Method for predicting corrosion shape, method for generating prediction model, method for managing metal structure, and device for predicting corrosion shape

Abstract

This prediction method involves: acquiring surface shape data and an average corrosion amount of three or more metal materials; calculating a range and a sill for each of the three or more metal materials; predicting a range, a sill, and an average corrosion amount of a metal material after the lapse of a prediction period; predicting the surface shape of the metal material by using the prediction results for the range and the sill; and predicting the corrosion shape of the metal material by subtracting the prediction result of the average corrosion amount of the metal material.

Description

Method for predicting corrosion shape, method for generating a prediction model, method for managing metal structures, and apparatus for predicting corrosion shape

[0001] This disclosure relates to a method for predicting the corrosion shape of a metallic material, a method for generating a prediction model used in the prediction method, a method for managing a metallic structure including a metallic material, and a device for predicting the corrosion shape of a metallic material.

[0002] Metal materials have long been a central material used in social infrastructure such as bridges and roads. However, in recent years, social infrastructure such as bridges and roads has been deteriorating. For example, major accidents such as the collapse of the Minneapolis Highway Bridge in 2007, the collapse of the Sasago Tunnel on the Yamanashi Chuo Expressway in 2012, and the collapse of the La Mondi Bridge in Italy in 2018 have drawn attention to the need for the development, maintenance, and renewal of social infrastructure. When developing, maintaining, and renewing metal structures as social infrastructure, it is necessary to estimate the long-term lifespan of the metal structures with high accuracy.

[0003] To predict the long-term lifespan of metal structures, it is necessary to calculate the change in the rate of corrosion of the metal materials used in the structure over time and the remaining load-bearing capacity of the metal materials. For example, the rate of corrosion of metal materials in an atmospheric corrosive environment is empirically calculated as Y = A·X Ｂ It is known that this can be expressed as a mathematical formula of the form Y = A * X. Y is the amount of corrosion of the metal material. X is the service life of the metal material. The unit of X is assumed to be years. A is a parameter that represents the amount of corrosion of the metal material in the first year. B is a parameter that represents the decay of the corrosion rate due to the formation of a rust layer on the surface of the metal material due to corrosion. The values ​​of parameters A and B change depending on the type of metal material or the atmospheric corrosion environment. In order to predict the long-term amount of corrosion of a metal material, the measurement results of the amount of corrosion of the metal material exposed to the target corrosion environment for a short period of time are used as described above: Y = A * X. Ｂ It is conceivable to apply this to a mathematical formula of the form and extrapolate it to a long period of use.

[0004] Furthermore, as described in Patent Document 1, it is known that the predicted amount of corrosion of weathering steel can be calculated using meteorological observation data, airborne salt content and sulfur oxide content, and intrinsic corrosion information relating to the components of the weathering steel, at the planned location where unpainted or painted weathering steel is to be used.

[0005] Furthermore, as described in Patent Document 2, it is known that information including weather observation data to which the weathering steel is exposed, the amount of airborne salt, and the amount of sulfur oxides can be estimated from map information or construction conditions alone, and that the amount of corrosion loss over a design service life of 100 years can be calculated based on this.

[0006] Patent No. 3909057 Patent No. 4143018

[0007] The estimated average corrosion rate of a metal material is used to calculate its remaining strength and predict the lifespan of metal structures made from it. However, even in mild environments where corrosion progresses relatively uniformly, corrosion-induced irregularities occur on the surface of the metal material. In particular, in harsh environments where the lifespan of metal structures is shortened, the surface irregularities caused by corrosion become very large. When the surface of a metal material is irregular, stress concentrates in the depressions where the remaining thickness of the metal material has decreased. As a result, failure or deformation of the metal structure is more likely to occur from these stress-concentrated depressions. In order to calculate the remaining strength of a metal material with high accuracy and predict the lifespan of metal structures made from it with high accuracy, it is necessary to be able to easily and accurately predict the shape of the irregularities caused by corrosion on the surface of the metal material.

[0008] Therefore, this disclosure aims to provide a method for predicting the corrosion shape of a metal material that can easily and accurately predict the corrosion shape of the metal material, a method for generating a prediction model used in the prediction method, a corrosion shape prediction device, and a method for managing a metal structure that can accurately predict the lifespan of a metal structure including a metal material and manage the metal structure according to its lifespan.

[0009] A method for predicting corrosion shape according to one embodiment of the present disclosure includes: obtaining a prediction model that predicts the changes due to corrosion after a prediction period, of a plurality of features including spatial features based on variogram analysis of the surface shape data of a metal material that has been exposed to a corrosive environment for a plurality of different periods, generated using surface shape data of the metal material; predicting the changes of the plurality of features after the prediction period using the prediction model; predicting the surface shape of the metal material after the prediction period using the prediction results of the changes of the plurality of features; and predicting the corrosion shape of the metal material after the prediction period from the prediction results of the surface shape of the metal material after the prediction period.

[0010] (2) In the corrosion shape prediction method described in (1) above, the plurality of feature quantities may include range and sill in variogram analysis.

[0011] (3) In the method for predicting corrosion shape described in (2) above, the plurality of feature quantities may include the average corrosion amount of the metal material.

[0012] (4) The corrosion shape prediction method described in (2) or (3) above may include predicting the surface shape of the metal material after the prediction period has elapsed as a surface roughness distribution characterized by a covariance function obtained from the predicted range and sill of the metal material after the prediction period has elapsed.

[0013] (5) The corrosion shape prediction method described in (4) above may include applying random numbers to the covariance function and adjusting the generation of the random numbers so that the predicted surface roughness distribution of the metal material after the prediction period has elapsed approaches the measured surface roughness distribution.

[0014] A method for generating a prediction model according to one embodiment of the present disclosure (6) generates a prediction model used in the corrosion shape prediction method described in any one of (1) to (5) above. The method for generating the prediction model includes: obtaining in advance surface shape data and average corrosion amount data of a metal material exposed to a corrosive environment for a plurality of different periods, for each of the data from the plurality of different periods; approximating the empirical semivariogram of the surface shape data to calculate a theoretical semivariogram from the surface shape data, and calculating the range and sill from the theoretical semivariogram; generating prediction models for the range and sill of the metal material when it is exposed to the corrosive environment using the range and sill of each of the plurality of surface shape data; and generating a prediction model for the average corrosion amount of the metal material when it is exposed to the corrosive environment using the average corrosion amount data from each of the plurality of different periods.

[0015] A method for managing a metal structure according to one embodiment of the present disclosure (7) includes calculating the remaining load-bearing capacity of a metal structure including the metal material after the prediction period has elapsed, using the corrosion shape of the metal material after the prediction period has elapsed, which is predicted by performing the corrosion shape prediction method described in any one of (1) to (5) above.

[0016] (8) The method for managing metal structures described in (7) above may include predicting the lifespan of the metal structure from the calculation results of the remaining load-bearing capacity of the metal structure after the elapsed prediction period.

[0017] A corrosion shape prediction device according to one embodiment of the present disclosure (9) comprises an acquisition unit and a calculation unit. The acquisition unit acquires a prediction model that predicts the changes due to corrosion after a prediction period has elapsed, of a plurality of feature quantities, including spatial feature quantities based on variogram analysis of the surface shape data of a metal material, which are generated using surface shape data of a metal material exposed to a corrosive environment for a plurality of different periods, and the prediction period. The calculation unit uses the prediction model to predict the changes of the plurality of feature quantities after the prediction period has elapsed, uses the prediction results of the changes of the plurality of feature quantities to predict the surface shape of the metal material after the prediction period has elapsed, and predicts the corrosion shape of the metal material after the prediction period has elapsed from the prediction results of the surface shape of the metal material after the prediction period has elapsed.

[0018] According to the method for predicting the corrosion shape of a metallic material, the method for generating a prediction model used in the prediction method, and the corrosion shape prediction device described herein, the corrosion shape of a metallic material can be predicted simply and with high accuracy. Furthermore, according to the method for managing metallic structures, the lifespan of metallic structures containing metallic materials can be predicted with high accuracy, and the metallic structures can be managed according to their lifespan.

[0019] This is a block diagram showing an example configuration of the prediction system related to this disclosure. This is a flowchart showing an example procedure for generating a prediction model and predicting corrosion shape related to this disclosure. This is a graph showing an example of a semivariogram of the surface shape of a corrosion sample. This is a graph showing an example of an aging model of range. This is a graph showing an example of an aging model of sill. This is a graph showing an example of an aging model of average corrosion amount. This is a graph showing an example of a two-dimensional map of the predicted surface shape of a metal material calculated from the predicted values ​​of range, sill, and average corrosion amount after the prediction period has elapsed. This is a flowchart showing an example procedure for setting a random number table to be applied to the prediction of the surface shape of a metal material. These are example images of corrosion morphologies predicted by various prediction models.

[0020] The following describes embodiments of the method and apparatus for predicting the corrosion shape of metallic materials, and the method for managing metallic structures containing metallic materials, based on the drawings. Each drawing is schematic and may differ from the actual one. Furthermore, the following embodiments are illustrative of an apparatus or method for embodying the technical idea of ​​this disclosure and do not limit the configuration to those described below. In other words, the technical idea of ​​this disclosure can be modified in various ways within the technical scope described in the claims.

[0021] (Outline of the method for predicting the corrosion shape of metal materials according to this disclosure) In order to accurately calculate the remaining load-bearing capacity of metal materials and accurately predict the lifespan of metal structures using metal materials, it is necessary to accurately predict the shape of the irregularities that occur on the surface of metal materials due to corrosion. As a result of diligent research to achieve the above objective, the inventors focused on the spatial autocorrelation structure of the irregularities that occur on the surface of corroded metal materials and considered predicting the shape of the irregularities by predicting the change in the spatial autocorrelation structure over time. Then, in order to clarify the spatial autocorrelation structure on the surface of corroded metal materials, they considered applying variogram analysis, which is used in spatial statistics, to three-dimensional data of the surface shape of corroded metal materials that were actually measured.

[0022] First, the inventors exposed three or more identical metal materials to a corrosive environment and measured the three-dimensional data of the corroded surface of each metal material obtained by allowing corrosion to progress on the surface of each metal material. Here, the length of exposure period for each metal material to the corrosive environment was different from that of the others. Furthermore, all other conditions were assumed to be the same for each metal material.

[0023] Next, the inventors applied variogram analysis to the three-dimensional data of the corroded surface of each metal material to calculate range and sill, which are spatial statistics in variogram analysis, and quantified the characteristics of the shape of the corroded surface of each metal material. As described above, the shape of the corroded surface of each metal material was obtained by exposing three or more identical metal materials to a corrosive environment for periods of different lengths. Therefore, range and sill values ​​representing the shape of the corroded surface of the metal material corresponding to each of the three or more different periods of time are calculated.

[0024] Next, the inventors used range and sill values ​​calculated from three-dimensional data of the shape of the corroded surface of a metal material exposed to a corrosive environment for three or more different periods of time to identify a function that approximates the relationship between the length of exposure to a corrosive environment and the values ​​of range and sill. Using the identified function, they calculated predicted values ​​of range and sill after an arbitrary prediction period had elapsed.

[0025] The inventors have found a method for calculating and visualizing the predicted range and sill values, which correspond to the spatial surface topography distribution, as predicted data for the three-dimensional shape of the corroded surface of a metal material. This is achieved by using the predicted range and sill values ​​as covariance parameters, capturing them as spatial probabilistic fluctuations, and performing inverse analysis.

[0026] Furthermore, the inventors actually measured the average corrosion amount of metal materials exposed to a corrosive environment for each of three or more different periods of time, identified a function that approximates the relationship between the length of exposure to the corrosive environment and the average corrosion amount, and used the identified function to calculate a predicted value of the average corrosion amount after an arbitrary predicted period. The inventors predicted the corrosion morphology of the metal material after an arbitrary predicted period by subtracting the predicted value of the average corrosion amount after an arbitrary predicted period from the predicted data of the three-dimensional shape of the corroded surface of the metal material after an arbitrary predicted period, in the thickness direction of the metal material.

[0027] The following describes detailed examples of the methods discovered by the inventors.

[0028] (Example of configuration of the metal material corrosion shape prediction system 1) As shown in Figure 1, the prediction system 1 according to this disclosure comprises a prediction device 10 and a measuring device 20.

[0029] <Measuring device 20> The measuring device 20 measures three-dimensional data of the shape of the corroded surface of the metal material and outputs the measurement results of the three-dimensional data to the prediction device 10. The measuring device 20 may include a laser displacement meter or a one-shot three-dimensional shape measuring machine, etc. The measuring device 20 is not limited to these examples and may include various devices capable of obtaining the shape of the corroded surface of the metal material as a numerical value.

[0030] The size of the irregularities caused by the corrosion of the surface of the metal material, that is, the depth to which the corrosion progresses, can be a value within a wide range from about 100 μm to about several tens of mm. The measuring device 20 may have a resolution in the depth direction of about 1 μm, for example, so that it can measure three-dimensional data with high precision even when the irregularities are small.

[0031] Further, the measuring device 20 measures the average corrosion amount of the metal material. The average corrosion amount of the metal material may be measured as a change in the thickness of the metal material. In this case, the measuring device 20 may be configured to measure the three-dimensional data of the front and back surfaces of the metal material in absolute coordinates and measure the difference between the absolute coordinates of the front and back surfaces as the thickness of the metal material.

[0032] The average corrosion amount of the metal material may also be measured as the difference between the weight of the metal material before exposure to the corrosion environment and the weight of the metal material after exposure to the corrosion environment, allowing the corrosion to progress and removing the rust layer to expose the base metal part. In this case, the measuring device 20 may include a load cell or the like.

[0033] The measuring device 20 may be connected to the prediction device 10 via a network such as a LAN (Local Area Network), and may be configured to output the measurement results to the prediction device 10 through the network.

[0034] <Predicting Device 10> The predicting device 10 includes an acquisition unit 15, a calculation unit 12, an output unit 13, a storage unit 14, and an input unit 11. The acquisition unit 15 acquires three-dimensional data of the shape of the corroded surface of the metal material from the measuring device 20 or another device. The acquisition unit 15 may be connected to the measuring device 20 or another device via a network such as a LAN, and may be configured to acquire the measurement results from the measuring device 20 through the network. The acquisition unit 15 may be configured to include a communication interface for communicably connecting to the measuring device 20 or another device in various communication methods, wired or wireless.

[0035] The calculation unit 12 performs various calculations to calculate and visualize predicted 3D data of the shape of the corrosion surface of a metal material after an arbitrary prediction period has elapsed, based on 3D data of the shape of the corrosion surface. The calculation unit 12 may be configured to include one or more processors. The processor may be a general-purpose processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The processor may be a dedicated processor specialized for a specific process. The processor is not limited to these and may be configured to include any processor. The processor may realize the functions of the prediction device 10 by reading and executing a program stored in the storage unit 14, which will be described later.

[0036] The storage unit 14 stores data or information used by the prediction device 10, such as three-dimensional data acquired by the acquisition unit 15. The storage unit 14 may also store programs executed by the calculation unit 12. The storage unit 14 may be configured to include one or more memories. The memories are, for example, semiconductor memories, magnetic memories, or optical memories, but are not limited to these and can be any memory. The storage unit 14 may also be configured to include an electromagnetic recording medium such as a hard disk drive (HDD).

[0037] The output unit 13 outputs the prediction data calculated by the calculation unit 12. The output unit 13 may include a display device. The display device may include various displays such as liquid crystal displays. The output unit 13 may be connected to other devices via a network such as a LAN and configured to output prediction data through the network. The output unit 13 may include a communication interface for connecting to other devices via various wired or wireless communication methods. The output unit 13 is not limited to these examples and may include various other devices or interfaces.

[0038] The input unit 11 may include an input device that receives input from an operator or the like. The input device may include, for example, a keyboard or physical keys, or may include a pointing device such as a touch panel, a touch sensor, or a mouse. The input device is not limited to these examples and may include various other devices. The input unit 11 may be connected to an external input device.

[0039] The prediction device 10 may be realized as a computer such as a desktop PC (Personal Computer), a notebook PC, or a tablet PC. The prediction device 10 may be configured to include at least one server. The prediction device 10 is not limited to these examples and may include various devices. Also, the prediction device 10 may be realized in an on-premises environment or may be realized using a cloud service.

[0040] (Operation example of the corrosion shape prediction system 1 of the metal material) In the prediction system 1 according to the present disclosure, the measurement device 20 measures three-dimensional data of the shape of the corroded surface of the metal material obtained by exposing the metal material to a corrosion environment. As a corrosion sample to be measured by the measurement device 20, the metal material is exposed to the corrosion environment for each of three or more different lengths of time to advance the surface corrosion, and the rust layer formed on the surface of the metal material is removed using a rust remover according to the type of metal, thereby preparing a metal material with the non-rusted base metal part exposed. That is, three or more corrosion samples in which each of three or more samples of the same type of metal material is exposed to the corrosion environment for different lengths of time are prepared.

[0041] The measurement device 20 measures three-dimensional data of the surface shape of the base metal part of each of three or more corrosion samples obtained by exposing the metal material to the corrosion environment for each of three or more different lengths of time and outputs it to the prediction device 10.

[0042] The surface shape data of each of three or more corrosion samples exposed to the corrosion environment for different lengths of time may be measured by another device different from the measurement device 20.

[0043] Furthermore, the measuring device 20 measures the average corrosion amount of each of the three or more corrosion samples described above and outputs the result to the prediction device 10. The average corrosion amount of the corrosion samples may be measured by a device other than the measuring device 20. The average corrosion amount of the corrosion samples may also be measured manually by an operator investigating the corrosion of metal materials.

[0044] The prediction device 10 acquires surface shape data and average corrosion amount of a corrosion sample from the measuring device 20 or other devices in the acquisition unit 15. If the average corrosion amount of a corrosion sample is measured manually by an operator, the prediction device 10 may acquire the measurement result of the average corrosion amount by accepting the manual input of the measurement result of the average corrosion amount by the operator in the input unit 11.

[0045] The prediction device 10, in its calculation unit 12, calculates feature quantities from the surface shape data of three or more corrosion samples exposed to a corrosive environment for periods of different lengths, and generates a prediction model for predicting the change in feature quantities due to corrosion after an arbitrary prediction period has elapsed, based on the relationship between the length of exposure to the corrosive environment and the change in feature quantities due to corrosion. The prediction model is configured to output predicted values ​​of the feature quantities of the surface shape data of corrosion samples exposed to the corrosive environment during an arbitrary prediction period, when an arbitrary prediction period is input. In this example, the prediction model is assumed to be a function that approximates the relationship between the length of exposure to the corrosive environment and the feature quantities. The prediction model is not limited to this example and may be any other model such as a machine learning model.

[0046] The features include spatial features based on variogram analysis of surface shape data. The spatial features may include range and sill. The prediction device 10 may calculate range and sill from the surface shape data of three or more corrosion samples and identify a function that approximates the relationship between the length of exposure to the corrosive environment and the values ​​of range and sill. The identified function may be used as a prediction model.

[0047] The prediction device 10 may pre-generate a prediction model and store it in the storage unit 14. The prediction model may be generated by a device other than the prediction device 10. The prediction device 10 may acquire a prediction model generated by another device.

[0048] The predictive model may be generated using features calculated from the surface morphology data of two corrosion samples exposed to a corrosive environment for two different periods of time. For example, if the function approximating the relationship between the length of exposure to the corrosive environment and the features is in a form that can be identified using only the features of the two periods, the predictive model can be generated using only the features of the two corrosion samples. For example, the function can be identified using only the features of the two periods by fixing its shape. Therefore, the predictive model can be generated using surface morphology data of metallic materials exposed to a corrosive environment for two or more, i.e., multiple different periods.

[0049] The prediction device 10 predicts feature quantities after an arbitrary prediction period using a prediction model. The prediction device 10 uses an identified function as the prediction model, and if the feature quantities are range and sill, it calculates predicted values ​​of range and sill after an arbitrary prediction period using the identified function. The prediction device 10 calculates the unevenness distribution corresponding to the predicted values ​​of the feature quantities after an arbitrary prediction period, i.e., the predicted values ​​of range and sill, as predicted data for the three-dimensional shape of the corrosion surface of the metal material to be predicted. The prediction device 10 predicts the corrosion shape by subtracting the predicted value of the average corrosion amount from the predicted data for the three-dimensional data. The prediction device 10 outputs the predicted data for the three-dimensional data of the corrosion surface of the metal material, or the predicted result of the corrosion shape after subtracting the predicted value of the average corrosion amount, at the output unit 13, and visualizes it by displaying it on a display device, for example.

[0050] The calculation unit 12 of the prediction device 10 may perform a method for generating a prediction model of the corrosion shape of a metal material, including an example procedure from steps S1 to S3 as illustrated in the flowchart of Figure 2, and a method for predicting the corrosion shape of a metal material, including an example procedure from steps S4 to S7, in order to predict the corrosion shape of a metal material after an arbitrary prediction period has elapsed. The method for generating the prediction model and the method for predicting the corrosion shape may be implemented as a generation program and a prediction program to be executed by a processor constituting the calculation unit 12. The generation program and the prediction program may be stored in a non-temporary computer-readable medium. The calculation unit 12 may generate a prediction model by performing a prediction model generation method in advance before executing the corrosion shape prediction method. The calculation unit 12 may also perform the corrosion shape prediction method by acquiring a prediction model generated by another device.

[0051] First, the calculation unit 12 generates a prediction model by executing a prediction model generation method that includes the steps S1 to S3. The calculation unit 12 uses the acquisition unit 15 to acquire surface shape data and average corrosion amount for each of the multiple corrosion samples, which are multiple samples of the same type of metal material exposed to a corrosive environment for periods of different lengths (step S1).

[0052] The calculation unit 12 calculates the range and sill of the surface shape data of the corroded sample by variogram analysis (step S2). The surface shape data of the corroded sample is assumed to be three-dimensional data in which the coordinate system is set so that the position on the surface of the corroded sample is represented by a grid of XY coordinates and the irregularities of the corroded sample are represented by the Z coordinate. The depth of the irregularities at one position on the surface of the corroded sample is represented by the XYZ coordinates of that position. The calculation unit 12 discretizes the data representing the irregularities in the surface shape data of the corroded sample using XY coordinates. The discretization width, i.e., the sampling period in the XY direction, is set to such an extent that an empirical semivariogram plot can be created and the range and sill can be calculated. The discretization width may be set to a range from, for example, about 10 μm to about 300 μm.

[0053] The calculation unit 12 creates a plot of the empirical semivariogram with the specified number of classes. The calculation unit 12 uses the surface shape data of the corrosion sample and Equation (1) to calculate the value of the semivariogram as the empirical semivariogram for each distance on the predetermined XY plane, and plots it as an actual value. In Equation (1), γ(h) represents the value of the semivariogram, h represents the Euclidean distance, N(h) represents all pairs of points with the Euclidean distance h, and z ｊ and z ｋ represent the corrosion depth of the points within the pair of points with the Euclidean distance h.

[0054]

[0055] The calculation unit 12 approximates the plot of the empirical semivariogram with the mathematical formula of the theoretical semivariogram, and calculates the range representing the influence range of spatial autocorrelation and the sill representing spatial dependence. As the model of the theoretical semivariogram, a spherical model, an exponential model, a Gaussian model, or the like is used. The calculation unit 12 selects a model that can appropriately approximate the plot of the empirical semivariogram and adjusts the parameters of the model.

[0056] The calculation unit 12 may select, for example, the spherical model shown in Equation (2) as the model of the theoretical semivariogram. In Equation (2), θ represents the adjustment parameter, WRSS(θ) represents the weighted least squares criterion, |N(h ｋ )| represents the Euclidean distance, that is, the total number of pairs of points with the lag h ｋ , K represents the number of lags of the variogram, γ(h ｋ ) represents the value of the empirical semivariogram at the lag k, and γ(h ｋ ; θ) represents the value of the theoretical semivariogram defined as the spherical model at the lag k.

[0057]

[0058] The calculation unit 12 determines the adjustment parameter θ in the spherical model so as to minimize the evaluation function represented by Equation (3). In Equation (3), θ １ represents the volume of the sphere, and θ ２represents an adjustment parameter that generalizes the diameter of the sphere. As a calculation algorithm to minimize the evaluation function, the least squares method (LS), the least absolute residual method (LAR), or the double square approximation method may be appropriately selected.

[0059]

[0060] The calculation unit 12 applies the determined adjustment parameter θ to the theoretical semivariogram model and calculates range and sill as spatial autocorrelation statistics. The calculation unit 12 may also calculate nuggets as statistics, which represent the measurement error of the 3D data, etc.

[0061] The range calculated from the 3D data of the corroded surface represents the range affected by the corrosion depth. The sill represents the degree of corrosion depth.

[0062] The calculation unit 12 may select a model such that the error between the plotted empirical semivariogram and the approximating model is minimized. The calculation unit 12 may calculate the error for each of the multiple models using equation (2) and select the model that minimizes the error. The model may also be selected by the operator. The operator may visually inspect the plotted empirical semivariogram and determine and select an appropriate model as a theoretical semivariogram. The calculation unit 12 may accept model selection input from the operator at the input unit 11.

[0063] The calculation unit 12 may approximate the plot of the empirical semivariogram, represented by a dashed line, with the theoretical semivariogram, represented by a solid line, as shown in the graph of Figure 3, for example. The calculation unit 12 may calculate the range and sill from the formula obtained by approximating the empirical semivariogram with the theoretical semivariogram.

[0064] Returning to the flowchart in Figure 2, the calculation unit 12 generates a prediction model for the range, sill, and average corrosion rate of the metallic material (step S3). The prediction model for the range, sill, and average corrosion rate of the metallic material is configured to output predicted values ​​for the range, sill, and average corrosion rate of the metallic material after the length of time the metallic material has been exposed to a corrosive environment has been input. The period input to the prediction model is also called the prediction period. The prediction model is configured to output predicted values ​​for the range, sill, and average corrosion rate of the metallic material after the prediction period has elapsed, when the prediction period has been input.

[0065] The prediction model is generated separately for range, sill, and mean corrosion rate. The prediction model may be generated to predict the aging changes of range and sill together. The prediction model may be generated to predict the aging changes of range or sill and the aging changes of mean corrosion rate together. The prediction model may be generated to predict the aging changes of range, sill, and mean corrosion rate together. The prediction model may be configured to predict changes in the surface morphological features of the metallic material due to corrosion.

[0066] The calculation unit 12 may identify a function that represents the relationship between the length of time the metal material has been exposed to a corrosive environment and the range, sill, and average corrosion amount of the metal material as the corrosion progresses, as a prediction model for the range, sill, and average corrosion amount of the metal material.

[0067] The calculation unit 12 may identify a function using multiple measured values ​​of range, sill, and average corrosion amount obtained by measuring corrosion samples obtained by actually exposing a metal material to a corrosive environment over several different periods of time, i.e., multiple measured values ​​of range, sill, and average corrosion amount. The calculation unit 12 may approximate the multiple measured values ​​of range, sill, and average corrosion amount of the metal material in any manner such as linear approximation, exponential approximation, or power approximation, and identify a function.

[0068] In this function, the range, sill, and average corrosion values ​​for a period longer than the actual period during which the metal material was exposed to the corrosive environment to obtain the corrosion sample correspond to data extrapolated from the measured range, sill, and average corrosion data. The calculation unit 12 may generate a predictive model for the range, sill, and average corrosion of a metal material by using the identified function to extrapolate the graphs of range, sill, and average corrosion to a period longer than the actual period during which the metal material was exposed to the corrosive environment to obtain the corrosion sample.

[0069] The calculation unit 12 uses the range and sill values ​​of multiple corrosion samples exposed to a corrosive environment for multiple periods of different lengths to approximate the relationship between the length of time a metal material is exposed to a corrosive environment and the range and sill of the three-dimensional data of the shape of the corrosion surface of the metal material as a function y = a・x ｂ It may be identified in the form of . Furthermore, the calculation unit 12 uses the average corrosion amount values ​​corresponding to each of several different length periods to approximate the relationship between the length of time the metal material is exposed to the corrosive environment and the average corrosion amount of the metal material as a function y = a・x ｂ It may be identified in this format.

[0070] As mentioned above, the form of the function to be identified is y = a・x ｂ The form is not limited to this, and may be any form capable of expressing changes in range, sill, and mean corrosion rate over time. The form of the function to be identified may be, for example, a logarithmic function or a polynomial.

[0071] The calculation unit 12 can generate a predictive model that predicts the characteristic quantities of the surface shape of a metal material due to corrosion, namely range, sill, or average corrosion amount, by executing the method for generating a predictive model that includes the steps S1 to S3 described above.

[0072] Next, the calculation unit 12 performs a corrosion shape prediction method including the steps S4 to S7 to predict the corrosion shape of the metal material after an arbitrary prediction period has elapsed. The calculation unit 12 acquires the prediction period (step S4). The calculation unit 12 may acquire the prediction period by receiving input of the prediction period from an operator or the like at the input unit 11. The calculation unit 12 may also acquire the prediction period from an external device at the acquisition unit 15. The calculation unit 12 predicts the range, sill, and average corrosion amount of the metal material after the prediction period has elapsed (step S5). The calculation unit 12 uses the model representing the aging changes of the range, sill, and average corrosion amount generated in step S3 to calculate the predicted values ​​of the range, sill, and average corrosion amount after an arbitrary period has elapsed as the prediction period.

[0073] The calculation unit 12 predicts the surface shape of the metal material after the prediction period using the predicted range and sill values ​​after the prediction period has elapsed (step S6). The predicted range and sill values ​​of the metal material after the prediction period, obtained in step S5, reflect the statistical characteristics of the surface irregularities of the metal material after the prediction period has elapsed. The calculation unit 12 calculates the covariance parameter by associating the predicted range and sill values ​​with the covariance function.

[0074] The calculation unit 12 predicts the surface shape of a metallic material having a distribution of surface irregularities characterized by a covariance parameter, treating it as a Gaussian random field. A Gaussian random field is a type of random field and is applied to the analysis or representation of surface irregularities of a plane in three-dimensional space in the field of spatial statistics, such as geostatistics, for the purpose of spatial data analysis. In this disclosure, the Gaussian random field is applied to the analysis of the distribution of surface irregularities of a metallic material.

[0075] A Gaussian random field is generated such that the distribution of surface irregularities due to corrosion of a metallic material is represented as a Gaussian or normal distribution characterized by a specified covariance function, thereby representing the distribution of heights on a plane. Known methods may be used as the algorithm for generating the Gaussian random field. For example, libraries or functions corresponding to various software development environments or programming languages ​​may be used as the Gaussian random field generation algorithm.

[0076] Here, depending on the functions provided in the software environment used, the change in height distribution may be regular. On the other hand, the distribution of surface irregularities on actual corroded metal materials is irregular. Therefore, irregularity in the change in height distribution may be achieved by using random noise random numbers for each height data. Alternatively, a library or function that generates a Gaussian random field, including random number generation, may be used. Irregularity may be adjusted by adjusting the pattern of the generated random numbers.

[0077] The calculation unit 12 may calculate three-dimensional data predicting the surface shape of a metal material by setting an arbitrary random number table from the parameters of a normal random field obtained by specifying the predicted values ​​of range and sill, the number of data points to be used to predict the surface shape, the data size in the width and height directions of the range to be predicted for the surface shape, and the data size in the depth direction of corrosion, and generating a normal random field using an arbitrary covariance function. Note that the random number table may be created separately and entered by a person, or it may be generated by an internal function by providing the number of occurrences and a seed.

[0078] The calculation unit 12 predicts the corrosion shape of the metal material by subtracting the average corrosion amount from the surface shape of the metal material (step S7). The surface shape of the metal material predicted in step S6 is a shape that represents the relative height difference between any point on the surface of the metal material and other points, and does not include information on the depth to which corrosion has progressed across the entire surface of the metal material. The depth to which corrosion has progressed across the entire surface of the metal material is represented as the average corrosion amount. The calculation unit 12 predicts the corrosion shape of the metal material by subtracting the average corrosion amount of the metal material after the predicted period predicted in step S5 from the surface shape of the metal material after the predicted period predicted in step S6. The corrosion shape of the metal material is a shape that reflects not only the surface shape after corrosion, but also the thickness of the metal material that has decreased as corrosion has progressed in the depth direction of the metal material.

[0079] After executing the procedure in step S7, the calculation unit 12 completes the execution of the procedure in the flowchart in Figure 2. By executing the corrosion shape prediction method including the procedures from steps S4 to S7 described above, the calculation unit 12 can predict the corrosion shape of the metal material after an arbitrary prediction period has elapsed using the prediction model.

[0080] Following the procedure in step S7, the calculation unit 12 may display the prediction results on the display device of the output unit 13. The calculation unit 12 may output the prediction results from the output unit 13 to another device. The calculation unit 12 may use the prediction results to perform a strength calculation of the metal material after the prediction period has elapsed. The calculation unit 12 may display the results of the strength calculation on the display device of the output unit 13. The calculation unit 12 may output the results of the strength calculation from the output unit 13 to another device.

[0081] (Example) As an example relating to the present disclosure, suppose that corrosion samples exposed to a corrosive environment for four different periods of time are prepared.

[0082] In this embodiment, the corrosive environment was set to an ambient temperature of 23.4°C (assuming an average annual temperature), a relative humidity of 76.7%, and an airborne salt content of 0.21 mg / dm³. ２ Set to per day, sulfur oxides (SO ２ ) The amount is 0.03 mg / dm ２ The unit for airborne salt and sulfur oxide content is expressed as the amount of substance in milligrams per 10 cm square area per day, and is abbreviated as mdd. The four different length periods were set to 1 year, 2 years, 3 years, and 5 years.

[0083] As a sample of metal material exposed to a corrosive environment, a piece of carbon steel with dimensions of 150 mm x 70 mm x 3 mm was used.

[0084] As a pretreatment for measuring the three-dimensional shape of the surface of a corroded sample, HCl:H as specified in ISO 8407 is used. ２ Rust formed on the surface of the corroded sample was removed using an acid pickling solution prepared by adding 3.5 g / L of hexamethylenetetramine as an inhibitor to an O=1:1 solution.

[0085] After removing the rust from the surface of the corroded sample, the three-dimensional shape of a 30 mm x 30 mm area in the center of the surface of the corroded sample was measured using a laser displacement meter at a 100 μm pitch in both the X and Y directions, with a depth resolution of 1 μm.

[0086] In this embodiment, an exponential model was selected as a model to approximate the empirical semivariograms generated from surface morphology data of corrosion samples over four different time periods with a theoretical semivariogram. The range and sill values ​​in the selected exponential model were calculated for each of the four different time periods.

[0087] As described above, the function that approximates the relationship between the ranges corresponding to each of the four different period lengths and the length of each period is shown as a solid line in the graph of Figure 4A, y = a・x, as a model of the range's change over time. ｂ The function was identified in the following format. The identified function outputs the predicted range value as y when the length of the prediction period is input as x. In the graph in Figure 4A, the x axis represents the prediction period and the y axis represents the predicted range value. In the identified function, the coefficient a was 37.284 and the coefficient b was 0.1489. Then, by inputting 50 years as the prediction period for x in the function, the predicted range value after 50 years was calculated as 66.8 μm as the value of y in the function.

[0088] As described above, the function that approximates the relationship between the value of Sil raised to the power of 0.5 corresponding to each of the four different lengths of periods calculated, and the length of each period, is y = a・x, which is shown as a solid line in the graph of Figure 4B, representing Sil's aging model. ｂThe function was identified in the form of . The identified function outputs a predicted value of Sil to the power of 0.5 as y when the length of the prediction period is input as x. In the graph in Figure 4B, the x axis represents the prediction period and the y axis represents the predicted value of Sil to the power of 0.5. In the identified function, the coefficient a was 22.398 and the coefficient b was 0.4844. Then, by inputting 50 years as the prediction period for x in the function, the predicted value of Sil to the power of 0.5 after 50 years was calculated as 149 μm as the value of y in the function. The predicted value of Sil can be easily calculated by squaring the predicted value of Sil to the power of 0.5.

[0089] Using predicted range and sill values ​​after 50 years, the surface shape of the metal material after corrosion progressed over 50 years was predicted. In this embodiment, the predicted surface shape of the metal material was calculated using the grf function included in geoR, an open-source library for the statistical analysis language R. The parameters used for prediction were set as follows: ・n: number of data points; in this embodiment, n was set to 90,000 points (300 x 300). ・nx: data size in the width direction; in this embodiment, nx was set to 300 points. ・ny: data size in the height direction; in this embodiment, ny was set to 300 points. ・cov.model: parameter specifying the variogram model; in this embodiment, cov.model was set to exponential, representing an exponential model. ・cov.pars: specification of variogram parameters; in this embodiment, cov.pars was set to sill and range. - mean: The average value of the simulation results. In this example, mean was used to set the zero point. Also, mean was set as the predicted value of the average corrosion amount.

[0090] Using the aforementioned grf function and the predicted range and sill values ​​for 50 years from now, and further processed through a random number table, three-dimensional data predicting the surface shape of the metal material 50 years from now was calculated.

[0091] The average corrosion amount of each of the four different period lengths corresponding to the corrosion samples was calculated from the weight change before and after rust removal. A function approximating the relationship between the average corrosion amount corresponding to each of the four different period lengths and the length of each period is shown as a model of the change in average corrosion amount over time, as shown by the solid line in the graph of Figure 5: y = a・x ｂ The function was identified in the form of . The identified function outputs a predicted average corrosion rate as y when the length of the prediction period is input as x. In the graph in Figure 5, the x axis represents the prediction period and the y axis represents the average corrosion rate. In the identified function, the coefficient a was 71.365 and the coefficient b was 0.5773. By inputting 50 years as the prediction period for x in the function, the predicted average corrosion rate after 50 years was calculated as 683 μm as the value of y in the function.

[0092] As described above, the 3D data of the corrosion shape was generated as a 2D map representing the corrosion amount in grayscale, as shown in Figure 6. This was obtained by subtracting the predicted average corrosion amount of 683 μm after 50 years from the 3D data predicting the surface shape of the metal material after 50 years. The less corrosion there is at each point on the surface of the metal material, the closer each point is to white. Conversely, the more corrosion there is at each point on the surface of the metal material, the closer each point is to black.

[0093] (Example of setting a random number table) As described above, when predicting the surface shape of a metal material after corrosion from predicted range and sill values, random numbers may be used to realize irregularity in the change in height distribution. The calculation unit 12 may adjust the generation of random numbers by setting a random number table. The calculation unit 12 may set a random number table by executing the example procedure illustrated in the flowchart of Figure 7.

[0094] The calculation unit 12 obtains predicted values ​​for the range and sill of the metal material after an arbitrary prediction period has elapsed, calculated in the procedure of step S5 in Figure 2 (step S11). The calculation unit 12 sets up a random number table (step S12). The calculation unit 12 applies the set random number table to predict the surface shape of the metal material after an arbitrary prediction period has elapsed from the predicted values ​​for the range and sill (step S13).

[0095] The calculation unit 12 determines whether the comparison between the predicted surface shape and the actual measurement satisfies the conditions (step S14). The calculation unit 12 uses the result predicted in the procedure of step S13 as the surface shape prediction. The calculation unit 12 may use the 3D surface data of the corrosion sample exposed to the corrosive environment for one of several different period lengths, acquired in the procedure of step S1 in Figure 2, as the actual surface shape measurement. The calculation unit 12 may also use the 3D surface data of the corrosion sample exposed to the corrosive environment for the longest period among several different period lengths as the actual surface shape measurement. The calculation unit 12 may also acquire and use 3D surface data of samples other than the corrosion sample used to predict range, sill, and average corrosion amount as the actual surface shape measurement.

[0096] The calculation unit 12 may set average value judgment or maximum value judgment as a comparison condition between the predicted surface shape and the measured surface shape. Specifically, in order to compare the predicted surface shape and the measured surface shape, the calculation unit 12 may calculate the difference between the height of the irregularities at each position in the XY direction along the surface of the metal material in the 3D data of the predicted surface shape and the height of the irregularities at each position in the XY direction in the 3D data of the measured surface shape. The calculation unit 12 may make a determination based on the condition that the sum of the absolute values, average value, or maximum value of the difference between the predicted surface shape and the measured surface shape in terms of the height of the irregularities at each position in the XY direction is less than a determination threshold.

[0097] The calculation unit 12 may set the frequency distribution of surface irregularity depth as a comparison condition between the predicted and measured surface shape.

[0098] The comparison between the predicted surface shape and the actual measurement may be determined by the operator's visual inspection. The calculation unit 12 may display the 3D data of the predicted surface shape and the 3D data of the actual measurement of the surface shape as a 3D map on the display device of the output unit 13. The operator may compare the predicted surface shape and the actual measurement by visually inspecting the 3D map and determine whether the predicted surface shape is close to the actual measurement. The calculation unit 12 may accept the operator's determination result as input at the input unit 11.

[0099] The calculation unit 12 is not limited to these examples, and may set various other comparison conditions.

[0100] If the comparison between the predicted and measured surface shape does not satisfy the conditions (step S14: NO), the calculation unit 12 returns to the procedure in step S12 to reset the random number table and re-predicts the surface shape in the procedure in step S13. The calculation unit 12 repeats the execution of the procedures from steps S12 to S14 until the comparison between the predicted and measured surface shape satisfies the conditions.

[0101] If the comparison between the predicted surface shape and the actual measurement satisfies the conditions (step S14: YES), the calculation unit 12 determines that the random number table has been set so that the predicted surface shape approaches the actual measurement, and terminates the execution of the steps in the flowchart of Figure 7. When the calculation unit 12 predicts the surface shape of the metal material after the prediction period has elapsed in step S6 of the flowchart of Figure 2, it may apply the random number table set according to the steps in the flowchart of Figure 7.

[0102] In the example described above, the comparison between predicted and measured surface shape was reflected in the setting of the random number table. The comparison between predicted and measured surface shape may also be reflected in the selection of a model that approximates the empirical semivariogram of the surface shape data. The model that approximates the empirical semivariogram may be selected so as to minimize the difference between predicted and measured surface shape. Furthermore, the comparison between predicted and measured surface shape may be reflected in the aging models of range and sill. The aging models of range and sill may be generated so as to minimize the difference between predicted and measured surface shape.

[0103] (Example 2) Example 2 explains that the predicted corrosion morphology differs for each variogram model. In Example 2, the surface shape of a metal material was calculated using the grf function included in geoR, a library openly available for the statistical analysis language R. The parameters used for prediction were set as follows: ・n: number of data points; in Example 2, n was set to 90,000 points (300 x 300). ・nx: data size in the width direction; in Example 2, nx was set to 300 points. ・ny: data size in the height direction; in Example 2, ny was set to 300 points. ・cov. model: parameter specifying the variogram model; in Example 2, cov. The model was set to one of 13 types: cauchy model, circular model, cubic model, exponential model, gaussian model, gencauchy model, generating model, material model, power model, powered exponential model, pure nugget model, spatial model, and wave model. ・cov. pars: Specifies the parameters of the variogram. In Example 2, cov. pars was set to sill and range. The value of sill was set to 100 and the value of range was set to 50. ・mean: Average value of the simulation results. In Example 2, mean was used to set the zero point. Mean was also set to the predicted value of the average corrosion amount.

[0104] Figure 8 shows example images of corrosion morphologies predicted by various models, i.e., the 13 models described above. It can be seen that the predicted corrosion morphologies differ significantly depending on the model. Therefore, in addition to fitting empirical and theoretical variograms, a subjective judgment of the output images is also necessary.

[0105] (Summary) As described above, according to the prediction device 10 and prediction method of this disclosure, the characteristic quantities of the corrosion shape of a metal material after the elapsed of an arbitrary prediction period are predicted as, for example, range and sill. Then, by inverse analysis from the predicted values ​​of characteristic quantities such as range and sill, the corrosion shape of the metal material after the elapsed of an arbitrary prediction period is predicted. In this way, the shape of the surface irregularities of the corrosion after the elapsed of an arbitrary prediction period is predicted without performing a simulation of the progression of corrosion in the metal material. As a result, the corrosion shape of the metal material can be predicted simply and with high accuracy.

[0106] (Example of a method for managing metal structures containing metal materials) Prediction results of the aging changes in the corrosion state of metal materials may be used to manage metal structures containing metal materials. For example, prediction results of the corrosion shape of metal materials after the prediction period may be used to calculate the remaining strength of metal materials after the prediction period. In other words, a method for managing metal structures may include a procedure for calculating the remaining strength of metal materials after the prediction period using prediction results of the corrosion shape of metal materials after the prediction period. The calculation of the remaining strength of metal materials may be performed by the calculation unit 12 of the prediction device 10. As a method for calculating the remaining strength of metal materials, for example, the elastoplastic finite displacement analysis method, which is a method for analyzing a composite nonlinear problem that takes into account both material nonlinearity and geometric nonlinearity, may be used, but is not limited to this.

[0107] The prediction device 10 and prediction method described herein predict the shape of the irregularities on the corroded surface of a metal material. When a metal material has irregularities on its surface, stress concentrates in the recesses where the remaining thickness of the metal material is thin. As a result, fracture or deformation of the metal structure is more likely to occur from the recesses where stress is concentrated. Therefore, by predicting the shape of the irregularities on the corroded surface of the metal material with high accuracy, the accuracy of calculating the remaining strength of the metal material can be improved. The calculation results of the remaining strength of the metal material can be used to reduce investment in metal structures by designing the corrosion allowance using quantitative data on the amount of corrosion of the metal material, i.e., by designing the margin.

[0108] The remaining yield strength of a metallic material may be used to predict the lifespan of a metallic structure containing the metallic material. In other words, a method for managing metallic structures may include a procedure for predicting the lifespan of a metallic structure containing the metallic material using the calculation result of the remaining yield strength of the metallic material. The prediction of the lifespan of the metallic structure may be performed by a calculation unit 12. For example, the calculation unit 12 may predict the change in the remaining yield strength of the metallic material over time and predict the number of years until the predicted value of the remaining yield strength falls below a judgment threshold as the lifespan of the metallic structure containing the metallic material. By improving the calculation accuracy of the remaining yield strength of the metallic material, the accuracy of predicting the lifespan of a metallic structure containing the metallic material can be improved.

[0109] Predicting the lifespan of metal structures is useful as a basis for making decisions when carrying out inspection or repair work on these structures. Improving the accuracy of lifespan predictions for metal structures can lead to more efficient planning of inspection and repair work. This efficient planning can reduce work costs. Furthermore, it can help prevent accidents or disasters such as equipment failures involving metal structures or building collapses.

[0110] While embodiments of this disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art can make various modifications or alterations based on this disclosure. Therefore, it should be noted that these modifications or alterations are included within the scope of this disclosure. For example, the functions included in each component or step can be rearranged in a logically consistent manner, and multiple components or steps can be combined into one or divided. Embodiments relating to this disclosure can also be realized as programs executed by a processor in the device or as storage media recording such programs. These should also be understood to be included within the scope of this disclosure.

[0111] 1 Prediction system 10 Prediction device (11: input unit, 12: calculation unit, 13: output unit, 14: storage unit, 15: acquisition unit) 20 Measurement device

Claims

1. A method for predicting corrosion shape, comprising: obtaining a prediction model that predicts the changes due to corrosion after a prediction period has elapsed, of a plurality of features, including spatial features based on variogram analysis of the surface shape data of a metal material, which is generated using surface shape data of a metal material exposed to a corrosive environment over multiple different periods; the prediction period; predicting the changes of the plurality of features after the prediction period has elapsed using the prediction model; predicting the surface shape of the metal material after the prediction period has elapsed using the prediction results of the changes of the plurality of features; and predicting the corrosion shape of the metal material after the prediction period has elapsed from the prediction results of the surface shape of the metal material after the prediction period has elapsed.

2. The method for predicting corrosion shape according to claim 1, wherein the plurality of features include range and sill in variogram analysis.

3. The method for predicting corrosion shape according to claim 2, wherein the plurality of feature quantities include the average corrosion amount of the metal material.

4. A method for predicting corrosion shape according to claim 2 or 3, comprising predicting the surface shape of the metal material after the prediction period has elapsed as a surface roughness distribution characterized by a covariance function obtained from the predicted range and sill of the metal material after the prediction period has elapsed.

5. A method for predicting corrosion shape according to claim 4, comprising applying random numbers to the covariance function and adjusting the generation of the random numbers so that the predicted surface roughness distribution of the metal material after the prediction period approaches the measured surface roughness distribution.

6. A method for generating a prediction model used in the corrosion shape prediction method according to any one of claims 1 to 5, comprising: acquiring in advance surface shape data and average corrosion amount data of a metal material exposed to a corrosive environment for a plurality of different periods, for each of the data for the plurality of different periods; approximating the empirical semivariogram of the surface shape data from each of the data for the plurality of different periods to calculate a theoretical semivariogram, and calculating the range and sill from the theoretical semivariogram; generating prediction models for the range and sill of the metal material when it is exposed to the corrosive environment using the respective range and sill of the plurality of surface shape data; and generating a prediction model for the average corrosion amount of the metal material when it is exposed to the corrosive environment using the average corrosion amount data for each of the plurality of different periods.

7. A method for managing a metal structure, comprising calculating the remaining load-bearing capacity of a metal structure including the metal material after the prediction period has elapsed, using the corrosion shape of the metal material after the prediction period has elapsed, which is predicted by performing the corrosion shape prediction method described in any one of claims 1 to 5.

8. A method for managing a metal structure according to claim 7, comprising predicting the lifespan of the metal structure from the calculation results of the remaining load-bearing capacity of the metal structure after the elapsed prediction period.

9. A corrosion shape prediction device comprising an acquisition unit and a calculation unit, wherein the acquisition unit acquires a prediction model that predicts the changes due to corrosion after a prediction period, of a plurality of feature quantities, including spatial feature quantities based on variogram analysis of the surface shape data of a metal material, which is generated using surface shape data of a metal material exposed to a corrosive environment for a plurality of different periods, and the prediction period; the calculation unit predicts the changes of the plurality of feature quantities after the prediction period using the prediction model; predicts the surface shape of the metal material after the prediction period using the prediction results of the changes of the plurality of feature quantities; and predicts the corrosion shape of the metal material after the prediction period from the prediction results of the surface shape of the metal material after the prediction period.

Images