Corrosion evaluation method, sensor current value derivation model, and control device

Abstract

To accurately estimate a sensor current value correlated with the corrosion rate of a metal without actually measuring the amount of deposits such as sea salt on the metal in the environment.SOLUTION: A corrosion evaluation method includes: a virtual sea salt variable derivation step of deriving a virtual sea salt variable correlated with an amount of sea salt adhering to a metal at predetermined measurement timing in the environment; and a sensor current value estimation step of estimating a sensor current value at a non-measurement point on the basis of the environmental factors which include a value obtained by summing the virtual sea salt variables derived at the measurement timing of the time point in the environmental factors of a sensor current value derivation model generation step as the cumulative virtual sea salt variable at the time point and serve as explanatory variables at the non-measurement point of the sensor current value and the sensor current value derivation model derived in the sensor current value derivation model generation step.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a corrosion evaluation method for evaluating the corrosiveness of metals in an environment, a sensor current value derivation model, and a control device.

Background Art

[0002] As a method for evaluating the corrosion rate of metals, a method of installing an ACM (Atmospheric Corrosion Monitor) sensor (atmospheric corrosion sensor) and monitoring the atmospheric environment is known. The ACM sensor records the galvanic current when a water film is formed on the sensor surface due to dew condensation or the like, and this current value is used for evaluating the corrosiveness in the atmospheric environment because it has a correlation with the corrosion rate.

[0003] Recently, attempts have been made to estimate and evaluate the corrosion rate of a target location without installing an ACM sensor by estimating the current value indicated by the ACM sensor from various environmental factors. For example, in the technique disclosed in Patent Document 1, when performing multiple regression analysis with the corrosion rate of a metal as the target variable and environmental factors that affect the corrosion rate as explanatory variables, it is proposed to include a virtual wetting time obtained by weighting and adding the relative humidity as an explanatory variable. Here, the current value indicated by the ACM sensor tends to increase depending on the amount of deposits such as sea salt. In other words, the sensor current value related to the corrosiveness of the metal to be evaluated tends to increase depending on the amount of deposits such as sea salt on the metal. In view of this point, Patent Document 1 shows that the amount of deposits on the metal should be taken into account in the weighting of the relative humidity.

[0004] In the technique disclosed in Patent Document 2, there is provided a corrosion evaluation method for evaluating the corrosiveness of a metal in an environment, in which a correlation relationship between temperature, humidity, the amount of sea salt particles which is the amount of sea salt adhered per unit area to the metal, and the sensor current amount of the ACM sensor is obtained in advance, and based on the correlation relationship, the temperature, humidity, and amount of sea salt particles measured in a predetermined environment, the sensor current amount is calculated.

Prior Art Documents

Patent Document

[0005]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0006] In the technology disclosed in Patent Document 1, regarding the amount of deposits such as sea salt that greatly affects the current value indicated by the ACM sensor, since weighting is performed based on the actual measured value, it is necessary to measure the amount of deposits at each point where the corrosion rate is evaluated. For this reason, there has been a problem that it is difficult to evaluate the corrosion rate over a wide range.

[0007] In the technology disclosed in Patent Document 2, since the amount of sea salt particles is included as an explanatory variable, it is necessary to measure the amount of sea salt particles at each evaluation point of the corrosion rate. For this reason, similar to the technology disclosed in Patent Document 1 described above, there has been a problem that it is difficult to realistically evaluate the corrosion rate over a wide range.

[0008] The present invention has been made in view of the above problems, and an object thereof is to provide a corrosion evaluation method, a sensor current value derivation model, and a control device that accurately estimate a sensor current value correlated with the corrosion rate of a metal without actually measuring the amount of deposits such as sea salt on the metal in the environment.

Means for Solving the Problems

[0009] The corrosion evaluation method for achieving the above object is A corrosion evaluation method for evaluating the corrosiveness of a metal in an environment, and its characteristic configuration is In the above environment, a sensor current value derivation model generation step is executed to generate a sensor current value derivation model for deriving the sensor current value by performing machine learning with the sensor current value measured by the ACM sensor as the target variable and environmental factors that affect the sensor current value in the above environment as the explanatory variables. A virtual sea salt variable derivation step is performed to derive a virtual sea salt variable that is derived from the sea wind speed, which is the wind speed component of the wind from the sea to the above environment, based on the wind direction and wind speed in the above environment, and the sensor wind speed, which is the wind speed component of the wind in the windward direction of the metal, based on the wind direction and wind speed in the above environment, and is correlated with the amount of sea salt adhering to the metal at a predetermined measurement timing in the above environment. The value obtained by adding up the virtual sea salt variables derived at the above measurement timing up to the current time is included as the integrated virtual sea salt variable at the current time in the environmental factors in the above sensor current value derivation model generation step. A sensor current value estimation step is performed to estimate the sensor current value at the non-measurement point based on the environmental factor as the explanatory variable at the non-measurement point of the sensor current value and the sensor current value derivation model derived in the sensor current value derivation model generation step.

[0010] As described so far, in the prior art, it has been necessary to measure the amount of deposits such as sea salt for each point where the sensor current value is estimated or the corrosion rate is evaluated, and it has been difficult to evaluate the corrosion rate over a wide range. The inventors of the present invention introduced the concept of a virtual sea salt variable that correlates with the amount of sea salt adhering to metal in the environment in a sensor current value derivation model for deriving a sensor current value by performing machine learning with the sensor current value measured by an ACM sensor in the environment as the target variable and the environmental factors that affect the sensor current value in the environment as the explanatory variables. Then, the virtual sea salt variable was defined as being derived from the sea wind speed, which is the wind speed component of the wind from the sea to the environment and is derived based on the wind direction and wind speed in the environment, and the sensor wind speed, which is the wind speed component of the wind in the direction of the wind received by the ACM sensor and is derived based on the wind direction and wind speed in the environment. Further, the concept of an integrated virtual sea salt variable was introduced, which is the sum of the virtual sea salt variables derived at the measurement timing up to that point.

[0011] That is, in the corrosion evaluation method having the above-described characteristic configuration, as an environmental factor in the sensor current value derivation model generation step for generating a sensor current value derivation model, the integrated virtual sea salt variable described above is included. In the sensor current value estimation step, based on the environmental factor as the explanatory variable at the non-measurement point of the sensor current value and the sensor current value derivation model derived in the sensor current value derivation model generation step, the sensor current value at the non-measurement point is estimated. Therefore, at each point where the sensor current value is estimated (the point where the corrosion evaluation is performed), it is not necessary to directly measure the sea salt, and it is sufficient to measure the sea wind speed and the sensor wind speed. As a result, compared with the conventional technique that requires direct measurement of sea salt, the sensor current value can be estimated (corrosion evaluation can be performed) over a wide range. In addition, the inventors have confirmed, based on the test results described later, that the corrosion evaluation method having the above-described characteristic configuration can estimate the sensor current value with relatively high accuracy.

[0012] A further characteristic configuration of the corrosion evaluation method is that the virtual sea salt variable as the environmental factor is the first-order product of the sea wind speed and the sensor wind speed.

[0013] Incidentally, the inventors have confirmed that when the conditions such as environmental factors are made the same, the estimation accuracy of the sensor current value is the highest (the value of the coefficient of determination is closest to 1 and the value of MSE is the smallest) when the virtual sea salt variable as an environmental factor is the first-order product of the sea wind speed and the sensor wind speed, compared with the case where the virtual sea salt variable is other values shown in the following [Table 1].

[0014]

Table 1

[0015] A further characteristic configuration of the corrosion evaluation method is obtaining the amount of electricity for the time from the two sensor current values continuously measured by the ACM sensor at the predetermined measurement timing in the environment and the time between the continuous measurement timings, and adding up the cumulative sensor electricity amount obtained in a predetermined measurement period, executing a conversion model generation step of generating a conversion model from the cumulative sensor electricity amount to the corrosion metal loss amount by performing a correlation analysis between the cumulative sensor electricity amount and the corrosion metal loss amount derived from the measurement value of the RCM sensor measured in the measurement period in the environment, The corrosion metal loss amount estimation step of estimating the corrosion metal loss amount from the cumulative sensor electricity amount obtained by sequentially estimating the sensor current value at the non-measurement point estimated in the sensor current value estimation step at a predetermined corrosion prediction period, converting it into an amount of electricity, and adding them up, and the conversion model. Incidentally, the RCM sensor is composed of the same material or the same type of material as the metal to be evaluated, and is a sensor that can accurately estimate the corrosion amount of the metal from the measurement value of the sensor.

[0016] According to the above characteristic configuration, in the conversion model generation step, the amount of electricity for the time is obtained from two sensor current values continuously measured by the ACM sensor at a predetermined measurement timing in the environment and the time between consecutive measurement timings, and the cumulative sensor electricity amount obtained by adding them up during a predetermined measurement period is subjected to correlation analysis with the amount of corrosion loss estimated from the measurement value of the RCM sensor measured during the measurement period in the environment, thereby generating a conversion model from the cumulative sensor electricity amount to the amount of corrosion loss. In the corrosion loss estimation step, the sensor current value estimated by the method described so far is sequentially estimated during the corrosion estimation period, converted into the amount of electricity, added up to obtain the cumulative sensor electricity amount, and the amount of corrosion loss can be estimated from the conversion model. Thereby, in both the step of generating the conversion model and the step of estimating the amount of corrosion loss, without directly measuring the amount of sea salt adhesion, the amount of corrosion loss at non-measurement points where the sensor current value is not directly measured can be estimated well.

[0017] A further characteristic configuration of the corrosion evaluation method is In the conversion model generation step, the amount of corrosion loss is estimated using the moving minimum value, which is the lowest value of the values measured over the measurement period, among the measurement values by the RCM sensor measured at the measurement timing of the measurement period in the environment.

[0018] In the conversion model generation step, it is preferable that the amount of corrosion loss is estimated using the moving minimum value, which is the lowest value of the values measured over the measurement period, among the measurement values by the RCM sensor measured at the measurement timing of the measurement period in the environment. The measurement value of the RCM sensor contains a lot of upward swing noise, but by using the moving minimum value as in the above characteristic configuration, it becomes possible to track the data trend well.

[0019] A further characteristic configuration of the corrosion evaluation method is The sensor current value derivation model is a multiple regression equation obtained by performing multiple regression analysis as the machine learning, or a regression tree obtained by performing analysis using a regression model as the machine learning.

[0020] As a sensor current value derivation model, the inventors of the present invention have adopted a multiple regression equation obtained by performing multiple regression analysis as machine learning, or a regression tree obtained by performing analysis using a regression model as machine learning. As shown in the test results described later, it has been confirmed that the sensor current value can be estimated well.

[0021] A further characteristic configuration of the corrosion evaluation method is In the sensor current value derivation model generation step, in addition to the environmental factors, machine learning is performed using a time factor representing the time-series change of the sensor current value as an explanatory variable.

[0022] The inventors of the present invention have confirmed that by performing machine learning using, as an explanatory variable, a time factor representing the time-series change of the sensor current value in addition to environmental factors in the sensor current value derivation model generation step, the estimation accuracy of the sensor current value can be improved as shown in the test results described later. Incidentally, the "time factor representing the time-series change of the sensor current value" refers to, in other words, a time factor that indirectly affects the sensor current value via temperature and humidity.

[0023] The sensor current value derivation model for achieving the above object is A sensor current value derivation model that derives the sensor current value by using, as an objective variable, the sensor current value measured by an ACM sensor in an environment, and performing machine learning using, as explanatory variables, environmental factors that affect the sensor current value in the environment. Its characteristic configuration is Derived from the sea wind speed, which is the wind speed component of the wind from the sea to the environment derived based on the wind direction and wind speed in the environment, and the sensor wind speed, which is the wind speed component of the wind in the windward direction of the metal derived based on the wind direction and wind speed in the environment, a virtual sea salt variable that is correlated with the amount of sea salt adhering to the metal at a predetermined measurement timing in the environment is derived. The value obtained by adding up the virtual sea salt variables derived at the measurement timing up to the current time is included in the environmental factors as the integrated virtual sea salt variable at the current time.

[0024] By using the sensor current value derivation model having the above-described characteristic configuration, it is not necessary to directly measure sea salt at each point where the sensor current value is estimated (the point where corrosion evaluation is performed), and it is sufficient to measure the sea wind speed and the sensor wind speed. As a result, the sensor current value can be estimated over a wide range as compared with the prior art in which it is necessary to directly measure sea salt.

[0025] A further characteristic configuration of the sensor current value derivation model is In addition to the environmental factors, the sensor current value is derived by performing machine learning with a time factor representing the time-series change of the sensor current value as an explanatory variable.

[0026] The inventors of the present invention have confirmed that, in the sensor current value derivation model, by performing machine learning with a time factor representing the time-series change of the sensor current value as an explanatory variable in addition to the environmental factors, the estimation accuracy of the sensor current value can be improved as shown in the test results described later.

[0027] The characteristic configuration of the control device for achieving the above object is Based on the above-described sensor current value derivation model and the environmental factors as explanatory variables at non-measurement points of the sensor current value, the sensor current value at the non-measurement points is estimated.

[0028] According to the above characteristic configuration, a control device capable of estimating the sensor current value over a wide range can be realized as compared with the prior art in which it is necessary to directly measure sea salt.

Brief Description of the Drawings

[0029]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Mode for Carrying Out the Invention

[0030] The corrosion evaluation method according to an embodiment of the present invention is a corrosion evaluation method for accurately estimating a sensor current value (and corrosion loss amount) correlated with the corrosion rate of a metal without actually measuring the amount of deposits such as sea salt on the metal in the environment, and provides a sensor current value derivation model and a control device. Hereinafter, embodiments of the corrosion evaluation method will be described with reference to FIGS. 1 to 10.

[0031] As shown in FIG. 1, the corrosion evaluation method according to the embodiment is a corrosion evaluation method for evaluating the corrosivity of a metal in the environment. A sensor current value measured by an ACM sensor AC in the environment is used as an objective variable, and environmental factors that affect the sensor current value in the environment are used as explanatory variables for machine learning to generate a sensor current value derivation model for deriving the sensor current value. In the corrosion evaluation method, a virtual sea salt variable correlated with the amount of sea salt deposited on the metal at a predetermined measurement timing in the environment is derived from a sea wind speed, which is a wind speed component of the wind from the sea to the environment derived based on the wind direction and wind speed in the environment, and a sensor wind speed, which is a wind speed component of the wind in the direction of the metal receiving the wind derived based on the wind direction and wind speed in the environment. A virtual sea salt variable derivation step is performed. A value obtained by adding up the virtual sea salt variables derived at the measurement timings up to the current time is used as the integrated virtual sea salt variable at the current time and included in the environmental factors in the sensor current value derivation model generation step. Based on the environmental factors as explanatory variables at non-measurement points of the sensor current value and the sensor current value derivation model derived in the sensor current value derivation model generation step, a sensor current value estimation step for estimating the sensor current value at the non-measurement points is performed. In addition, in the present embodiment, it is assumed that the measured values of the ACM sensor AC and the RCM sensor RC are values obtained by performing predetermined numerical processing such as peak cutting unless otherwise specified. Incidentally, as the metal, for example, metals facing in all directions at all locations inside and outside a certain factory can be targeted.

[0032] The corrosion evaluation method is effective in an environment where sea salt may adhere to the metal for which corrosion is being evaluated, and learning data is acquired by a corrosion evaluation apparatus 100 related to corrosion evaluation as shown in FIG. 1. More specifically, as shown in FIG. 1, the corrosion evaluation apparatus 100 according to the present embodiment includes a control device S that can receive and analyze output values of various sensors, and is installed in an environment exposed to wind from the sea (for example, near the coastline KY). As an environmental factor measurement sensor group K, it includes a wind speed sensor K1 for measuring wind speed, a rainfall sensor K2 for measuring rainfall, a humidity sensor K3 for measuring the humidity of the atmosphere (relative humidity or absolute humidity), a temperature sensor K4 for measuring the temperature of the atmosphere, and a solar radiation sensor K5 for measuring solar radiation. The control device S is an information processing device realized by the cooperation of hardware such as a memory and a CPU and software.

[0033] Furthermore, the corrosion evaluation apparatus 100 includes, as a target variable related value measurement sensor group F, an ACM sensor AC and an RCM sensor RC capable of measuring a change amount of an electric resistance value measured during a predetermined measurement period. In addition, when measuring environmental factors that affect the sensor current value at a non-measurement point described later, only the above-described environmental factor measurement sensor group K may be provided at the non-measurement point, and the acquired data may be sent to a data center (not shown) different from the non-measurement point via an electric communication line (not shown) or the like.

[0034] Now, the control device S can derive, from the wind speed V measured by the wind speed sensor K1, as shown in FIG. 1, a sea wind speed Vcosθ1 which is a wind speed component of the wind from the sea to the environment derived based on the wind direction and wind speed in the environment, and a sensor wind speed Vcosθ2 which is a wind speed component of the wind in the wind receiving direction of the ACM sensor AC and the RCM sensor RC derived based on the wind direction and wind speed in the environment.

[0035] When an explanation is added, the sea wind speed Vcosθ1 is a component along the line segment D1 passing through the wind speed sensor K1 and orthogonal to the coastline KY closest to the corrosion evaluation device 100 among the wind speeds V measured by the wind speed sensor K1, and is represented by the following [Equation 1].

[0036]

Equation

[0037] Furthermore, the sensor wind speed Vcosθ2 is a wind speed component in the wind receiving direction of the metal (the line segment D2 along the direction orthogonal to the wind receiving surface of the wind speed sensor K1) among the wind speeds V measured by the wind speed sensor K1, and is represented by the following [Equation 2]. Incidentally, in the corrosion evaluation device 100, although detailed illustration is omitted, it is assumed that the wind receiving direction of the metal (for example, the opening direction of the wind inlet of the wind box when the metal is installed in the wind box) coincides with the direction orthogonal to the wind receiving surface of the wind speed sensor K1.

[0038]

Equation

[0039] Now, in the corrosion evaluation method according to the present embodiment, the first-order product of the above-described sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2 at a predetermined measurement timing is set as a virtual sea salt variable correlated with the amount of sea salt adhering to the ACM sensor AC in the environment at that measurement timing. And after a predetermined measurement period has elapsed since the start of measurement, the integrated virtual sea salt variable correlated with the total amount of sea salt adhering to the ACM sensor AC is the sum of the first-order products of the sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2 for each of a plurality of measurement timings included in the measurement period.

[0040] Hereinafter, the virtual sea salt variable (more specifically, the integrated virtual sea salt variable) is used as an environmental factor in the sensor current value derivation model generation step to estimate the sensor current value and to estimate the corrosion reduction amount. Next, regarding the estimation of the sensor current value and the estimation of the corrosion loss amount, an explanation will be given along the processing flow of the corrosion evaluation method shown in FIG. 2.

[0041] In a predetermined environment, the control device S executes a virtual sea salt variable derivation step of deriving a virtual sea salt variable that is the first-order product of the measured sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2 at a predetermined measurement timing (#01). Furthermore, after a predetermined measurement period has elapsed since the start of measurement, an integrated virtual sea salt variable correlated with the total amount of sea salt adhering to the ACM sensor AC is derived as the sum of the first-order products of the sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2 for each of a plurality of measurement timings included in the measurement period. Hereinafter, the "integrated virtual sea salt variable at a predetermined measurement timing" shall mean the sum of the virtual sea salt variables derived from the start of measurement to the predetermined measurement timing.

[0042] At the measurement timing for measuring the above-mentioned sea wind speed Vcosθ1 and sensor wind speed Vcosθ2, the control device S measures the sensor current value with the ACM sensor AC as learning data for the target variable, and measures the rainfall with the rainfall sensor K2, the humidity of the atmosphere with the humidity sensor K3, the temperature of the atmosphere with the temperature sensor K4, and the solar radiation amount with the solar radiation amount sensor K5 as environmental factors that are learning data for the explanatory variable (#02). The control device S obtains, as learning data, a set of learning data for each measurement timing, that is, a set of sensor current values, integrated virtual sea salt variables, rainfall, humidity, temperature, and solar radiation amount for each predetermined measurement timing, at the measurement timing of a predetermined measurement period. Here, it is assumed that the measurement period and the measurement timing are, for example, at 30-minute intervals for one month.

[0043] The control device S executes a sensor current value derivation model generation step of generating a sensor current value derivation model by performing machine learning using a part of the learning data obtained in steps #01 to #02 (#03). Furthermore, as the sensor current value derivation model, a multiple regression equation obtained by performing multiple regression analysis as machine learning, or a regression tree obtained by performing analysis using a regression model (e.g., XGBoost) as machine learning can be preferably used. For example, when the sensor current value derivation model is a multiple regression equation, detailed calculation results will be described later, but it is as follows in [Equation 3] below. Here, a step of verifying the accuracy of the generated sensor current value derivation model can be executed using the rest of the learning data.

[0044]

Equation

[0045] In [Equation 3], a is the relative humidity (%), b is the temperature (°C), c is the rainfall (mm), d is the solar irradiance (μW / cm 2 ), e is the integrated virtual sea salt variable, and α1 to α6 are coefficients calculated by machine learning.

[0046] Next, at an unmeasured point as the derivation target point of the estimated value of the sensor current value, the sea wind speed, sensor wind speed, humidity, temperature, rainfall, and solar irradiance, which are environmental factors, are measured (#04), and the measured environmental factors are substituted into the sensor current value derivation model to execute a sensor current value estimation step of deriving an estimated value of the sensor current value (#05). More specifically, in the sensor current value estimation step of step #05, at the unmeasured point, the integrated virtual sea salt variable obtained by adding the first-order product of the sea wind speed and the sensor wind speed measured up to the time when the sensor current value estimation step is executed to the sensor current value derivation model, and the humidity, temperature, rainfall, and solar irradiance at the time of execution of the sensor current value estimation step are used to derive the sensor current value. That is, in order to derive the sensor current value according to this embodiment, since it is necessary to derive the integrated virtual sea salt variable (the value obtained by adding the virtual sea salt variables), at least the sea wind speed and the sensor wind speed for deriving the virtual sea salt variable need to be measured from the time (the start time of the estimation period for estimating the corrosion thinning amount of the metal) when the metal is installed in the environment (unmeasured point) where it is corroded.

[0047] The control device S obtains the amount of electricity for the time from two sensor current values continuously measured by the ACM sensor AC at a predetermined measurement timing in the environment and the time between consecutive measurement timings, and adds up the amount of electricity for the predetermined measurement period. The cumulative sensor electricity amount obtained is subjected to correlation analysis with the corrosion loss amount derived from the measurement value of the RCM sensor measured during the measurement period in the environment, and a conversion model generation step for generating a conversion model from the cumulative sensor electricity amount to the corrosion loss amount is executed (#06).

[0048] Note that the corrosion loss amount R is obtained by converting the measurement value i of the RCM sensor RC based on a known predetermined conversion formula. Incidentally, as the measurement value i of the RCM sensor RC, the moving minimum value, which is the minimum value of the values measured over a measurement period (for example, one day), can be preferably used (Reference: Tomoyasu Suzuki, "Examination of the Application of ACM Sensors in the Corrosion Evaluation of Steel in Freshwater and Treated Water" (Materials and Environment, 68, 201 - 204 (2019))). R is converted based on a known predetermined conversion formula. Incidentally, as the measurement value i of the RCM sensor RC, R the moving minimum value, which is the minimum value of the values measured over a measurement period (for example, one day), can be preferably used (Reference: Tomoyasu Suzuki, "Examination of the Application of ACM Sensors in the Corrosion Evaluation of Steel in Freshwater and Treated Water" (Materials and Environment, 68, 201 - 204 (2019))).

[0049] The cumulative sensor electricity amount E is expressed by the following [Equation 4] when the sensor current value measured by the ACM sensor AC is i. A (Reference: Masahiro Yamamoto, "Continuous Measurement of Corrosion Rate of Steel in Outdoor Environment Using Alternating Current Impedance Method" (Journal of the Japan Institute of Metals, Vol. 65, No. 6 (2001) 465 - 469)). The [Equation 4] converts the output value I t [μA] of the ACM sensor AC at the elapsed time t [minutes] measured every 30 minutes into the cumulative sensor electricity amount Q t [C]. Here, I k-1 is the output value 30 minutes before I k . Also, "10 -6 " is a value for unit conversion from μA to A, "30" means 30 minutes, and "60" is a value for unit conversion from minutes to seconds. In other words, it is the result of performing time - series integration on the current value.

[0050]

Equation

[0051] Here, although the details of the generated conversion model are omitted, as an example of the conversion model, a logarithmic approximation formula shown by a solid line in FIG. 5 can be cited. In addition, a more accurate approximation formula can be generated by using machine learning or the like.

[0052] The control device S executes a corrosion loss amount estimation step of estimating the corrosion loss amount from the cumulative sensor electrical quantity obtained by sequentially estimating the sensor current value at a non-measured point estimated in the sensor current value estimation step of step #05, converting it into an electrical quantity, and adding them together, and the conversion model (#07).

[0053] Next, the calculation results are shown below for the case where the sensor current value derivation model is a multiple regression equation obtained by performing multiple regression analysis as machine learning, and the case where it is a regression tree obtained by performing analysis using a regression model as machine learning.

[0054] 〔When the sensor current value derivation model is a multiple regression equation〕 The measured values of the above-described ACM sensor AC and RCM sensor RC and the values related to the above-described environmental factors were measured at a plurality of predetermined points in Japan from January to March 2023 for three months, and using the measurement data, the sensor current value was estimated and the corrosion loss amount was estimated.

[0055] In the sensor current value derivation model generation step, the sensor current value was estimated by substituting the environmental factors measured at another point (non-measured point) different from the above-mentioned one point into the sensor current value derivation model (multiple regression equation shown in [Equation 3]) obtained in the sensor current value derivation model generation step, using the data obtained at one point among the plurality of points. The relationship between the actually measured value and the estimated value of the sensor current value is shown in FIGS. 3 and 4. The coefficient of determination (0 ≦ R), which is an index for evaluating how well the sensor current value derivation model fits the data with respect to the estimated value of the sensor current value. 2≦1) is "0.45", which is relatively close to 1, and it can be said that the prediction (estimation) accuracy is high. Also, the MSE (Mean Squared Error), which is a general index for evaluating the performance of a machine learning model, was "1.10". The smaller the MSE, the smaller the difference between the predicted value and the actual observed value. The calculated value "0.88" is small enough, and it can be said that the generated sensor current value derivation model is a model with good accuracy.

[0056] By executing the conversion model generation step and the corrosion thinning amount estimation step, from the correlation relationship (shown in FIG. 5) between the cumulative charge amount (hereinafter referred to as "ACM sensor cumulative charge amount") obtained by converting the sensor current value measured by the ACM sensor according to the above [Equation 4] and the corrosion thinning amount (hereinafter referred to as "RCM sensor thinning amount") derived from the measurement value of the RCM sensor, a conversion formula (the formula shown by the solid line in FIG. 5) for converting the ACM sensor cumulative charge amount into the RCM sensor thinning amount is obtained. FIG. 6 shows a comparison between the estimated value and the measured value of the RCM sensor thinning amount obtained by substituting the sensor current value at a certain point estimated above into the conversion formula. It can be seen from FIG. 6 that the estimated value estimates the measured value with relatively high accuracy.

[0057] 〔When the sensor current value derivation model is a regression tree (XGBoost)〕 Similar to the case where the sensor current value derivation model is a multiple regression equation, the sensor current value was estimated. The relationship between the measured value and the estimated value of the sensor current value is shown in FIGS. 7 and 8. Regarding the estimated value of the sensor current value, the coefficient of determination (0 ≦ R 2 ≦1) is "0.47", and the MSE (Mean Squared Error) was "1.18". From the values of the coefficient of determination and the MSE, it can be said that for the current measurement data, the estimation accuracy in the sensor current value derivation model is equivalent between the multiple regression equation and the regression tree (XGBoost).

[0058] Furthermore, by performing the conversion model generation step and the corrosion metal loss estimation step, a conversion formula (the formula shown by the solid line in Fig. 5) for converting the ACM sensor cumulative charge amount into the RCM sensor metal loss can be obtained. Fig. 9 shows a comparison between the estimated value and the measured value of the RCM sensor metal loss obtained by substituting the sensor current value at a certain point into the conversion formula. It can be seen from Fig. 9 that the estimated value of the corrosion metal loss estimates the measured value with relatively high accuracy.

[0059] 〔Alternative Embodiment〕 (1) In the corrosion evaluation method according to the above embodiment, in addition to the step of estimating the sensor current value (#01~#05), the control for performing the step of estimating the corrosion metal loss (#06~#07) was described. The corrosion evaluation method according to the present invention may omit the step of estimating the corrosion metal loss (#06~#07) and only perform the step of estimating the sensor current value (#01~#05).

[0060] (2) In the above embodiment, the sea wind speed and the sensor wind speed may be configured to be estimated from the wind direction and wind speed data included in meteorological data such as AMeDAS of the Japan Meteorological Agency without directly measuring them. Also, regarding the rainfall, humidity, temperature, and solar radiation amount as environmental factors, they may be obtained or estimated from meteorological data such as AMeDAS of the Japan Meteorological Agency. In this case, the environmental factor measurement sensor group K in the corrosion evaluation apparatus 100 for a predetermined environment can be omitted.

[0061] (3) In the above embodiment, rainfall, humidity, temperature, and solar radiation amount were exemplified as environmental factors that affect the sensor current value, but not all of them need to be included as environmental factors, or other environmental factors may be included.

[0062] (4) In the above embodiment, the environment where the teacher data is acquired and the non-measurement points related to the data of environmental factors for estimating the sensor current value etc. are not limited to the vicinity of the coastline KY, and may be various points such as at sea or on land.

[0063] (5) In the above embodiment, the virtual sea salt variable is assumed to be the first-order product of the measured sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2. The reason for setting the virtual sea salt variable as the first-order product of the sea wind speed and the sensor wind speed is that, as the virtual sea salt variable, various relational expressions using the sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2 as variables were used for machine learning. As a result, it was estimated that setting the first-order product of the sea wind speed Vcosθ1 and the sensor wind speed Vcosθ2 as the virtual sea salt variable gave the highest accuracy of the estimated sensor current value. However, the virtual sea salt variable may be defined by other expressions using the sea wind speed and the sensor wind speed as variables, and a sensor current value derivation model may be generated.

[0064] (6) In the above embodiment, as the sensor current value derivation model, an example was given of a multiple regression equation obtained by performing multiple regression analysis as machine learning, or a regression tree obtained by performing analysis using a regression model (for example, XGBoost) as machine learning. However, without being limited to the above configuration, models such as random forest and time series prediction may be adopted.

[0065] (7) In the above embodiment, in the sensor current value derivation model generation step, an evaluation method was shown for generating a sensor current value derivation model that derives a sensor current value by performing machine learning with environmental factors that affect the sensor current value in the environment as explanatory variables. As another evaluation method, in the sensor current value derivation model generation step, in addition to environmental factors, it may also be a method of performing machine learning with a time factor representing the time series change of the sensor current value as an explanatory variable. For example, as an index related to "the number of days elapsed from the starting day" as the time factor that is an explanatory variable, there are "the number of days elapsed from the starting day (DOY in [Equation 5], [Equation 6], and Figure 10)" itself, "a sine function with the number of days elapsed from the starting day as a variable (hereinafter DOY_sin in [Equation 5] and Figure 10)", and "a cosine function with the number of days elapsed from the starting day as a variable (hereinafter DOY_cos in [Equation 6] and Figure 10)". Note that DOY is obtained by subtracting the number of leap days (N) in that period.

[0066]

Number

[0067]

Number

[0068] Here, the "number of days elapsed from the starting date (DOY in [Number 5] and [Number 6])" is calculated as "the date for which the corrosion weight loss is to be determined - the starting date". Regarding the "number of days elapsed from the starting date", various starting dates can be used as long as they are reasonably determined for evaluating the time change in the corrosion weight loss of the target metal. However, the date when the target metal for evaluating the corrosion weight loss is exposed to the corrosion environment at the evaluation site can preferably be adopted. In addition, as an index related to the "number of days elapsed from the starting date", it has been confirmed that by adopting at least one or more of the above-mentioned ones, the derivation accuracy of the sensor current value can be improved.

[0069] Furthermore, as an index related to the "elapsed time from the starting time" as a time factor of the explanatory variable, there are "the elapsed time from the starting time (TOD in [Number 7], [Number 8], and Figure 11])" itself, "a sine function with the elapsed time from the starting time as a variable (hereinafter TOD_sin in [Number 7] and Figure 11)", and "a cosine function with the elapsed time from the starting time as a variable (hereinafter [Number 8], TOD_cos in Figure 11)".

[0070]

Number

[0071]

Number

[0072] Here, the "elapsed time from the starting time (TOD in [Equation 7] and [Equation 8])" is calculated as "target time (target time of the target day for determining corrosivity) - starting time (starting time of the starting day)". More specifically, if the "target time of the target day for determining the corrosion weight loss" is set to "9:00 (24-hour format) on November 22, 2024" and the "starting time of the starting day" is set to "0:00 (24-hour format) on January 1, 2022", the "elapsed time from the starting time" is calculated as "9:00 - 0:00 = 9 hours" ignoring the date. Regarding the "elapsed time from the starting time", various starting times can be used as long as they are reasonably determined for evaluating the time change of the corrosion weight loss of the target metal. For example, the time when the target metal for evaluating the corrosion weight loss starts to be exposed to the above environment can be preferably adopted. In addition, as an index related to the "elapsed time from the starting time", it has been confirmed that the derivation accuracy of the sensor current value can be improved by adopting at least one or more of the above-mentioned ones.

[0073] Now, in the sensor current value derivation model generation step, in addition to environmental factors, when using multiple regression analysis as machine learning to derive the sensor current value by using, as explanatory variables, time factors that represent the time series change of the sensor current value, the following [Equation 9] can be preferably used instead of the above [Equation 3].

[0074]

Equation

[0075] In addition, even when the sensor current value derivation model is a regression tree (XGBoost), it has been confirmed that the calculation accuracy can be improved by performing machine learning with the above-mentioned time factors as explanatory variables.

[0076] When the above six indicators are adopted as time factors, in multiple regression analysis, the coefficient of determination is "0.45" and the MSE is "1.02", and it has been confirmed that the accuracy is improved compared with the numerical values shown in [Table 1]. On the other hand, even when using regression trees (XGBoost), the coefficient of determination is "0.50" and the MSE is "0.95", and it has been confirmed that the accuracy is improved compared with the numerical values shown in [Table 1].

[0077] In addition, the configurations disclosed in the above embodiments (including other embodiments, the same applies hereinafter) can be applied in combination with the configurations disclosed in other embodiments as long as there is no contradiction, and the embodiments disclosed in this specification are illustrative, and the embodiments of the present invention are not limited thereto, and can be appropriately modified within the scope not departing from the object of the present invention.

Industrial Applicability

[0078] The corrosion evaluation method of the present invention can be effectively used as a corrosion evaluation method, a sensor current value derivation model, and a control device that accurately estimate the sensor current value correlated with the corrosion rate of a metal without actually measuring the amount of deposits such as sea salt on the metal in the environment.

Explanation of Symbols

[0079] 100: Corrosion evaluation device AC: ACM sensor E: Cumulative sensor electrical quantity R: Corrosion reduction amount RC: RCM sensor Vcosθ1: Sea wind speed Vcosθ2: Sensor wind speed

Claims

1. A corrosion evaluation method for evaluating the corrosiveness of metals in an environment, comprising: Performing a sensor current value derivation model generation step of generating a sensor current value derivation model for deriving the sensor current value by performing machine learning with the sensor current value measured by an ACM sensor in the environment as the target variable and the environmental factors affecting the sensor current value in the environment as the explanatory variables; Performing a virtual sea salt variable derivation step of deriving a virtual sea salt variable that is derived from the sea wind speed, which is the wind speed component of the wind from the sea to the environment, based on the wind direction and wind speed in the environment, and the sensor wind speed, which is the wind speed component of the wind in the direction of the metal receiving wind, based on the wind direction and wind speed in the environment, and is correlated with the amount of sea salt adhering to the metal at a predetermined measurement timing in the environment; Including, as the integrated virtual sea salt variable at the current time, the value obtained by adding up the virtual sea salt variables derived at the measurement timing up to the current time in the environmental factors of the sensor current value derivation model generation step; Performing a sensor current value estimation step of estimating the sensor current value at the non-measurement point based on the environmental factors as the explanatory variables at the non-measurement point of the sensor current value and the sensor current value derivation model derived in the sensor current value derivation model generation step. A corrosion evaluation method.

2. The corrosion evaluation method according to claim 1, wherein the virtual sea salt variable as the environmental factor is a first-order product of the sea wind speed and the sensor wind speed.

3. Obtaining the amount of electricity for the time from the two sensor current values continuously measured by the ACM sensor at a predetermined measurement timing in the environment and the time between the consecutive measurement timings, and adding up the amount of electricity for the time in a predetermined measurement period to obtain the cumulative sensor electricity amount; By performing a correlation analysis between the cumulative sensor electricity amount and the corrosion metal loss amount derived from the measurement value of the RCM sensor measured in the measurement period in the environment; Performing a conversion model generation step of generating a conversion model from the cumulative sensor electricity amount to the corrosion metal loss amount; Performing a corrosion metal loss amount estimation step of estimating the corrosion metal loss amount from the cumulative sensor electricity amount obtained by sequentially estimating the sensor current value at the non-measurement point estimated in the sensor current value estimation step over a predetermined corrosion prediction period, converting it into an amount of electricity, and adding it up, and the conversion model. The corrosion evaluation method according to claim 1 or 2.

4. The corrosion evaluation method according to claim 3, wherein in the conversion model generation step, the corrosion metal loss amount is estimated using a moving minimum value which is the lowest value among the measured values by the RCM sensor measured at the measurement timing during the measurement period in the environment over the measurement period.

5. The corrosion evaluation method according to claim 1 or 2, wherein the sensor current value derivation model is a multiple regression equation obtained by performing multiple regression analysis as the machine learning, or a regression tree obtained by performing analysis using a regression model as the machine learning.

6. The corrosion evaluation method according to claim 1 or 2, wherein in the sensor current value derivation model generation step, in addition to the environmental factors, a time factor representing the time-series change of the sensor current value is used as an explanatory variable for machine learning.

7. A sensor current value derivation model for deriving a sensor current value by performing machine learning with the sensor current value measured by an ACM sensor in an environment as an objective variable and environmental factors affecting the sensor current value in the environment as explanatory variables, derived from a sea wind speed which is a wind speed component of the wind from the sea to the environment derived based on the wind direction and wind speed in the environment, and a sensor wind speed which is a wind speed component of the wind in the direction of the metal receiving wind derived based on the wind direction and wind speed in the environment, and deriving a virtual sea salt variable correlated with the amount of sea salt adhering to the metal at a predetermined measurement timing in the environment, a sensor current value derivation model included in the environmental factors, wherein a value obtained by adding up the virtual sea salt variables derived at the measurement timing up to the current time is used as the integrated virtual sea salt variable at the current time.

8. The sensor current value derivation model according to claim 7, wherein in addition to the environmental factors, a time factor representing the time-series change of the sensor current value is used as an explanatory variable for machine learning to derive the sensor current value.

9. A control device for estimating the sensor current value at a non-measurement point based on the sensor current value derivation model according to claim 7 and the environmental factors as explanatory variables at the non-measurement point of the sensor current value.

Images