WO2020162098A1

WO2020162098A1 - Method for generating metallic material corrosion level prediction model, metallic material corrosion level predicting method, metallic material selecting method, metallic material corrosion level prediction program, and metallic material corrosion level prediction device

Info

Publication number: WO2020162098A1
Application number: PCT/JP2020/000519
Authority: WO
Inventors: 一浩中辻; 真孝面田; 水野　大輔; 山口　収
Original assignee: Ｊｆｅスチール株式会社
Priority date: 2019-02-08
Filing date: 2020-01-09
Publication date: 2020-08-13
Also published as: JP6939995B2; JPWO2020162098A1

Abstract

This method for generating a corrosion level prediction model comprises a learning step for generating a prediction model for predicting the corrosion level of a metallic material in accordance with usage duration, by means of machine learning using data including: a plurality of environmental parameters indicating the environment in which a metallic material is used; the corrosion level of the metallic material; and the usage duration of the metallic material.

Description

Method of generating corrosion amount prediction model of metal material, corrosion amount prediction method of metal material, selection method of metal material, corrosion amount prediction program of metal material, and corrosion amount prediction device of metal material

The present invention relates to a method for generating a metal material corrosion amount prediction model, a metal material corrosion amount prediction method, a metal material selection method, a metal material corrosion amount prediction program, and a metal material corrosion amount prediction device.

Several methods have been proposed in the past for predicting the amount of corrosion of a metal material by using an environmental parameter indicating the usage environment of the metal material. For example, in Non-Patent Document 1 and Patent Documents 1 to 5, an empirical formula is constructed to determine the amount of corrosion. Non-Patent Document 2 predicts the amount of corrosion using multiple regression analysis.

Japanese Patent No. 3909057 Japanese Patent No. 4706254 Japanese Patent No. 5895522 Japanese Patent No. 5066160 Japanese Patent No. 4706279

However, in the methods proposed in Non-Patent Document 1 and Patent Documents 1 to 5, since the amount of corrosion of a metal material is obtained using an empirical formula, for example, when it is desired to respond to changes in the environment or when different metal materials are used. To predict the amount of corrosion, it is necessary to modify the model formula each time. Therefore, there is a problem that it is not realistic in terms of maintenance. Further, in the method proposed in Non-Patent Document 2, there is a problem that a long-term corrosion amount cannot be predicted because a prediction model using the years of use of the metal material is not constructed.

Further, Patent Document 5 proposes a method of predicting a corrosion amount of a steel material after one year and deriving a damping parameter from the corrosion amount to predict a long-term corrosion amount. However, in the method proposed in Patent Document 5, when deriving the attenuation parameter, it is not predicted using the actual results of the corrosion amount for a year or longer, and the prediction model is calculated using an empirical formula and manual adjustment. I'm building. Therefore, problems remain in the utilization of the long-term corrosion amount record and the maintenance as described above. Furthermore, the method proposed in Patent Document 5 does not consider the SO _X concentration that affects corrosion when predicting the amount of corrosion.

SO _X dissolves in the water film on the metal surface, acidifies the water film, and produces sulfite ion HSO ₃ ⁻ . Then, this acidification promotes the anode reaction in which the metal is dissolved. Further, the sulfite ion HSO ₃ ⁻ also reacts with an oxidizing agent such as NO ₂ or O ₃ to become SO ₄ ²⁻ , which promotes corrosion. As described above, the SO _X concentration has a great influence on the corrosion rate, and is therefore important for predicting the amount of corrosion.

The present invention has been made in view of the above, has a high prediction accuracy, and a method of generating a corrosion amount prediction model of a metal material capable of performing long-term corrosion prediction in an atmospheric corrosion environment, corrosion of a metal material An object of the present invention is to provide an amount prediction method, a metal material selection method, a metal material corrosion amount prediction program, and a metal material corrosion amount prediction device.

In order to solve the above problems and achieve the object, a method of generating a corrosion amount prediction model of a metal material according to the present invention is a plurality of environmental parameters indicating a usage environment of the metal material, and a corrosion amount of the metal material. A learning step of generating a prediction model for predicting a corrosion amount of the metal material according to the usage period by machine learning using data including the usage period of the metal material.

Further, in the method of generating a corrosion amount prediction model for a metal material according to the present invention, in the above invention, the learning step generates a first prediction model for predicting a corrosion amount of the metal material for a predetermined period set in advance. The method may include a first learning step and a second learning step of generating a second prediction model that predicts a damping parameter indicating a decay of the corrosion rate of the metallic material for a period exceeding the predetermined period.

Further, the method for generating a corrosion amount prediction model of a metal material according to the present invention, in the above invention, as the machine learning method, a learning method including decision tree regression, random forest, neural network, support vector regression may be used. Good.

In order to solve the above-mentioned problems and achieve the object, the corrosion amount prediction method of the metal material according to the present invention, the prediction amount generated by the generation method of the corrosion amount prediction model of the metal material, the corrosion amount The method includes a prediction step of predicting a corrosion amount of the metal material by inputting a plurality of environmental parameters indicating a usage environment of the metal material to be predicted and a usage period of the metal material.

In order to solve the above-mentioned problems and achieve the object, the corrosion amount prediction method of the metal material according to the present invention is the first prediction model generated by the generation method of the corrosion amount prediction model of the metal material described above, A plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted, and a predetermined predetermined period as an input, a first prediction step of predicting the corrosion amount of the metal material in the predetermined period, and the above By the second prediction model generated by the method of generating a corrosion amount prediction model of the metal material, a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted, and a period exceeding the predetermined period are input. The second prediction step of predicting a damping parameter indicating the decay of the corrosion rate of the metal material in a period exceeding the predetermined period, the corrosion amount of the metal material in the predetermined period, and the damping parameter, A third prediction step of predicting a corrosion amount of the metal material for a period exceeding the predetermined period.

Further, the corrosion amount prediction method of the metal material according to the present invention, in the above invention, the plurality of environmental parameters, temperature, relative humidity, absolute humidity, at least one of the wet time and rainfall, flying salt content, At least one of the SO _X concentration and the NO _X concentration may be included.

In order to solve the above-mentioned problems and to achieve the object, the metal material selection method according to the present invention uses the above-described metal material corrosion amount prediction method to select a metal material according to the usage environment.

In order to solve the above problems and achieve the object, the corrosion amount prediction program of the metal material according to the present invention, a computer, by the prediction model generated by the generation method of the corrosion amount prediction model of the metal material, By inputting a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted and the usage period of the metal material, the function is performed as a prediction means for predicting the corrosion amount of the metal material.

In order to solve the above-mentioned problems and to achieve the object, a metal material corrosion amount prediction program according to the present invention causes a computer to perform a first prediction generated by the method for generating a metal material corrosion amount prediction model described above. By the model, a plurality of environmental parameters indicating the use environment of the metal material to predict the amount of corrosion, and a predetermined predetermined period as an input, the first prediction means for predicting the corrosion amount of the metal material of the predetermined period, The second prediction model generated by the method of generating a corrosion amount prediction model of the metal material, a plurality of environmental parameters indicating the usage environment of the metal material to predict the corrosion amount, and a period exceeding the predetermined period. As an input, a second predicting means for predicting a damping parameter indicating the decay of the corrosion rate of the metal material in a period exceeding the predetermined period, a corrosion amount of the metal material in the predetermined period, and based on the decay parameter. , Third prediction means for predicting the amount of corrosion of the metal material in a period exceeding the predetermined period.

In order to solve the above-mentioned problems and achieve the object, a corrosion amount prediction device for a metal material according to the present invention, a plurality of environmental parameters indicating the usage environment of the metal material, the corrosion amount of the metal material, the metal Using the data including the usage period of the material, by machine learning, learning means for generating a prediction model for predicting the corrosion amount of the metal material according to the usage period, and by the prediction model, to predict the corrosion amount A plurality of environmental parameters indicating a usage environment of the metal material and a usage period of the metal material are input, and a prediction unit that predicts a corrosion amount of the metal material is provided.

Further, in the above-described invention, the apparatus for predicting the amount of corrosion of a metallic material according to the present invention is characterized in that the learning means generates a first prediction model for predicting an amount of corrosion of the metallic material for a predetermined period set in advance. Learning means, and a second learning means for generating a second prediction model for predicting a damping parameter indicating the decay of the corrosion rate of the metal material for a period exceeding the predetermined period, the prediction means, With the first prediction model, a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted, and a predetermined predetermined period are input, and the corrosion amount of the metal material during the predetermined period is predicted. One prediction means, by the second prediction model, a plurality of environmental parameters indicating the use environment of the metal material to predict the amount of corrosion, and a period exceeding the predetermined period as an input, a period exceeding the predetermined period Second prediction means for predicting a damping parameter indicating the decay of the corrosion rate of the metallic material, the amount of corrosion of the metallic material in the predetermined period, and the damping parameter, based on the decay parameter, The third prediction means for predicting the corrosion amount of the metal material may be provided.

According to the present invention, it is possible to accurately predict long-term corrosion of a metallic material in an atmospheric corrosive environment, and it is possible to select an optimal metallic material according to the usage environment.

FIG. 1 is a block diagram showing a schematic configuration of a metal material corrosion amount prediction device according to an embodiment of the present invention. FIG. 2 is a flowchart showing a flow of a method for generating a corrosion amount prediction model for metallic materials according to the first embodiment of the present invention. FIG. 3 is a flowchart showing a flow of a method for predicting a corrosion amount of a metal material according to the first embodiment of the present invention. FIG. 4 is an example of the method for predicting the amount of corrosion of a metallic material according to the first embodiment of the present invention, and is a graph showing an error between the predicted value and the actual value of the amount of corrosion of the metallic material. FIG. 5 is a flowchart showing the flow of generation of a corrosion amount prediction model in the method for generating a corrosion amount prediction model of a metal material according to the second embodiment of the present invention. FIG. 6 is a flowchart showing a flow of generation of a damping parameter prediction model in the method of generating a corrosion amount prediction model of a metal material according to the second embodiment of the present invention. FIG. 7: is a flowchart which shows the flow of the corrosion amount prediction method of the metallic material which concerns on Embodiment 2 of this invention. FIG. 8 is an example of the method for predicting the amount of corrosion of a metal material according to the second embodiment of the present invention, and is a graph showing the error between the predicted value and the actual value of the amount of corrosion of the metal material.

Drawing about the generation method of the corrosion amount prediction model of the metal material, the corrosion amount prediction method of the metal material, the selection method of the metal material, the corrosion amount prediction program of the metal material, and the corrosion amount prediction device of the metal material according to the embodiment of the present invention Will be described with reference to. Hereinafter, a corrosion amount prediction device, a corrosion amount prediction model generation method, and a corrosion amount prediction method will be described for each embodiment. The present invention is not limited to the embodiments described below.

[Embodiment 1]
(Corrosion amount prediction device)
The configuration of the metal material corrosion amount prediction device according to the first embodiment of the present invention will be described with reference to FIG. The corrosion amount prediction device 1 includes an input unit 10, an actual result database (actual result DB) 20, a calculation unit 30, and a display unit 40.

The input unit 10 is an input means for the arithmetic unit 30 and is realized by an input device such as a keyboard, a mouse pointer, a ten-key pad, or the like.

The actual result database 20 stores actual result data regarding the amount of corrosion of metal materials (for example, steel materials) for each steel type. The actual data regarding the amount of corrosion includes a period of use of the metal material, the amount of corrosion of the metal material during the period of use, and a plurality of environmental parameters (for example, annual average) indicating the environment of use of the metal material. In addition, examples of the above-mentioned "plurality of environmental parameters" include temperature (temperature), relative humidity, absolute humidity, wetting time, rainfall amount, flying salt content, SO _X concentration and NO _X concentration. The actual data of these environmental parameters is, for example, annual average data.

The arithmetic unit 30 is realized by, for example, a processor including a CPU (Central Processing Unit) and the like, and a memory (main storage unit) including a RAM (Random Access Memory) and a ROM (Read Only Memory).

The arithmetic unit 30 realizes a function that matches a predetermined purpose by loading and executing a program in the work area of the main storage unit and controlling each component etc. through the execution of the program. The arithmetic unit 30 functions as a learning unit (learning unit) 31 and a corrosion amount prediction unit (prediction unit) 32 through the execution of the program. In the present embodiment, the functions of the learning unit 31 and the corrosion amount prediction unit 32 are realized by one calculation unit (≈computer), but the learning unit 31 and the corrosion amount prediction are performed by two calculation units (≈computer). You may implement|achieve the function of the part 32, respectively.

The learning unit 31 uses actual data including a plurality of environmental parameters, the corrosion amount of the metal material, and the usage period of the metal material, and uses a machine learning to predict the corrosion amount of the metal material according to the usage period by using the corrosion amount. Generate a predictive model. A specific method of generating the corrosion amount prediction model in the learning unit 31 will be described later (see "Method of generating corrosion amount prediction model" described later).

The corrosion amount prediction unit 32 predicts the corrosion amount of the metal material by using the corrosion amount prediction model generated by the learning unit 31 as an input with the environmental parameter of the metal material whose corrosion amount is to be predicted and the usage period of the metal material. .. A specific method of predicting the corrosion amount in the corrosion amount prediction unit 32 will be described later (see "Corrosion amount prediction method" described later).

The display unit 40 is realized by a display device such as an LCD display or a CRT display, and displays the prediction result of the corrosion amount of the metal material, for example, based on the display signal input from the calculation unit 30.

(Method of generating corrosion prediction model)
A method for generating a corrosion amount prediction model for a metal material according to the first embodiment of the present invention will be described with reference to FIG. The method of generating the corrosion amount prediction model is implemented mainly by the learning unit 31 of the calculation unit 30. It should be noted that the generation of the corrosion amount prediction model is performed in advance before the corrosion amount prediction described below is performed.

First, the environmental parameter and the usage period of the metal material are selected (step S11). In this step, the environment used as an explanatory variable when generating a corrosion amount prediction model from among temperature (temperature), relative humidity, absolute humidity, wetting time, rainfall, flying salt content, SO _X concentration and NO _X concentration Select parameters.

In step S11, all the environmental parameters may be selected and a corrosion amount prediction model considering all the environmental parameters may be generated. However, among the environmental parameters described above, some environmental parameters exhibit multicollinearity, such as the relationship between the amount of incoming salt and the SO _X concentration. Therefore, in step S13 in the subsequent stage, when a corrosion amount prediction model is generated using a regression model that cannot eliminate the problem of multicollinearity between environmental parameters, in this step, the correlation between environmental parameters is checked in advance and a strong correlation is established. It is preferable to select the environmental parameters so that the environmental parameters that they have are not included.

Subsequently, the learning unit 31 acquires the corrosion amount of the metal material in the use period selected in step S11 and its environmental parameter from the performance database (step S12).

Subsequently, the learning unit 31 generates a corrosion amount prediction model by a regression model that can be fitted to the non-linearity of data (specifically, environmental parameter) (step S13), and ends this flow. Since the environmental parameter has non-linearity, the prediction accuracy cannot be improved by the multiple regression analysis used in Non-Patent Document 2, for example. Therefore, in this step, a corrosion amount prediction model that predicts the corrosion amount of the metal material according to the period of use is generated by machine learning using a regression model capable of footing to non-linearity. That is, in this step, a corrosion amount prediction model is generated by learning the regression model by inputting the usage period of the metal material and the environmental parameter in the usage period and using the corrosion amount of the metal material as the output in the usage period. ..

Here, examples of regression models (machine learning methods) that can be fitted to non-linearity include decision tree regression, random forest, neural network, support vector regression, and gradient boosting.

(Corrosion amount prediction method)
A method of predicting the amount of corrosion of a metal material according to the first embodiment of the present invention will be described with reference to FIG. The corrosion amount prediction method is implemented mainly by the corrosion amount prediction unit 32 of the calculation unit 30. In addition, in the corrosion amount prediction method, the corrosion amount of the metal material in an arbitrary period of use is predicted by using the corrosion amount prediction model generated by the above-described corrosion amount prediction model generation method.

Specifically, the corrosion amount prediction unit 32 calculates a predicted value of the corrosion amount in the environmental parameter to be predicted from the corrosion amount prediction model generated by the above-described method of generating the corrosion amount prediction model (step S21), and End the flow. Then, the corrosion amount prediction unit 32 displays the corrosion amount prediction result on the display unit 40.

In step S21, the corrosion amount prediction model is used to predict the corrosion amount of the metal material by inputting a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted and the usage period of the metal material. That is, in this step, by inputting the period of use of the metal material and the environmental parameter in the period of use to the corrosion amount prediction model, the amount of corrosion of the metal material in the period of use and the environmental parameter is obtained as an output. Further, in this step, the corrosion amount in the arbitrary usage period can be obtained by inputting the arbitrary usage period into the corrosion amount prediction model.

(Example)
An example of the method of predicting the amount of corrosion of a metal material according to the first embodiment of the present invention will be described with reference to FIG. This figure is a graph showing the error between the predicted value and the actual value of the corrosion amount of the metal material, obtained using the corrosion amount prediction model used for the bridge. (A) of the same figure is a comparative example, and shows the result of predicting the corrosion amount by using this corrosion amount prediction model to generate a corrosion amount prediction model using the multiple regression analysis in the learning step described above. Further, (b) to (d) of the same figure are examples of the present invention. In the learning step described above, a corrosion amount prediction model is generated using decision tree regression, support vector regression, and random forest, and these corrosion amounts are calculated. The result of having predicted the amount of corrosion by a prediction model is shown.

Here, in the decision tree regression described above, the objective variable is predicted using a rule expressed in a tree format based on the value of the explanatory variable. Examples of this rule include “corrosion amount y when the temperature is x° C. or higher, and corrosion amount z when the temperature is lower than x° C.”. Generally, the amount of corrosion of a metal material increases as the temperature rises, but when the temperature rises to some extent, the amount of corrosion decreases. Therefore, by using such a rule for prediction, it is possible to improve the prediction accuracy as compared with linear regression such as multiple regression. In addition, the decision tree regression has a feature that the prediction process is very easy to understand because the prediction is performed using rules. In the verification (see FIG. 4B) in this embodiment, the depth of the decision tree is “6”.

In support vector regression, non-linear transformation of explanatory variables is performed by the kernel to create new variables, and a prediction model is constructed using the new variables. Support vector regression is characterized in that a noise-resistant model can be created by ignoring errors smaller than a predetermined value when constructing a prediction model. The amount of corrosion of a metallic material has a factor that cannot be explained only by the explanatory variables, and the environmental parameters of the same explanatory variable have slightly different amounts of corrosion. Therefore, it is possible to improve the prediction accuracy by applying the support vector regression, which is a noise-resistant regression model that allows such an error, in advance. In the verification in this embodiment (see (c) of FIG. 4), a Gaussian kernel is used for the nonlinear conversion.

　Random forest creates multiple decision trees by randomly omitting samples. Then, prediction is performed by taking the average of the prediction results of each decision tree. By combining a plurality of decision trees in this way, it is more resistant to data noise than the decision tree alone, and the prediction accuracy is improved. In the verification in this embodiment (see (d) of FIG. 4), 500 decision trees are created and two samples are randomly omitted for each decision tree. The depth of each decision tree is "6".

In addition, in the comparative example and the example of the present invention, the usage period (year) of the metal material, the annual average temperature (°C), the annual average humidity (%), and the average amount of flying salt in daily units (mg/m ² /day(= mmd) (Cl ⁻ conversion)), and the average SO _X concentration (mg/m ² /day (=mmd) (So ₂ conversion)) on a daily basis are used as explanatory variables, and the corrosion amount (μm) of the metal material is a predictor variable. As a result, a regression model was trained to generate a corrosion amount prediction model. Note that the average amount of incoming salt and the average SO _X concentration are data on a daily basis as described above, but the value used for prediction is the average value in a predetermined period, such as the annual average value of each value. Here, the annual average value is used.

As shown in FIG. 4, an example using multiple regression analysis (see (a)) that cannot deal with nonlinearity and a regression model (see (b) to (d)) that can deal with nonlinearity were used. Comparing with the example, it was confirmed that the latter regression model reduces the rRMSE (relative root mean square of error) and improves the prediction accuracy of the corrosion amount of metal materials.

According to the method for generating a corrosion amount prediction model for a metal material and the corrosion amount prediction method for a metal material according to the first embodiment of the present invention as described above, a corrosion amount prediction model is generated by machine learning, and the corrosion amount prediction is performed. By using the model, it is possible to accurately predict long-term corrosion of metallic materials in an atmospheric corrosive environment. Further, according to the corrosion amount prediction method according to the present embodiment, the corrosion amount of the metal material during an arbitrary use period can be predicted with a simple configuration and process. Further, by using the corrosion amount prediction method according to the present embodiment, it becomes possible to select an optimum metal material according to the usage environment.

[Embodiment 2]
(Corrosion amount prediction device)
The configuration of the metal material corrosion amount prediction device according to the second embodiment of the present invention will be described. As shown in FIG. 1, the corrosion amount prediction device 1A according to the present embodiment has the same hardware configuration as the corrosion amount prediction device 1 described above, and is different only in the processing performed by the calculation unit 30. Therefore, description other than the processing in the arithmetic unit 30 will be omitted.

The learning unit 31 generates two prediction models by machine learning using a regression model. Of the two prediction models, the first prediction model is a corrosion amount prediction model that predicts the corrosion amount of a metal material in a predetermined period (for example, one year), and the second prediction model is the above-described predetermined amount. It is a damping parameter prediction model that predicts a damping parameter that indicates the decay of the corrosion rate of a metallic material for a period exceeding the period (for example, more than one year).

The learning unit 31 generates a corrosion amount prediction model and a decay parameter prediction model by machine learning using actual data including a plurality of environmental parameters, the corrosion amount of the metal material, and the usage period of the metal material. A specific method of generating the two prediction models in the learning unit 31 will be described later (see "Method of generating corrosion amount prediction model" described later).

The corrosion amount prediction unit 32 uses the corrosion amount prediction model (first prediction model) generated by the learning unit 31 to determine the environmental parameter of the metal material whose corrosion amount is to be predicted and the usage period (for example, one year) of the metal material. Is input to predict the corrosion amount of the metal material for a predetermined period.

In addition, the corrosion amount prediction unit 32 uses the attenuation parameter prediction model (second prediction model) generated by the learning unit 31 and the environmental parameter of the metal material whose corrosion amount is to be predicted, and a period exceeding a predetermined period (for example, one year). (Above) is input to predict the decay parameter of the corrosion amount of the metallic material over a predetermined period. Then, the corrosion amount prediction unit 32 predicts the corrosion amount of the metal material in the period exceeding the predetermined period based on the corrosion amount of the metal material in the predetermined period and the attenuation parameter. A specific method of predicting the corrosion amount in the corrosion amount prediction unit 32 will be described later (see "Corrosion amount prediction method" described later).

(Method of generating corrosion prediction model)
A method of generating a corrosion prediction model for metallic materials according to the second embodiment of the present invention will be described with reference to FIGS. 5 and 6. The method of generating the corrosion amount prediction model is implemented mainly by the learning unit 31 of the calculation unit 30.

In the method of generating the corrosion amount prediction model, the generation of a corrosion amount prediction model that predicts the corrosion amount of the metal material in a predetermined period (for example, one year) (see FIG. 5) and the corrosion of the metal material in the period exceeding the predetermined period are performed. The generation of the attenuation parameter of the amount (see FIG. 6) is performed, but the generation of the attenuation parameter may be performed only when it is desired to predict the corrosion amount of the metal material over a predetermined period. It may not be performed when only predicting. Further, the generation of the corrosion amount prediction model and the generation of the deceleration parameter are performed in advance before the corrosion amount prediction described below is performed.

(1) Generation of Corrosion Prediction Model First, a corrosion amount prediction model is generated from the temperature (temperature), relative humidity, absolute humidity, wetting time, rainfall, flying salt content, SO _X concentration, and NO _X concentration. An environment parameter used as an explanatory variable is selected (step S31). The method of selecting the environment parameter in this step is the same as that in step S11 (see FIG. 2) described above, and thus the description thereof is omitted.

Subsequently, the learning unit 31 acquires the corrosion amount of the metal material for a predetermined period and the environmental parameter thereof from the performance database (step S32).

Subsequently, the learning unit 31 generates a corrosion amount prediction model by a regression model that can be fitted to the nonlinearity of the data (specifically, environmental parameter) (step S33), and ends this flow. Since the method of generating the corrosion amount prediction model in this step is the same as that in step S13 (see FIG. 2) described above, the description thereof will be omitted.

(2) Generation of Attenuation Parameter Prediction Model First, the learning unit 31 determines the amount of corrosion of the metal material and its environmental parameter for a predetermined period, and the amount of corrosion of the metal material and its environmental parameter for a period exceeding the predetermined period. Is acquired from the performance database (step S41).

Subsequently, the learning unit 31 generates a damping parameter prediction model by a regression model that can be fitted to the non-linearity of data (specifically, environmental parameter) (step S42), and ends this flow.

Here, as shown in Non-Patent Document 1, it is known that the corrosion amount of a metal material in an atmospheric corrosive environment is expressed by the following formula (1) as an empirical formula.

In the above formula (1), Y is the amount of corrosion of the metal material after X years of use, A is the amount of corrosion of the metal material for one year, X is the period of use of the metal material, and B is formed by corrosion. It is a damping parameter indicating the decay of corrosion rate due to the effect of the rust layer.

By taking the logarithm of the above formula (1) and rearranging it, it can be transformed as shown in the following formula (2).

Here, in step S42 described above, the prediction model (Y prediction model of the above formula (1)) that directly predicts the corrosion amount of the metal material in the period exceeding the predetermined period is not directly generated, but is described above. A prediction model for predicting B in Expression (1) is generated. That is, the value of “logY−logA” in the above equation (2) is obtained for each actual data acquired in step S41, and the value of A prediction model for predicting the value of "logY-logA" is generated. This prediction model is the attenuation parameter prediction model in this embodiment. A value obtained by multiplying logX is used for each environment parameter used in the regression model.

(Corrosion amount prediction method)
A method of predicting the amount of corrosion of a metal material according to Embodiment 2 of the present invention will be described with reference to FIG. 7. The corrosion amount prediction method is mainly performed by the corrosion amount prediction unit 32. Further, in the following description, the predetermined period determined in advance will be described as “one year”.

First, the corrosion amount prediction unit 32 calculates the predicted value of the corrosion amount in the environmental parameter to be predicted from the corrosion amount prediction model generated in step S33 described above (step S51). In this step, the corrosion amount prediction model described above is used to predict the corrosion amount of the metal material for one year by inputting a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted. That is, in this step, by inputting the environmental parameter in the usage period (=one year) of the metal material to the corrosion amount prediction model, the corrosion amount of the metal material in the usage period and the environmental parameter is obtained as an output.

Subsequently, the corrosion amount prediction unit 32 determines whether or not to predict the corrosion amount for which the usage period is one year (step S52). When it is determined in step S52 that the corrosion amount for one year is to be predicted (Yes in step S52), the corrosion amount prediction unit 32 ends this flow and displays the corrosion amount prediction result on the display unit 40. ..

On the other hand, when it is determined in step S52 that the corrosion amount for which the usage period exceeds one year is predicted (No in step S52), the corrosion amount prediction unit 32 predicts the attenuation parameter in the environmental parameter to be predicted from the attenuation parameter prediction model. A value is calculated (step S53).

In step S52, by inputting a plurality of environmental parameters indicating the environment of the metal material whose corrosion amount is to be predicted and the usage period (=more than one year) of the metal material into the above-described attenuation parameter prediction model, The value of "logY-logA" in equation (2) is obtained as the output. Then, by dividing logX from the value of “logY−logA”, the attenuation parameter of the corrosion amount of the metal material in the use period and the environmental parameter is predicted.

Subsequently, the corrosion amount prediction unit 32 uses the predicted value of the corrosion amount of the metallic material calculated in step S51 (use period: one year) and the predicted value of the attenuation parameter calculated in step S53 to predict the use period ( =the predicted value of the corrosion amount in more than one year) is calculated (step S54), and this flow ends. Then, the corrosion amount prediction unit 32 displays the corrosion amount prediction result on the display unit 40.

(Example)
An example of the method for predicting the amount of corrosion of a metal material according to the second embodiment of the present invention will be described with reference to FIG. This figure is a graph showing the error between the predicted value and the actual value of the corrosion amount of the metal material, obtained using the corrosion amount prediction model used for the bridge. (A) of the same figure is a comparative example, and shows the result of predicting the corrosion amount by using this corrosion amount prediction model to generate a corrosion amount prediction model using the multiple regression analysis in the learning step described above. Further, (b) to (d) of the same figure are examples of the present invention. In the learning step described above, a corrosion amount prediction model is generated using decision tree regression, support vector regression, and random forest, and these corrosion amounts are calculated. The result of having predicted the amount of corrosion by a prediction model is shown.

Also, in the verification using the decision tree regression (see (b) of FIG. 8), the depth of the decision tree is “6”. In the verification using the support vector regression (see (c) of FIG. 8), the Gaussian kernel is used for the non-linear conversion. Further, in the verification using the random forest (see (d) of FIG. 8), 500 decision trees are created and two samples are randomly omitted for each decision tree. The depth of each decision tree is "6".

In addition, in the comparative example and the example of the present invention, the usage period (year) of the metal material, the annual average temperature (°C), the annual average humidity (%), and the average amount of flying salt in days (mg/m ² /day(= mmd) (Cl ⁻ conversion)), and the average SO _X concentration (mg/m ² /day (=mmd) (So ₂ conversion)) on a daily basis are used as explanatory variables, and the corrosion amount (μm) of the metal material is a predictor variable. As a result, a regression model was trained to generate a corrosion amount prediction model. The average flying salt content and the average SO _X concentration are data on a daily basis as described above, but the value used for prediction is the average value in a predetermined period, such as the annual average value of each value. Here, the annual average value is used.

As shown in FIG. 8, an example using multiple regression analysis (see (a)) that cannot deal with nonlinearity and a regression model (see (b) to (d)) that can deal with nonlinearity were used. Comparing with the example, it was confirmed that the latter regression model reduces the rRMSE and improves the prediction accuracy of the corrosion amount of the metal material. In addition, as compared with the results of the example of Embodiment 1 described above (see FIG. 4 ), it was confirmed that the relative mean square error is reduced in all regression models, and the accuracy of predicting the corrosion amount of the metal material is further improved. did it.

According to the method of generating the corrosion amount prediction model of a metal material and the corrosion amount prediction method of a metal material according to the second embodiment of the present invention as described above, the long-term corrosion prediction of a metal material can be performed more accurately in an atmospheric corrosive environment. It can be carried out. Further, by using the corrosion amount prediction method according to the present embodiment, it becomes possible to select an optimum metal material according to the usage environment.

As described above, the invention relates to a method for generating a corrosion amount prediction model for a metal material, a metal material corrosion amount prediction method, a metal material selection method, a metal material corrosion amount prediction program, and a metal material corrosion amount prediction device according to the present invention. Although the embodiments and the examples for carrying out the present invention have been specifically described, the gist of the present invention is not limited to these descriptions and should be broadly construed based on the description of the claims. Further, it goes without saying that various changes and modifications based on these descriptions are also included in the spirit of the present invention.

For example, in the above-described first embodiment, the method for generating a corrosion amount prediction model (see FIG. 2) and the corrosion amount prediction method using this prediction model (see FIG. 3) are performed at different timings. Step S21 of FIG. 3 may be performed after steps S11 to S13 of FIG.

Similarly, in the above-described second embodiment, a method for generating a corrosion amount prediction model (see FIG. 5), a method for generating a damping parameter prediction model (see FIG. 6), and a corrosion amount prediction method using these prediction models (see FIG. 7). However, for example, when steps S51 and S52 of FIG. 7 are performed after step S31 to S33 of FIG. 5 and a negative determination is made in step S52, step S41 of FIG. S42 may be performed, and steps S53 and S54 of FIG. 7 may be performed after step S42.

Here, in the above-described Embodiment 2, the initial corrosion amount A of the metal material in the initial year and the attenuation parameter B indicating the attenuation of the corrosion rate of the metal material are separately predicted, and the corrosion amount of the initial year is calculated. Although the long-term corrosion amount was predicted as a standard, the standard for predicting the long-term corrosion amount is not limited to the corrosion amount in the initial year.

That is, in the initial corrosion amount prediction step, the corrosion amount of the metal material in any predetermined predetermined period (initial period) is predicted, and in the long-term corrosion amount prediction step, the long-term corrosion amount is calculated based on the corrosion amount in the predetermined period described above. You may predict.

For example, when the corrosion amount of 1.5 years is given as A'as the initial corrosion amount, the formula for predicting the corrosion amount after X years is expanded from the above formula (1) to obtain the following formula (3). It can be described as follows.

If this is generalized, the following equation (4) can be obtained by setting the corrosion amount in a certain initial period X ₀ year as A′ and the damping parameter with reference to X ₀ year as B′. By using this formula (4), the corrosion amount in the period of X>X ₀ can be calculated as the corrosion amount based on X ₀ years.

The initial corrosion amount A′ of the metal material in an arbitrary initial period and the attenuation parameter B′ are separately predicted, and the elapsed years X′ after the initial period is calculated as the attenuation parameter B′ as shown in the above formula (4). The long-term corrosion amount after the initial period can be predicted by raising to the power of. However, the initial corrosion amount A in the above equation (1) is based on the corrosion amount for one year. Therefore, the period of X _{0 in} the above formula (4) is not assumed to be greatly deviated from one year, and it is considered that the practical range is about half a year to two years.

1, 1A Corrosion amount prediction device 10 Input unit 20 Result database 30 Calculation unit 31 Learning unit 32 Corrosion amount prediction unit 40 Display unit

Claims

Corrosion of the metal material according to the usage period by machine learning using a plurality of environmental parameters indicating the usage environment of the metal material, the amount of corrosion of the metal material, and the data including the usage period of the metal material. A method for generating a corrosion amount prediction model for a metal material, the method including a learning step for generating a prediction model for predicting the amount of corrosion.
The learning step is
A first learning step of generating a first prediction model for predicting the amount of corrosion of the metal material for a predetermined period of time,
A second learning step of generating a second prediction model for predicting a damping parameter indicating the decay of the corrosion rate of the metallic material for a period exceeding the predetermined period,
A method of generating a corrosion amount prediction model for a metal material according to claim 1, including:
The method for generating a corrosion amount prediction model for a metal material according to claim 1 or 2, wherein a learning method including decision tree regression, random forest, neural network, and support vector regression is used as the machine learning method.
The corrosion amount of the metal material is input by inputting a plurality of environmental parameters indicating a use environment of the metal material whose corrosion amount is to be predicted by the prediction model generated by the method according to claim 1 and a usage period of the metal material. A method of predicting the amount of corrosion of a metal material, the method including a prediction step of predicting.
With the first prediction model generated by the method according to claim 2, a plurality of environmental parameters indicating a usage environment of a metal material whose corrosion amount is to be predicted and a predetermined predetermined period are input, and the predetermined period of the predetermined period is input. A first prediction step of predicting the amount of corrosion of the metal material,
With the second prediction model generated by the method according to claim 2, a plurality of environmental parameters indicating a usage environment of the metal material whose corrosion amount is to be predicted, and a period exceeding the predetermined period are input, and the predetermined period is set. A second predicting step of predicting a damping parameter indicative of a decay of the corrosion rate of said metallic material over a period of time,
Corrosion amount of the metal material of the predetermined period, based on the attenuation parameter, a third prediction step of predicting the corrosion amount of the metal material in a period exceeding the predetermined period,
A method for predicting the amount of corrosion of metallic materials including.
The environmental parameters include at least one of temperature, relative humidity, absolute humidity, wetting time, and rainfall, and at least one of flying salt content, SO X concentration, and NO X concentration. Alternatively, the method of predicting the corrosion amount of a metal material according to claim 5.
A method of selecting a metal material that selects the metal material according to the usage environment by using the corrosion amount prediction method of the metal material according to claim 4 or claim 5.
The metal material is input by a computer using a plurality of environmental parameters indicating a usage environment of the metal material whose corrosion amount is to be predicted by the prediction model generated by the method according to claim 1 and a usage period of the metal material. A corrosion amount prediction program for metallic materials that functions as a prediction means for predicting the corrosion amount of metal.
Computer,
With the first prediction model generated by the method according to claim 2, a plurality of environmental parameters indicating a usage environment of a metal material whose corrosion amount is to be predicted and a predetermined predetermined period are input, and the predetermined period of the predetermined period is input. A first prediction means for predicting the amount of corrosion of the metal material,
With the second prediction model generated by the method according to claim 2, a plurality of environmental parameters indicating a usage environment of the metal material whose corrosion amount is to be predicted, and a period exceeding the predetermined period are input, and the predetermined period is set. A second predicting means for predicting a damping parameter indicating the decay of the corrosion rate of the metallic material over a period of time,
Corrosion amount of the metal material in the predetermined period, based on the attenuation parameter, a third predicting means for predicting the corrosion amount of the metal material in a period exceeding the predetermined period,
A program to predict the amount of corrosion of metallic materials to function as.
A plurality of environmental parameters indicating the usage environment of the metal material, the amount of corrosion of the metal material, and the usage period of the metal material are used, and the metal material is used according to the usage period by machine learning. Learning means for generating a prediction model for predicting the amount of corrosion of
With the prediction model, a plurality of environmental parameters indicating the use environment of the metal material to predict the amount of corrosion, and a period of use of the metal material as an input, a prediction means for predicting the amount of corrosion of the metal material,
An apparatus for predicting the amount of corrosion of metallic materials.
The learning means is
A first learning means for generating a first prediction model for predicting the amount of corrosion of the metal material for a predetermined period of time,
Second learning means for generating a second prediction model for predicting a damping parameter indicating the decay of the corrosion rate of the metal material in a period exceeding the predetermined period,
Equipped with
The prediction means is
By the first prediction model, by inputting a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted and a predetermined predetermined period, the corrosion amount of the metal material during the predetermined period is predicted. A first predictor,
By the second prediction model, a plurality of environmental parameters indicating the usage environment of the metal material whose corrosion amount is to be predicted and a period exceeding the predetermined period are input, and the corrosion of the metal material during the period exceeding the predetermined period is input. A second predicting means for predicting a damping parameter indicating speed damping,
Corrosion amount of the metal material in the predetermined period, based on the attenuation parameter, a third predicting means for predicting the corrosion amount of the metal material in a period exceeding the predetermined period,
The corrosion amount prediction device for a metal material according to claim 10, further comprising: