time. The adjustment time data is data expressed as a fixed numerical value, such as "10000" hours.

[0040] The data acquisition unit 42 provides each of the acquired data to the neurons of the input layer neuron group 30 of the function approximator 15, as follows: That is, the data acquisition unit 42 provides numerical data indicating the current pipe thickness to neuron 31. The data acquisition unit 42 provides numerical data indicating the masses of multiple elements included in the current chemical composition data to neurons 32-1, 32-2, 32-3, ..., which are associated with the corresponding elements. The data acquisition unit 42 provides numerical data indicating the temperature included in the current heat treatment condition data to neuron 33-1, and provides numerical data indicating the time included in the current heat treatment condition data to neuron 33-2. The data acquisition unit 42 provides numerical data indicating the adjustment temperature to neuron 34. The data acquisition unit 42 provides numerical data indicating the adjustment time to neuron 35.

[0041] Note that currently used pipes are pipes of the same standard as the pipes tested in the creep rupture test but with a different heat. The currently used pipe thickness values ​​are values ​​obtained by actually measuring the thickness of the currently used pipes. The currently used chemical composition values ​​are values ​​obtained by taking samples from the currently used pipes and analyzing their chemical composition. The currently used heat treatment condition values ​​are values ​​obtained from the mill sheet of the currently used pipes. The values ​​indicating the conditioning temperature and the conditioning time are predetermined, as described above.

[0042] The prediction formula adjustment unit 43 adjusts the prediction formula used to predict a predetermined creep rupture time so that when the output value of the learning model 20a and the adjustment temperature are substituted into the prediction formula, the creep rupture time calculated by the prediction formula matches the adjustment time. Here, the prediction formula is, for example, a second-order LMP (Larson-Miller Parameter) regression formula shown in the following formula (1), which is a so-called life evaluation formula.

[0043]

number

[0044] In the above formula (1), "tr" on the left side is the creep rupture time in hours. "T" on the right side is the temperature of the object whose creep rupture time is to be predicted in K. "σ" on the right side is the load stress applied to the object in MPa. In formula (1), "C," "a0," "a1," and "a2" on the right side are coefficients that are predetermined for each standard of the metal material of the object. In other words, formula (1) is a formula that calculates the creep rupture time of a predicted object based on the temperature of the object and the load stress applied to the object.

[0045] The values ​​of the coefficients "C," "a0," "a1," and "a2" in formula (1) are calculated in advance assuming that prediction objects of the same specification have the same strength, and are not calculated taking into account differences in chemical composition that arise due to differences in heat. Therefore, the prediction formula adjustment unit 43 adjusts the values ​​of the coefficients "C," "a0," "a1," and "a2" in formula (1) so that differences in chemical composition that arise due to differences in heat are taken into account.

[0046] Figure 6 is a graph called a stress creep rupture diagram, with the horizontal axis representing the common logarithm of creep rupture time and the vertical axis representing the common logarithm of applied stress. If "T" in equation (1) is the adjustment temperature, the characteristics of equation (1) can be plotted on the stress creep rupture diagram, for example, as shown by the solid curve 50 in Figure 6.

[0047] The output value of the learning model 20a that the prediction equation adjuster 43 substitutes into equation (1) corresponds to the creep rupture strength value and is obtained by providing the learning model 20a with data on the current pipe thickness, the current chemical composition, the current heat treatment conditions, the adjustment temperature, and the adjustment time. Therefore, the output value of the learning model 20a represents the magnitude of the load stress applied to the current pipe, taking into account the difference in chemical composition resulting from the difference in heat when the creep rupture time coincides with the adjustment time and the temperature of the current pipe coincides with the adjustment temperature. Therefore, the conditions under which the output value of the learning model 20a was obtained are shown in the graph of FIG. 6, where the horizontal axis represents the adjustment time and the vertical axis represents the output value of the learning model 20a, as indicated by reference numeral 61.

[0048] In contrast, the condition under which the value on the left side of equation (1) becomes the adjustment time is the position indicated by reference numeral 51. When taking into account differences in chemical composition resulting from differences in heat, the position indicated by reference numeral 51 needs to become the position indicated by reference numeral 61. Therefore, the prediction formula adjustment unit 43 adjusts the values ​​of the coefficients "C," "a0," "a1," and "a2" so that the curve 50 representing the characteristics of equation (1) shifts parallel to the vertical axis from the position indicated by reference numeral 51 to the position indicated by reference numeral 61, resulting in the dashed curve 60. The new prediction formula obtained through this adjustment takes into account differences in chemical composition resulting from differences in heat.

[0049] The creep rupture time calculation unit 44 calculates the creep rupture time of the currently used pipe by substituting the actual load stress, which is the load stress actually applied to the currently used pipe, and the actual use temperature, which is the temperature at which the currently used pipe is actually used, into the prediction formula after being adjusted by the prediction formula adjustment unit 43. The remaining life calculation unit 45 calculates the time indicating the remaining life of the currently used pipe by subtracting the actual use time, which is the time the currently used pipe has actually been used, from the creep rupture time calculated by the creep rupture time calculation unit 44.

[0050] (Example of remaining life prediction device operation) 7 is a flowchart showing the flow of processing by the remaining life prediction device 2. As described above, the learned coefficient storage unit 16a stores learned coefficients in advance. When the remaining life prediction device 2 is started, the coefficient reading unit 17 reads out the learned coefficients from the learned coefficient storage unit 16a and applies each of the read coefficients to the corresponding neurons in the hidden layer neuron group 36 and the output layer neuron 37. This results in the construction of a learned learning model 20a (SP1).

[0051] The data acquisition unit 42 acquires externally provided data on the current pipe thickness, current chemical composition, current heat treatment conditions, adjustment temperature, and adjustment time, and uses the acquired data as input data (SP2). The data acquisition unit 42 provides each piece of data included in the input data to a corresponding neuron in the input layer neuron group 30 of the trained learning model 20a. The neurons in the input layer neuron group 30 and the neurons in the hidden layer neuron group 36 perform forward propagation calculations in the forward propagation neural network. The output layer neuron 37 outputs an output value calculated based on the output value output by the neuron in the hidden layer neuron group 36 connected to it in the lower layer, the coefficient applied to it, and the activation function (SP3).

[0052] The prediction formula adjustment unit 43 receives the output value output by the trained learning model 20a. The prediction formula adjustment unit 43 receives externally provided data on the adjustment temperature and the adjustment time. The prediction formula adjustment unit 43 substitutes the received output value as the load stress into the prediction formula of Equation (1), and calculates new coefficients "C," "a0," "a1," and "a2" to adjust the prediction formula so that when the numerical value indicated in the unit [°C] of the received data on the adjustment temperature is converted to a numerical value indicated in the unit [K] and then substituted, the creep rupture time calculated by the prediction formula matches the numerical value indicated by the received data on the adjustment time. The prediction formula adjustment unit 43 outputs data indicating the adjusted prediction formula, i.e., data indicating the new coefficients "C," "a0," "a1," and "a2," to the creep rupture time calculation unit 44 (SP4).

[0053] The creep rupture time calculation unit 44 receives data indicating the coefficients "C," "a0," "a1," and "a2" output by the prediction formula adjustment unit 43. The creep rupture time calculation unit 44 receives data on the actual load stress and data on the actual operating temperature, both of which are externally provided. The creep rupture time calculation unit 44 applies the received coefficients "C," "a0," "a1," and "a2" to the prediction formula (1) to generate an adjusted prediction formula. The creep rupture time calculation unit 44 then substitutes the values ​​indicated by the received actual load stress data into the adjusted prediction formula, and further converts the values ​​indicated in units of °C indicated by the received actual operating temperature data into values ​​indicated in units of K, thereby calculating the creep rupture time of the currently used piping. The creep rupture time calculation unit 44 outputs data indicating the calculated creep rupture time of the currently used piping to the remaining life calculation unit 45 (SP5).

[0054] The remaining life calculation unit 45 receives data indicating the creep rupture time of the currently used pipe output by the creep rupture time calculation unit 44 and data on the actual use time given from an external source. Based on the received data, the remaining life calculation unit 45 subtracts the actual use time from the creep rupture time of the currently used pipe to calculate the time indicating the remaining life of the currently used pipe (SP6).

[0055] (Actions and Effects) When a load stress equivalent to that expected under normal operating conditions is applied to a pipe, the creep rupture time becomes extremely long. Since it is not possible to perform many creep rupture tests that result in long creep rupture times, the number of records with long creep rupture times is limited in the creep rupture test table 12. In this case, the number of training data with long creep rupture times generated by the learning processing unit 14 of the learning device 1 will also be small.

[0056] For example, suppose that the learning device 1 is configured to provide a creep rupture strength value as an input element to the function approximator 15 instead of the creep rupture time included in the input element, and perform a learning process so that the output value of the function approximator 15 approaches the creep rupture time corresponding to the input element. In this case, it becomes possible to directly obtain the creep rupture time using only a trained learning model in which trained coefficients are applied to the function approximator 15, without using the prediction formula shown in Equation (1). However, because the learning model constructed in this way has a small amount of training data with long creep rupture times, even if sufficient learning processing is performed, it will only be able to output creep rupture times with low accuracy for load stresses expected in normal use conditions.

[0057] Considering the limited amount of training data with long creep rupture times, the learning device 1 according to the above embodiment includes creep rupture time as an input element to the function approximator 15, and performs a learning process so that the output value of the function approximator 15 approaches the creep rupture strength value corresponding to the input element. Furthermore, the remaining life prediction device 2 provides input data including an adjustment time, which is a time representing the representative value of the creep rupture time category with the largest number of records, to the trained learning model 20a. In other words, the trained learning model 20a utilizes the largest portion of the training data. Therefore, the accuracy of the creep rupture strength value obtained as the output value of the trained learning model 20a is high. The prediction formula adjustment unit 43 uses this highly accurate creep rupture strength value to calculate new coefficients "C," "a0," "a1," and "a2" for the prediction formula (1). Therefore, by using the prediction formula after adjusting it with the new coefficients "C", "a0", "a1", and "a2", it is possible to obtain a highly accurate creep rupture time.

[0058] By using the remaining life prediction device 2 configured in this manner, it is possible to obtain, for example, ten prediction formulas adjusted according to the chemical compositions of the metallic materials 100-1 to 100-10 of the ten pipes shown in FIG. 9 . Using each of the obtained ten prediction formulas, it is possible to calculate the remaining life of each of the metallic materials 100-1 to 100-10 of the pipes with high accuracy, taking into account differences in chemical composition. Based on the calculated remaining life of the metallic materials 100-1 to 100-10 of the pipes, it is possible to identify, for example, metallic materials 100-1 to 100-10 of the pipes that have remaining life at the time of the current periodic inspection but will reach the end of their life by the next periodic inspection. Therefore, it is possible to determine the content and duration of the periodic inspection, such as by inspecting the metallic materials 100-1 to 100-10 of the pipes that will reach the end of their life by the next periodic inspection. In other words, according to the configuration of the above embodiment, it is possible to predict with high accuracy the remaining life of metal materials such as pipes, taking into account differences in chemical composition between heats, and thereby to identify with high accuracy the metal materials to be inspected during periodic inspections.

[0059] (Other embodiments) In the above embodiment, the creep rupture test table 12 stored in the creep rupture test data storage unit 11 of the learning device 1 may have a data format that does not include the heat treatment condition item. In this case, the learning processing unit 14 generates training data that does not include data on the temperature and time of the heat treatment conditions. Furthermore, the input layer neuron group 30 of the function approximator 15 does not include the heat treatment condition neuron group 33. Furthermore, it is not necessary to provide data on the current heat treatment conditions from an external source to the remaining life prediction device 2. Although not including the heat treatment conditions reduces the accuracy of the remaining life calculated by the remaining life prediction device 2, the following becomes possible instead. For example, suppose that a mill sheet for a pipe is unavailable and the heat treatment conditions of the pipe are unknown. Even in such a case, a numerical value indicating the pipe's thickness can be obtained by actually measuring the pipe to be tested in the creep rupture test or the currently used pipe. Furthermore, a numerical value indicating the pipe's chemical composition can be obtained by analyzing the chemical composition of samples obtained from the pipe to be tested in the creep rupture test or the currently used pipe. Therefore, it is possible to predict the remaining life of currently used pipes using only numerical data that can be obtained at the site where the pipes are located.

[0060] In the above embodiment, if the mill sheet of the piping to be tested in the creep rupture test indicates the pipe thickness and chemical composition of the piping, the pipe thickness and chemical composition of the piping indicated in the mill sheet may be stored in the creep rupture test table 12. Also, if the mill sheet of the piping currently in use indicates the pipe thickness and chemical composition of the piping, the pipe thickness of the piping indicated in the mill sheet may be set as the currently in use pipe thickness, and the chemical composition of the piping indicated in the mill sheet may be set as the currently in use chemical composition and provided to the remaining life prediction device 2.

[0061] In the above embodiment, the pipe thickness of the pipe to be tested in the creep rupture test is stored in the creep rupture test table 12, and the pipe thickness of the currently used pipe is provided to the remaining life prediction device 2. However, the pipe thickness is just one example of data indicating the characteristics of the pipe's shape, and any data indicating the characteristics of the pipe's shape may be used. For example, the outer diameter of the pipe may be used instead of the pipe thickness, or a combination of data indicating multiple characteristics of the pipe's shape may be used, such as using both the pipe thickness and the outer diameter.

[0062] In the above embodiment, an example of a metallic material is shown in which piping is used, but in a plant, metallic materials other than piping are used whose soundness is periodically inspected. Therefore, metallic materials other than piping may be used as test objects in a creep rupture test, and the remaining life of the metallic materials may be predicted by the remaining life prediction device 2.

[0063] In the above embodiment, other parameters related to the piping may be added to the material condition data. For example, if post-weld heat treatment (PWHT) has been performed on the piping to be tested in the creep rupture test and the currently used piping, data indicating the PWHT conditions may be added to the material condition data. Furthermore, if the heat treatment conditions differ in cooling methods, such as water cooling, data indicating the cooling method used during the heat treatment may be added to the material condition data. In this case, when generating the training data, the learning processing unit 14 includes data on the added parameters in the input elements of the training data, and the function approximator 15 has neurons in the input layer neuron group 30 that incorporate the added parameter data. Furthermore, the data import unit 42 of the remaining life prediction device 2 is provided with the added parameter data for the currently used piping. In this way, adding other parameters related to the piping to the material condition data can further improve the accuracy of the remaining life calculated by the remaining life prediction device 2.

[0064] In the above embodiment, the creep rupture test table 12 includes a sub-item for "Test Temperature" in the "Material Condition Data" item to store records obtained when creep rupture tests are conducted at various test temperatures. The function approximator 15 also includes a neuron 34 that incorporates numerical data indicating the test temperature, allowing for learning processing that includes the test temperature. In contrast, the remaining life prediction device 2 provides a fixed value, known as the adjustment temperature, to the data input unit 42 and the prediction formula adjustment unit 43. This utilizes only a portion of the predictive ability of the trained learning model 20a to predict creep rupture strength values—in other words, only the adjustment temperature portion. Considering this, the present embodiment can also be configured as follows. That is, the input layer neuron group 30 of the function approximator 15 is configured without including the neuron 34 that incorporates numerical data indicating the test temperature. The learning processor 14 extracts only records in which the test temperature matches the adjustment temperature from the creep rupture test table 12, generates training data from the extracted records that does not include numerical data indicating the test temperature, and performs a training process using the generated training data and a function approximator 15 that does not include a neuron 34. The trained learning model 20a obtained through this training process is a model that predicts creep rupture strength values ​​when the test temperature is fixed at the adjustment temperature. Therefore, in the remaining life prediction device 2, the creep rupture strength value for the adjustment temperature can be obtained as the output value of the trained learning model 20a without providing numerical data indicating the adjustment temperature to the data acquisition unit 42. Furthermore, since the number of data included in the training data can be reduced, the loss value can be more easily converged in the training process in the learning device 1, and the time required for the training process can be shortened. Note that, in this configuration, it is desirable to conduct many creep rupture tests using the adjustment temperature as the test temperature to obtain a sufficient amount of training data.

[0065] In the above embodiment, the adjustment temperature is a predetermined temperature selected in advance from within the range of design temperatures that were assumed to be used when the current pipe was designed. However, the adjustment temperature is not limited to temperatures within the range of design temperatures, and may be, for example, the actual use temperature or the test temperature with the largest number of records stored in the creep rupture test table 12.

[0066] In the above embodiment, the learning processing unit 14 of the learning device 1 generates training data from the data stored in the creep rupture test table 12. The data provided to the data acquisition unit 42 of the remaining life prediction device 2 is also expressed in the same units as the data stored in the creep rupture test table 12. When performing training using a neural network, the data provided as input data to the neural network is typically subjected to preprocessing, such as normalization and standardization, to efficiently converge the loss value. Therefore, even in the above embodiment, the learning processing unit 14 may generate the training data by performing preprocessing, such as normalization and standardization, on the data stored in the creep rupture test table 12. In this case, the data acquisition unit 42 of the remaining life prediction device 2 performs the same preprocessing on each piece of externally provided data as the preprocessing performed by the learning processing unit 14 on the corresponding data.

[0067] In the above embodiment, the activation function applied to the neurons of the hidden layer neuron group 36 and the output layer neuron 37 of the function approximator 15 is, for example, the ReLU function, but an activation function other than the ReLU function may also be applied.

[0068] In the above embodiment, the loss function predetermined in the learning processing unit 14 is, for example, a function that calculates a square sum error, but a loss function other than a function that calculates a square sum error may also be applied.

[0069] In the above embodiment, the function approximator 15 is, for example, a neural network as shown in Fig. 3. However, the function approximator 15 may be realized by a machine learning means other than a neural network.

[0070] The embodiments of the present disclosure have been described in detail above with reference to the drawings, but the specific configurations are not limited to these embodiments and include designs within the scope of the gist of the present disclosure.

[0071] (Computer Configuration) FIG. 9 is a schematic block diagram showing the configuration of a computer that realizes each of the learning device 1 and the remaining life prediction device 2 according to the above-described embodiment. The computer 90 includes a processor 91, a main memory 92, a storage 93, and an interface 94. The creep rupture test data storage unit 11 and the learning unit 13 of the learning device 1, and the learned coefficient storage unit 16a, the coefficient reading unit 17, the estimation unit 41, the prediction formula adjustment unit 43, the creep rupture time calculation unit 44, and the remaining life calculation unit 45 of the remaining life prediction device 2 are implemented in the computer 90. The operations of these functional units are stored in the storage 93 in the form of a program. The processor 91 reads the program from the storage 93 and loads it into the main memory 92, and executes the processes shown in FIGS. 4 and 7 according to the program. The processor 91 also allocates storage areas in the main memory 92 corresponding to the creep rupture test data storage unit 11, the coefficient storage unit 16, and the learned coefficient storage unit 16a, respectively, according to the program.

[0072] The program may be for realizing some of the functions to be performed by the computer 90. For example, the program may be combined with other programs already stored in storage or other programs implemented in other devices to perform the functions. In other embodiments, the computer may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or instead of the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions realized by the processor may be realized by the integrated circuit.

[0073] Examples of storage 93 include a hard disk drive (HDD), a solid state drive (SSD), a magnetic disk, a magneto-optical disk, a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), and a semiconductor memory. Storage 93 may be an internal medium directly connected to the bus of computer 90, or an external medium connected to computer 90 via interface 94 or a communication line. Furthermore, when this program is distributed to computer 90 via a communication line, computer 90 that receives the program may load the program into main memory 92 and execute the above-mentioned processing. Furthermore, storage 93 is a non-transitory tangible storage medium.

[0074] <Additional Notes> The remaining life prediction device 2 described in each embodiment can be understood, for example, as follows.

[0075] (1) A remaining life prediction device 2 according to a first aspect includes an estimation unit 41 that acquires an output value output by a learning model 20a by inputting a predetermined adjustment time and material condition data of a metallic material of the standard that is currently in use to a learning model 20a that has been trained using teacher data that uses creep rupture strength values ​​and creep rupture times of each of a plurality of metallic materials that have the same standard but are manufactured by different heats, and material condition data that indicates the conditions of each of the metallic materials, the creep rupture times and the material condition data including at least the shape and chemical composition of the metallic materials, and that outputs the creep rupture strength value corresponding to the input; and a prediction unit 42 that acquires an output value output by the learning model 20a by inputting a predetermined adjustment time and material condition data of a metallic material of the standard that is currently in use, using a predetermined prediction formula that calculates the creep rupture strength value of the metallic material from a load stress applied to the metallic material of the standard and a temperature of the metallic material. The creep rupture time-prediction system includes a prediction formula adjustment unit that adjusts the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time when the output value acquired by the estimation unit is substituted as the load stress and an adjustment temperature, which is a predetermined temperature, is substituted, a creep rupture time-prediction unit that calculates the creep rupture time of the current metallic material by substituting an actual use temperature, which is the temperature at which the current metallic material is actually used, and an actual load stress actually applied to the current metallic material into the prediction formula adjusted by the prediction formula adjustment unit, and a remaining life calculation unit that calculates the remaining life of the current metallic material by subtracting the actual use time, which is the time the current metallic material is actually used, from the creep rupture time of the current metallic material calculated by the creep rupture time-prediction unit. According to this aspect and each of the following aspects, the remaining life of a metallic material can be predicted with high accuracy while taking into account differences in chemical composition between heats.

[0076] (2) The remaining life prediction device 2 according to the second aspect is the remaining life prediction device 2 of (1), wherein the shape of the metal material is either the outer diameter or the pipe thickness of the pipe manufactured from the metal material, or both.

[0077] (3) A third aspect of the remaining life prediction device 2 is the remaining life prediction device 2 of (1) or (2), wherein the material condition data further includes heat treatment conditions used when the metallic material was manufactured. In this case, the addition of the heat treatment conditions to the material condition data increases the accuracy of the creep rupture strength value predicted by the learning model 20a, thereby increasing the accuracy of the final calculated remaining life of the metallic material.

[0078] (4) A fourth aspect of the remaining life prediction device 2 is the remaining life prediction device 2 of any one of (1) to (3), wherein the material condition data further includes the creep rupture strength value of the metallic material corresponding to the material condition data and the temperature of the metallic material when a creep rupture test was conducted to measure the creep rupture time. The estimation unit 41 acquires an output value from the learning model 20a by providing as input to the learning model 20a a predetermined adjustment time and material condition data for the currently used metallic material, including the predetermined adjustment temperature, instead of the temperature of the metallic material when the creep rupture test was conducted. In this case, an output value corresponding to an arbitrarily determined adjustment temperature can be obtained from the learning model 20a, making it possible to select an adjustment temperature that is easy to adjust the prediction formula from multiple adjustment temperatures.

[0079] (5) A fifth aspect of the remaining life prediction device 2 is the remaining life prediction device 2 of any one of (1) to (3), wherein the learning model 20a is trained using training data that uses the creep rupture time and the material condition data as inputs and the creep rupture strength value corresponding to the inputs, using the creep rupture strength value and the creep rupture time measured in a creep rupture test conducted with the metallic material at a predetermined adjustment temperature, which is a creep rupture strength value and the creep rupture time measured in a creep rupture test conducted with the metallic material at a predetermined adjustment temperature, and the material condition data. In this case, the temperature of the metallic material does not need to be included as a target parameter in the learning process performed when constructing the learning model 20a, which makes it easier for the loss value to converge in the learning process. [Explanation of symbols]

[0080] 1 Learning device 2. Remaining life prediction device 11. Creep rupture test data storage section 12 Creep rupture test table 13 Learning Department 14 Learning processing unit 15 Function Approximators 16,16a Coefficient storage section 17 Coefficient readout unit 20,20a Learning Model 42 Data acquisition section 43 Prediction equation adjustment section 44 Creep rupture time calculation section 45 Remaining life calculation section

Claims

1. an estimation unit that uses creep rupture strength values ​​and creep rupture times of a plurality of metallic materials manufactured by different heats but of the same standard, and material condition data indicating the conditions of each of the metallic materials, the material condition data including at least the shape and chemical composition of the metallic materials, to obtain an output value output by the learning model by providing as input a predetermined time and material condition data of a metallic material of the standard that is currently in use to a learning model trained with training data that takes as input the creep rupture times and the material condition data and outputs the creep rupture strength value corresponding to the input; a prediction formula adjustment unit that, when the output value acquired by the estimation unit is substituted as the load stress into a predetermined prediction formula that predicts the creep rupture time of the metallic material of the standard from the load stress applied to the metallic material and the temperature of the metallic material, and when a predetermined temperature is substituted, adjusts the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time; a creep rupture time calculation unit that calculates the creep rupture time of the currently used metallic material by substituting an actual use temperature, which is the temperature at which the currently used metallic material is actually used, and an actual load stress that is actually applied to the currently used metallic material, into the prediction formula adjusted by the prediction formula adjustment unit; a remaining life calculation unit that calculates the remaining life of the currently used metallic material by subtracting an actual use time, which is the time the currently used metallic material has actually been used, from the creep rupture time of the currently used metallic material calculated by the creep rupture time calculation unit; and A remaining life prediction device comprising:

2. The shape of the metal material refers to either one or both of the outer diameter and the pipe thickness of the pipe manufactured from the metal material. The remaining life prediction device according to claim 1 .

3. The material condition data further includes heat treatment conditions when the metal material is manufactured. The remaining life prediction device according to claim 1 or 2.

4. the material condition data further includes the temperature of the metallic material when a creep rupture test is conducted to measure the creep rupture strength value and the creep rupture time of the metallic material corresponding to the material condition data; The estimation unit The learning model includes: and obtaining an output value output by the learning model by providing as input the predetermined time and material condition data of the currently used metallic material, which includes the predetermined temperature instead of the temperature of the metallic material when the creep rupture test was performed. The remaining life prediction device according to any one of claims 1 to 3.

5. The learning model is using the creep rupture strength value and the creep rupture time measured in a creep rupture test conducted with the temperature of the metallic material at the predetermined temperature and the material condition data, the creep rupture time and the material condition data are input, and learning is performed using training data in which the creep rupture strength value corresponding to the input is output. The remaining life prediction device according to any one of claims 1 to 3.

6. an estimation step of obtaining an output value output by a learning model that has been trained using training data in which the creep rupture time and the material condition data are input and the creep rupture strength value corresponding to the input is output by providing as input a predetermined time and material condition data of a metallic material of the standard that is currently in use, using the creep rupture strength value and creep rupture time of each of a plurality of metallic materials that have the same specification but are manufactured by different heats, and material condition data indicating the conditions of each of the metallic materials, the material condition data including at least the shape and chemical composition of the metallic material; a prediction formula adjustment step of substituting the output value acquired in the estimation step as the load stress into a predetermined prediction formula that predicts the creep rupture time of the metallic material of the standard from the load stress applied to the metallic material and the temperature of the metallic material, and adjusting the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time when a predetermined temperature is also substituted; a creep rupture time calculation step of calculating the creep rupture time of the currently used metallic material by substituting an actual use temperature, which is a temperature at which the currently used metallic material is actually used, and an actual load stress actually applied to the currently used metallic material into the prediction formula adjusted in the prediction formula adjustment step; a remaining life calculation step of calculating a remaining life of the currently used metallic material by subtracting an actual use time, which is the time the currently used metallic material has actually been used, from the creep rupture time of the currently used metallic material calculated in the creep rupture time calculation step; A remaining life prediction method including:

7. Computer, an estimation means for obtaining an output value output by a learning model trained with training data that uses creep rupture strength values ​​and creep rupture times of a plurality of metallic materials manufactured by different heats and having the same specification, and material condition data indicating the conditions of each of the metallic materials, the creep rupture times and the material condition data including at least the shape and chemical composition of the metallic materials, and inputs the creep rupture times and the material condition data of metallic materials of the specification that are currently in use; a prediction formula adjustment means for adjusting a prediction formula that is determined in advance and that predicts the creep rupture time of the metallic material based on the load stress applied to the metallic material of the standard and the temperature of the metallic material, by substituting the output value acquired by the estimation means as the load stress and substituting a predetermined temperature into the prediction formula so that the creep rupture time calculated by the prediction formula coincides with the predetermined time; a creep rupture time calculation means for calculating the creep rupture time of the currently used metallic material by substituting an actual use temperature, which is the temperature at which the currently used metallic material is actually used, and an actual load stress actually applied to the currently used metallic material, into the prediction formula adjusted by the prediction formula adjustment means; a remaining life calculation means for calculating a remaining life of the currently used metallic material by subtracting an actual use time, which is the time the currently used metallic material has actually been used, from the creep rupture time of the currently used metallic material calculated by the creep rupture time calculation means; A program to function as a

Images