Metal member remaining life diagnosis device and metal member texture determination device

Abstract

To provide a metal member remaining life diagnosis device and a metal member texture determination device capable of obtaining accurate diagnosis results in a short time without relying on the skill level of an operator.SOLUTION: Provided is a metal member remaining life diagnosis device comprising: a texture determination system for inputting a metal texture image of the metal member to be evaluated, determining multiple kinds of metal textures included in the metal texture image on the basis of a texture determination learning model, and quantizing and outputting for each metal texture; and a remaining life diagnosis system for inputting the material data of the metal member to be evaluated, as well as inputting a feature amount from the texture determination system, and diagnosing the remaining life on the basis of a remaining life diagnosis dataset in which the feature amount and the material data are adopted as explanatory variables and the remaining life diagnosis result is adopted as an objective variable.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] An embodiment of the present invention relates to a remaining life assessment device for a metallic member and a structure determination device for a metallic member. [Background technology]

[0002] Low alloy steels such as 2.25Cr-1Mo steel and Cr-Mo-V steel are used for the main steam lead pipes and casings of steam turbines and the piping of heat recovery steam generators (HRSGs). Since these are used at high temperatures and pressures, there is concern about creep and fatigue accompanied by changes in the metal structure, and damage due to the combination of these factors. In severe cases, cracks penetrate the wall thickness and lead to leakage of the fluid flowing inside. To avoid this, the amount of damage to the components is grasped and life assessments are carried out, and methods for this include the A-parameter method, the void area fraction method, and the grain deformation method (see, for example, Patent Document 1). However, since the results of life assessments from these diagnoses vary widely, it is common to combine multiple methods to comprehensively assess life.

[0003] The work process for diagnosis is mirror polishing of the outer surface of the part → etching with chemicals to reveal the metal structure → taking a replica by the SUMP method → ​​observing the structure with a metallurgical microscope or electron microscope after deposition, and analyzing it for each life evaluation method. These tasks take a huge amount of time, and the evaluation accuracy depends heavily on the experience of the worker, so evaluation by an experienced person is essential.

[0004] In order to solve the problem of working hours, image processing such as binarization and machine learning have been used to reduce the labor required for evaluation. Patent Document 2 discloses a method of evaluating remaining life by binarizing a microstructure photograph using image processing to extract only precipitates. It is known that in ferritic heat-resistant steels, as creep damage progresses, precipitates in the structure coagulate and coarsen, widening the spacing between precipitates, and this method evaluates remaining life from the spacing between precipitates based on this known relationship.

[0005] Patent Document 3 discloses a method for evaluating remaining life based on a learning model that has learned the relationship between metal structural morphology and remaining life through machine learning. It is known that structural morphology such as the state of voids, the state of crystal grains, and the properties of precipitates obtained from a structural photograph are each related to the amount of creep damage, and this method uses machine learning to learn structural photographs and evaluation results that have been evaluated for remaining life based on these known relationships, constructing a learning model, and evaluating the remaining life based on this learning model. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2010-223823 A [Patent Document 2] JP 2016-191556 A [Patent Document 3] JP 2023-44835 A Summary of the Invention [Problem to be solved by the invention]

[0007] Conventional life assessment methods such as those described in the above-mentioned Patent Document 1 require many work steps before obtaining the assessment results, and therefore require a large assessment cost and lead time. For this reason, even if there are multiple parts where leakage is a concern, it is not possible to assess the lifespan of all parts due to the cost and lead time. In addition, there is a problem that the reliability of the assessment results is low because the assessment accuracy depends on the skill level of the worker.

[0008] In the life assessment method of the above-mentioned Patent Document 2, the only feature extracted from the structural photograph is the precipitates. Not limited to the above disclosure, image processing such as binarization has difficulty in identifying multiple elements (grain boundaries, precipitates, voids, etc.), so only a single life assessment method can be applied, and there is a problem in that the assessment results are less reliable than those of conventional life assessments using multiple methods.

[0009] In the remaining life assessment method of Patent Document 3 described above, the correlation between the structural photograph and the remaining life assessment is determined by machine learning. Generally, the correlation determined by machine learning is a black box, and the lack of explainability is a problem. This method also has a problem in that it is unclear which element of the structural photograph was used to assess the remaining life, and the assessment results lack explainability. In addition, in order to build a learning model for this assessment method, it is necessary to perform a remaining life assessment from the structural photograph using the conventional A-parameter method or grain deformation method. Therefore, a lot of work is required when building a learning model.

[0010] The present invention has been made to solve the above-mentioned problems, and has an object to provide a remaining life assessment device for a metal member and a structure determination device for a metal member that are capable of obtaining accurate diagnostic results in a short period of time without depending on the skill level of an operator. [Means for solving the problem]

[0011] A remaining life assessment device for a metallic component according to an embodiment of the present invention is characterized in that it comprises: a structure determination system that inputs a metallographic structure image of a metallic component to be evaluated, determines a plurality of types of metallic structures contained in the metallographic structure image based on a structure determination learning model, and outputs the determined results as quantified feature values ​​for each metallic structure; and a remaining life assessment system that inputs material data of the metallic component to be evaluated and inputs the feature values ​​from the structure determination system, and assesses the remaining life based on a remaining life assessment data set in which the feature values ​​and the material data are used as explanatory variables and a remaining life assessment result is used as a target variable.

[0012] The structure determination device for a metallic component according to an embodiment of the present invention is a structure determination device for a metallic component that inputs a metal structure image of a metallic component to be evaluated, determines multiple types of metal structures contained in the metal structure image based on a structure determination learning model, and outputs quantified features for each metal structure, and is characterized in that the structure determination learning model is constructed by supervised machine learning and is capable of determining grain boundaries, voids, and precipitates. Effect of the Invention

[0013] According to the embodiments of the present invention, it is possible to provide a remaining life assessment device for a metal component and a structure determination device for a metal component that are capable of obtaining accurate diagnostic results in a short period of time without relying on the proficiency of an operator. [Brief description of the drawings]

[0014] [Figure 1] FIG. 4 is a flowchart showing a processing procedure for assessing remaining life of a metal member according to the first embodiment. [Diagram 2] 1 is a block diagram showing a schematic configuration of a remaining life assessment device for a metal member according to a first embodiment. [Diagram 3] FIG. 13 is a schematic diagram showing an example of a metal structure image. [Figure 4] FIG. 13 is a diagram showing an example of a remaining life assessment data set. [Diagram 5] FIG. 11 is a flowchart showing a processing procedure for assessing remaining life of a metal member according to the second embodiment. [Figure 6] FIG. 11 is a flowchart showing a processing procedure for assessing remaining life of a metal member according to the third embodiment. [Figure 7] FIG. 13 is a flowchart showing a processing procedure for assessing remaining life of a metal member according to the fourth embodiment. [Figure 8] FIG. 13 is a flowchart showing a processing procedure for determining the structure of a metal member according to the fifth embodiment. [Figure 9] FIG. 13 is a block diagram showing a schematic configuration of a structure assessment device for a metal member according to a fifth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, a remaining life assessment device for a metallic member and a structure determination device for a metallic member according to embodiments will be described with reference to the drawings.

[0016] (First embodiment) First, a first embodiment will be described. Fig. 1 is a flow chart showing a processing procedure for assessing remaining life span of a metal member in the first embodiment, and Fig. 2 is a block diagram showing a configuration of a remaining life span assessment device 100 for a metal member in the first embodiment.

[0017] As shown in Fig. 2, the remaining life assessment device 100 for a metal component according to the first embodiment includes a structure determination system 110 and a life assessment system 120. In order to solve the problem that the accuracy of remaining life assessment varies depending on the experience of the worker, the first embodiment uses a remaining life assessment system 120 that performs machine learning based on past life assessment results. Also, in order to solve the problem that remaining life assessment takes an enormous amount of work time, the structure determination system 110 is used in which feature amount regions of metal structure images can be identified by supervised machine learning.

[0018] The structure determination system 110 includes a metal structure image input unit 111, a metal structure determination unit 112, a structure determination learning model 113, and a feature value output unit 114. In addition, the metal structure determination unit 112 includes a grain boundary identification unit 115, a precipitate identification unit 116, and a void identification unit 117.

[0019] The life assessment system 120 includes a material data input unit 121, a remaining life assessment unit 122, a remaining life assessment data set 123, and a remaining life output unit .

[0020] The structure determination system 110 can individually identify multiple elements of a metal structure image, and the remaining life assessment system 120 performs remaining life assessment based on those elements and material data (including the operating history and composition of the component) from the material data input unit 121. This provides the same reliability as the remaining life assessment results using conventional multiple methods.

[0021] Furthermore, the structure determination system 110 extracts and quantifies regions of structure factors such as crystal grains (grain boundaries), voids, and precipitates from the input metal structure image. By inputting these quantified structure factors and the above-mentioned material data to the remaining life assessment unit 122, the remaining life is output from the remaining life output unit 124. In other words, it is possible to understand which part of the structure morphology is being focused on, thereby obtaining explainability. Furthermore, since the structure determination system 110 can automatically quantify the structure morphology of the metal structure, it is possible to significantly reduce the amount of work required to build a learning model for remaining life assessment.

[0022] This metallic member remaining life assessment device 100 can be used, for example, to assess the remaining life of a steam turbine main steam lead pipe made of 2.25Cr1Mo steel. To obtain a metallographic image, first, the surface of the lead pipe to be assessed is mirror-polished and corroded with a chemical to reveal the metallographic structure. Then, a microscope or the like is used to capture an image of the metallographic structure or a replica image based on the metallographic structure. When this metallographic structure image and material data such as the chemical composition and use history of the material to be assessed are input into the metallic member remaining life assessment device 100, a remaining life assessment result is output.

[0023] As described above, the metallic component remaining life assessment device 100 is composed of a two-stage structure of the structure determination system 110 and the remaining life assessment system 120. First, a metal structure image is captured and input to the structure determination system 110, which outputs the feature quantities of the structure. Next, the feature quantities output by the structure determination system 110 and material data such as the chemical composition and the usage history of the component are input to the remaining life assessment system 120, which outputs the remaining life assessment result.

[0024] (Development procedure for the texture determination system 110) The machine learning of the texture determination system 110 employs a machine learning method that extracts image features and identifies regions. As this machine learning method, for example, the Trainable Weka Segmentation method (TWS method) can be used. The development procedure of the texture determination learning model 113 constructed by the TWS method will be described below.

[0025] When constructing the structure determination learning model 113 using the TWS method, an arbitrary metal structure image is read and the feature amount of the structure is marked. For example, in order to identify the grain boundary, marking is applied at several points along the grain boundary portion on the read metal structure image.

[0026] Other features to be recognized, such as voids and precipitates, are similarly marked in several locations, and supervised learning is started, allowing the features of all regions of the metallographic image to be identified. If the learning results for a region using the TWS method are insufficient or excessive, the markings are corrected as appropriate and additional learning is performed on the learning model. In addition, by loading metallographic images with different shooting conditions, such as brightness and contrast, and performing additional learning in the same manner, a learning model can be constructed that can identify regions in metallographic images under any conditions. Furthermore, by applying an image processing filter, it is possible to prevent the misidentification of polishing marks on the metallographic structure.

[0027] (Method of operating the organization determination system 110) As shown in Fig. 1, first, the metal structure of the component to be diagnosed for its life span is photographed (1). An optical microscope, an electron microscope, etc. are used for the photographing. The photographed metal structure image is input to a metal structure image input unit 111 of a structure determination system 110.

[0028] 3 is a schematic diagram showing an example of a metal structure image, and as shown in Fig. 3, in this metal structure image, crystal grains 30, crystal grain boundaries 31, precipitates 32, and voids 33 can be confirmed. In a metal structure determination unit 112 of a structure determination system 110, a crystal grain boundary identification unit 115, a precipitate identification unit 116, and a void identification unit 117 perform structure determination (10) based on a structure determination learning model 113 machine-learned by supervised learning, and automatically perform grain boundary identification (11), precipitate identification (12), and void identification (13), and output the size, shape, and the like of each from a feature output unit 114 as a feature (14).

[0029] According to the structure determination system 110 of this embodiment, the metal structure can be identified by the structure determination learning model 113 constructed by machine learning, and multiple elements such as grain boundaries, precipitates, and voids can be identified. This makes it possible to significantly reduce the number of work steps, and also to eliminate differences in evaluation accuracy due to the experience of workers (the evaluation can be standardized).

[0030] (Development procedure for remaining life assessment system 120) The remaining life assessment system 120 constructs a learning model by machine learning based on the remaining life assessment data set 123. FIG. 4 is a diagram showing an example of the remaining life assessment data set 123. In the remaining life assessment data set 123, parameters (feature amounts of metal structure) that are indexes in conventional remaining life assessments, such as the A-parameter method and the grain elongation method, and parameters (material data) that are not considered as indexes in conventional remaining life assessments but are thought to affect the remaining life of materials are adopted as explanatory variables. In the example shown in FIG. 4, the feature amounts of the metal structure are the grain inclination distribution, grain size, precipitate size, and void size. In the example shown in FIG. 4, the material data are the chemical structure of the component, the operating time of the component application equipment, the component application location, the initial strength of the component, and the use temperature of the component. However, the contents of the remaining life assessment data set 123 are not limited to these. In the example shown in FIG. 4, the objective variable is the remaining life assessment result, which is output as the remaining life of the component.

[0031] Parameters (feature amounts of metal structure) that are used as indexes in conventional remaining life assessments are automatically output by inputting any metal structure image into the structure determination system 110. Since the relationship between each feature amount of metal structure and remaining life differs for each steel type, it is necessary to prepare a remaining life assessment data set 123 for each steel type. However, in this case, the amount of work required to build a learning model by machine learning can be significantly reduced.

[0032] (Method of operating remaining life assessment system 120) As shown in Fig. 1, when the remaining life assessment system 120 is operated, the structural feature quantities (14) output by the structural determination system 110 and material data are input to the remaining life assessment unit 122 (2). The remaining life assessment unit 122 performs a remaining life assessment (20) using the input data and a remaining life assessment data set 123, and obtains a remaining life output (21) that is the result of the assessment. The remaining life output is output as a continuous value, for example, X hours. For the machine learning of this system, for example, a neural network or multiple regression analysis can be used.

[0033] As described above, according to the remaining life assessment system 120 of this embodiment, the remaining life assessment is performed based on each feature amount of the structure and material data by using machine learning with the remaining life assessment dataset 123, so that the work hours can be significantly reduced while ensuring the reliability of the results by being able to consider multiple remaining life assessment factors.

[0034] Second embodiment Next, a second embodiment will be described. Fig. 5 is a flow chart showing a procedure for assessing remaining life span of a metal member in the second embodiment. Note that the same reference numerals are used for parts corresponding to those in the first embodiment shown in Fig. 1, and duplicated explanations will be omitted.

[0035] 5, in the second embodiment, in a second-stage remaining life assessment system 120, a remaining life output (21) from a remaining life assessment is input as additional learning to a remaining life assessment data set 123 (22). This allows data to be accumulated, and machine learning is further performed, which is different from the first embodiment.

[0036] In the second embodiment, first, feature amounts are extracted from a metal structure image by a structure determination system 110 in the same manner as in the first embodiment (14). Next, a remaining life assessment system 120 assesses the remaining life based on the feature amounts and material data (2) (20). The remaining life assessment dataset 123 can be updated by additionally learning (22) the output of this remaining life assessment (21) and the input of the feature amounts (14) and material data (2). The updated remaining life assessment dataset 123 can be held separately from the existing remaining life assessment dataset 123, and the user can arbitrarily select which remaining life assessment dataset 123 to use.

[0037] Third embodiment Next, a third embodiment will be described. Fig. 6 is a flow chart showing a procedure for assessing the remaining life span of a metal member in the third embodiment. Note that the same reference numerals are used to designate parts corresponding to those in the first embodiment shown in Fig. 1, and duplicated explanations will be omitted.

[0038] In the third embodiment, as shown in Fig. 6, in a first-stage structure determination system 110, after grain boundaries, precipitates, and voids are region-identified from a metal structure image in the same manner as in the first embodiment (11, 12, 13), a user (operator) judges whether or not the region identification result is correct (14a). If correction is required, the user corrects the region identification result (15), and the corrected data is added to the structure determination learning model 113 as additional learning (16). On the other hand, if correction is not required, the result is output as it is as a feature (14).

[0039] As described above, in the third embodiment, if a region identified on a metal structure image in the microstructure determination system 110 needs to be corrected, the user can correct the region, and the microstructure determination learning model 113 can be updated by correcting the region and performing additional learning. The updated microstructure determination learning model 113 can be stored separately from the existing microstructure determination learning model 113, and the user can arbitrarily select which microstructure determination learning model 113 to use. Thereafter, the remaining life is evaluated in the remaining life assessment system 120 from the obtained feature amount and material data, as in the first embodiment.

[0040] (Fourth embodiment) Next, a fourth embodiment will be described. Fig. 7 is a flow chart showing a procedure for assessing remaining life of a metal member in the fourth embodiment. Note that the same reference numerals are used for parts corresponding to those in the first embodiment shown in Fig. 1, and duplicated explanations will be omitted.

[0041] The fourth embodiment differs from the first embodiment in that the user can correct the quantification results after the grain boundaries, precipitates, and voids are area-identified and quantified as feature values ​​in the first-stage structure determination system 110. Also, the fourth embodiment differs from the third embodiment in that the user corrects the quantitative feature values, not the area identification results.

[0042] That is, as shown in Fig. 7, in the fourth embodiment, first, as in the first embodiment, the structure determination system 110 identifies the regions of grain boundaries, precipitates, and voids from a metal structure image (11, 12, 13), and outputs the size, shape, and the like of each as a feature (14). The user can arbitrarily select, correct, or add to the output results (14b). Thereafter, as in the first embodiment, the remaining life is diagnosed in the remaining life assessment system 120 from the obtained feature (14) or the feature changed by the user (14b) and the material data (2).

[0043] According to the fourth embodiment, the feature quantities identified by machine learning in the structure determination system 110 are passed through the user's judgment, so that the remaining life assessment results are not a black box but have explainability and can be made reliable.

[0044] Fifth embodiment Next, a fifth embodiment will be described. Fig. 8 is a flow chart showing a processing procedure in the fifth embodiment, and Fig. 9 is a block diagram showing the configuration of a structure determination apparatus 200 for a metal member in the fifth embodiment. Note that the same reference numerals are used to designate parts corresponding to those in the first embodiment shown in Figs. 1 and 2, and duplicated explanations will be omitted.

[0045] In the fifth embodiment, a quality inspection of a component is performed using only the first-stage structure determination system 110 in the first embodiment. This quality inspection can be used not only for checking whether the remaining life design of the component is appropriate, but also for quality inspection during component manufacturing for uneven heat treatment, non-uniform composition, etc. Furthermore, the target component in the fifth embodiment is not limited to a high-temperature component, and can be applied to metal components other than high-temperature components.

[0046] As shown in FIG. 9, the structure determination apparatus 200 for a metallic member has a similar configuration to that of the structure determination system 110, and also has, at the downstream side of the feature output unit 114, a quality inspection unit 118 that performs the above-mentioned quality inspection by using the feature output from the feature output unit 114.

[0047] In the fifth embodiment, as shown in Fig. 8, first, as in the structure determination system 110 in the first embodiment, grain boundaries, precipitates, and voids are identified by region from a metallographic image by structure determination (10) (11, 12, 13), and feature quantities are output (14). The structure determination (10) is applied to multiple locations of the same component, and the obtained feature quantity parameters of the metallographic structure are compared to confirm the quality uniformity within the component and perform a quality inspection (30). Alternatively, the feature quantity parameters of the metallographic structure obtained as described above are compared with a reference value of the metallographic structure of the component to perform a quality inspection (30).

[0048] According to the fifth embodiment, the metal structure is automatically identified based on the structure determination learning model 113 constructed by machine learning, so that the number of work steps can be significantly reduced and the difference in quality inspection accuracy between workers can be eliminated (the inspection can be standardized).

[0049] Although several embodiments of the present invention have been described above, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0050] 100...remaining life assessment device for metal components, 110...structure determination system, 111...metal structure image input unit, 112...metal structure determination unit, 113...structure determination learning model, 114...feature output unit, 115...grain boundary identification unit, 116...precipitate identification unit, 117...void identification unit, 118...quality inspection unit, 120...remaining life assessment system, 121...material data input unit, 122...remaining life assessment unit, 123...remaining life assessment data set, 124...remaining life output unit.

Claims

1. a structure determination system that receives a metal structure image of a metal component to be evaluated, determines a plurality of types of metal structures contained in the metal structure image based on a structure determination learning model, and outputs a quantified feature value for each metal structure; a remaining life assessment system which inputs material data of the metallic component to be evaluated and the feature quantities from the structure determination system, and diagnoses a remaining life based on a remaining life assessment data set in which the feature quantities and the material data are explanatory variables and a remaining life assessment result is an objective variable; A remaining life assessment device for a metal component, comprising:

2. The remaining life assessment device for a metal member according to claim 1, The structure determination learning model is constructed by supervised machine learning and is capable of determining grain boundaries, voids, and precipitates. A remaining life assessment device for a metal member.

3. The remaining life assessment device for a metal member according to claim 1 or 2, The remaining life assessment system assesses the remaining life based on a learning model by supervised machine learning using the remaining life assessment data set as training data. A remaining life assessment device for a metal member.

4. The remaining life assessment device for a metal member according to claim 3, The remaining life assessment system accumulates the results of the remaining life assessment in the learning model, performs machine learning on the model, and uses the results for the next and subsequent remaining life assessments. A remaining life assessment device for a metal member.

5. The remaining life assessment device for a metal member according to claim 1 or 2, The material data includes at least one of an application location of the metal component to be evaluated, an operating time of an application device, an initial strength, and a usage temperature, and a feature amount of the metal structure of the metal component to be evaluated. A remaining life assessment device for a metal member.

6. 3. The remaining life assessment device for a metal member according to claim 2, The operator can correct the judgment results of the multiple types of metal structures contained in the metal structure image, which are performed based on the structure judgment learning model, and when the judgment results are corrected, the structure judgment learning model is updated by machine learning using the corrected judgment results and the metal structure image as training data. A remaining life assessment device for a metal member.

7. The remaining life assessment device for a metal member according to claim 1 or 2, The operator can modify the feature amount quantified for each metal structure in the structure determination system, and when the feature amount is modified, the modified feature amount is input to the remaining life assessment system. A remaining life assessment device for a metal member.

8. 1. A structure determination device for a metallic component, which inputs a metal structure image of a metallic component to be evaluated, determines a plurality of types of metal structures contained in the metal structure image based on a structure determination learning model, and outputs a quantified feature amount for each metal structure, The structure determination learning model is constructed by supervised machine learning and is capable of determining grain boundaries, voids, and precipitates.

13. A metal structure determination device comprising:

Images