Corrosion amount prediction device

Abstract

To provide a corrosion amount prediction device capable of more quantitatively evaluating progression of rust of a rusted part from an image of the rusted part of a prediction target object.SOLUTION: A corrosion amount prediction device of the present invention is used for predicting an amount of corrosion of a rusted part of a prediction target object, and comprises a prediction unit configured to predict the amount of corrosion by acquiring the amount of corrosion from an image of the rusted part of the prediction target object using a prediction model, where the prediction model is a learning model constructed by performing machine learning using teacher data consisting of sets of images and corrosion amounts of the rusted part of the prediction target object.SELECTED DRAWING: Figure 1

Description

Technical Field

[0004] , , , , , , , , , , , , , , , , , ,

[0003]

[0001] The present invention relates to a corrosion amount prediction device that predicts the corrosion amount of a rusted part of a prediction object.

Background Art

[0002] In recent years, systems and devices have been developed to evaluate the occurrence status of rust on objects composed of steel materials and the like from images obtained by imaging the objects.

[0003] As such a system or device, for example, there is an evaluation system that evaluates the degree of rust on an evaluation object using a captured image of the evaluation object, including an image acquisition unit that acquires the captured image, a correction unit that generates an evaluation image by correcting the captured image, an evaluation unit that performs an evaluation based on the evaluation image, and an output unit that outputs the evaluation result by the evaluation unit. The correction unit extracts an evaluation region, which is an image of a range having a predetermined area on the surface of the evaluation object from the captured image, and generates an evaluation image based on the evaluation region (Patent Document 1). Also, there is known an information processing device including an acquisition unit that acquires an image obtained by imaging target power distribution equipment, an identification unit that identifies a rust region from the image acquired by the acquisition unit based on a learned first learning model of the image of the power distribution equipment and the rust region in the image, and a determination unit that determines the degree of deterioration of the target power distribution equipment from the data indicating the rust region identified by the identification unit based on a learned second learning model of the data indicating the rust region and the degree of deterioration of the power distribution equipment (Patent Document 2).

Prior Art Documents

Patent Documents

[0005] Conventional systems and devices for evaluating rust formation could not predict quantitative information such as height, which is an indicator of rust progression, from images of rusted areas on an object. Therefore, there was a need to be able to more quantitatively evaluate rust progression from images of rusted areas on an object.

[0006] This invention has been made in view of these points, and its objective is to provide a corrosion amount prediction device that can more quantitatively evaluate the progression of rust in a rusted area of ​​an object from an image of the rusted area of ​​the object to be predicted. [Means for solving the problem]

[0007] To solve the above problems, the present invention provides a corrosion amount prediction device that predicts the amount of corrosion of a rusted part of an object to be predicted, and comprises a prediction unit that predicts the amount of corrosion by obtaining the amount of corrosion from an image of the rusted part of the object to be predicted using a prediction model, and the prediction model is a learning model constructed by performing machine learning using training data which is a pair of an image of the rusted part of the object to be predicted and the amount of corrosion. [Effects of the Invention]

[0008] According to the present invention, the progression of rust in a rusted area of ​​an object to be predicted can be evaluated more quantitatively from an image of the rusted area. [Brief explanation of the drawing]

[0009] [Figure 1] (a) is a schematic diagram showing a corrosion amount prediction device according to one embodiment, and (b) is a schematic diagram showing the configuration of a computer that implements the corrosion amount prediction device according to one embodiment. [Figure 2] (a) is a flowchart of a method for constructing a corrosion amount prediction device according to one embodiment, and (b) is a flowchart of machine learning in the method for constructing a corrosion amount prediction device according to one embodiment. [Figure 3](a) and (b) are diagrams illustrating the preparation of training data in a method for constructing a corrosion amount prediction device according to one embodiment. [Figure 4] This figure illustrates a method for expanding training data in teacher data according to one embodiment. [Figure 5] (a) is a flowchart of a method for predicting the amount of corrosion of a rusted part of an object to be predicted according to one embodiment, and (b) is a flowchart of the prediction of the amount of corrosion using the method for predicting the amount of corrosion of a rusted part of an object to be predicted according to one embodiment. [Figure 6] This is a scatter plot showing the prediction error for images of 26 rusted areas that were targeted for prediction. [Modes for carrying out the invention]

[0010] The following describes embodiments of the corrosion amount prediction device of the present invention.

[0011] [One embodiment] First, an overview of the corrosion amount prediction device according to one embodiment will be explained by illustrating the corrosion amount prediction device according to one embodiment. Figure 1(a) is a schematic diagram showing the corrosion amount prediction device according to one embodiment, and Figure 1(b) is a schematic diagram showing the configuration of the computer that realizes the corrosion amount prediction device according to one embodiment. Figure 2(a) is a flowchart of the method for constructing the corrosion amount prediction device according to one embodiment, and Figure 2(b) is a flowchart of the machine learning in the method for constructing the corrosion amount prediction device according to one embodiment. Figures 3(a) and 3(b) are diagrams illustrating the preparation of training data in the method for constructing the corrosion amount prediction device according to one embodiment. Figure 4 is a diagram illustrating the method for expanding the training data in the training data according to one embodiment. Figure 5(a) is a flowchart of the method for predicting the amount of corrosion of the rusted part of the target object according to one embodiment, and Figure 5(b) is a flowchart of the corrosion amount prediction in the method for predicting the amount of corrosion of the rusted part of the target object according to one embodiment.

[0012] (Corrosion amount prediction device) As shown in Figure 1(a), a corrosion amount prediction device 1 according to one embodiment is a device that predicts the maximum height Hmax (corrosion amount) of the rusted portion of an electrodeposited coated steel sheet (object to be predicted) after corrosion. The corrosion amount prediction device 1 comprises an input unit 2, a storage unit 4, a processing unit 6 including a prediction unit 6a and a calculation unit 6b, and an output unit 8. The corrosion amount prediction device 1 is implemented, for example, by a computer 100 shown in Figure 1(b). The computer 100 includes a processing unit (computer) 110, a storage device 120, an input device 130, an output device 140, and an input / output interface (I / F) 150, etc.

[0013] The input unit 2 of the corrosion amount prediction device 1 receives information such as images of rusted areas on electrodeposited steel sheets after corrosion. The storage unit 4 stores the prediction model (learning model) 4a, as well as other data 4b, including the training data described later, images of the rusted areas, and predicted values ​​of the maximum height Hmax of the rusted areas. The prediction unit 6a of the processing unit 6 uses the prediction model 4a to predict the maximum height Hmax of the rusted areas (dependent variable) by obtaining the maximum height Hmax of the rusted areas (dependent variable) from images of the rusted areas (explanatory variable) on the electrodeposited steel sheets after corrosion. The prediction model 4a is a learning model constructed by performing machine learning using training data consisting of pairs of images of the rusted areas and the maximum height Hmax of the electrodeposited steel sheets after corrosion. The output unit 8 outputs the prediction result (predicted value of the maximum height Hmax of the rusted areas).

[0014] (Method for constructing a corrosion rate prediction device) In a method for constructing a corrosion amount prediction device according to one embodiment, first, as shown in Figure 2(a), training data is prepared. In preparing the training data, first, as shown in Figure 3(a), multiple rust areas to be targeted for machine learning are selected from an image of the surface of an electrodeposited steel sheet sample after corrosion, and images of multiple regions of the same shape (rectangles whose vertical and horizontal directions are parallel to the x and y directions of the xy plane parallel to the surface of the electrodeposited steel sheet) containing the selected multiple rust areas are cropped. Then, the images of the cropped regions are converted into multiple images with a specified pixel ratio (a pixels × b pixels) without changing the aspect ratio, and the converted multiple images are obtained as images of the rust areas of multiple sets of training data, respectively. Next, as shown in Figure 3(b), using a VR-5000 manufactured by Keyence Corporation, the height H [mm] of the rust area at each point in the xy plane parallel to the surface of the electrodeposited steel sheet is measured in the rust area imaging area of ​​the surface of the electrodeposited steel sheet where the images of the rust areas of each set of training data have been captured. Then, the maximum height H of the rust in the rust imaging region is obtained as the correct label for each set of training data, which is the maximum height Hmax [mm] of the rust. Note that the electrodeposited coated steel sheet comprises a steel sheet, a zinc protective film provided on the surface of the steel sheet, and an electrodeposited coating provided on the surface of the zinc protective film. The height H of the rust at each point in the xy plane is the height of the rust with the surface of the electrodeposited coated steel sheet (surface of the electrodeposited coating) as the reference (H=0). In this way, multiple sets of training data, which consist of an image of the rust and the maximum height Hmax (correct label) of the rust, are obtained and stored in the memory unit. This prepares multiple sets of training data.

[0015] Next, as shown in Figure 2(a), machine learning is performed using multiple sets of training data by employing ResNet (Residual Neural Networks), a type of deep learning, as a machine learning method. In machine learning, first, as shown in Figure 2(b), the training data is loaded. Specifically, multiple sets of training data stored in the memory unit are loaded. Next, the training data is split. Specifically, multiple sets of training data are split into multiple sets of training data and multiple sets of validation data. Then, as shown in Figure 2(b), learning is repeated to update the learning model for the training data, which includes multiple sets of training data and multiple sets of validation data, until the number of epochs (number of learning iterations) reaches a predetermined number.

[0016] In this process, for each epoch (one learning cycle), training is performed to train the learning model with all sets of training data in a predetermined number of iterations, as shown in Figure 2(b). During training, all sets of training data are divided into multiple minibatches, and the learning model is trained with the training data belonging to each minibatch in sequence. In training the training data belonging to each minibatch, training data augmentation, forward propagation, error calculation, backpropagation, and weight updating are performed in this order. Specifically, in training data augmentation, as shown in Figure 4, multiple new images of rust areas are generated from the rust area images (reference images) of each training data belonging to each minibatch using augmentation methods such as blurring, cropping, brightness change, horizontal flipping, vertical flipping, and sharpening. Then, by combining each of the multiple newly generated rust area images with the maximum height Hmax (ground truth label) of the rust area in the reference image training data, multiple new sets of rust area images and maximum rust area height Hmax are created, and these multiple sets are also used as training data to increase the variation of the training data and enable efficient training. In forward propagation, the prediction unit of the processing unit uses the learned model to predict the maximum height Hmax (target variable) of the rust from the rust image (explanatory variable) of each training data. In error calculation, the predicted maximum height Hmax (predicted value) from the rust image of each training data is compared with the actual maximum height Hmax (ground truth label) of the rust for that training data, and the prediction error is calculated by subtracting the ground truth label (actual value) from the predicted value. In backpropagation and weight updating, the prediction error calculated for each training data is reflected in the learned model, and parameters such as the weights of the learned model are updated to reduce the prediction error. In this training, the trained learned model is obtained by sequentially training the learned model with the training data belonging to each mini-batch for multiple mini-batches.

[0017] Next, as shown in Fig. 2(b), the trained learning model is verified using all sets of verification data. In the verification, prediction of the verification data, error calculation, and accuracy calculation are performed in this order. In the prediction of the verification data, the prediction unit of the processing unit uses the learning model to obtain the maximum height Hmax (target variable) of the rust part from the image of the rust part (explanatory variable) of each verification data, thereby predicting the maximum height Hmax of the rust part. In the error calculation, the maximum height Hmax (predicted value) of the rust part predicted from the image of the rust part of each verification data is compared with the maximum height Hmax (correct label) of the rust part of each of the verification data, and the prediction error obtained by subtracting the correct label (actual measured value) from the predicted value is calculated. In the accuracy calculation, from the calculation results of the prediction errors of all sets of verification data, MAE (Mean Absolute Error) is calculated as the accuracy index. At this time, in addition to MAE, for example, RMSE (Root Mean Squared Error), RMSE of the relative error, MAE of the relative error, etc. may also be calculated as the accuracy index.

[0018] Next, as shown in Fig. 2(b), the prediction model is updated. In the update of the prediction model, every 1 epoch (one learning), it is confirmed whether the MAE of the accuracy index becomes smaller than the MAE in the previous epoch (learning), and if it becomes smaller, the prediction model is updated with the trained learning model. As described above, learning is repeated until the number of epochs reaches a predetermined number. In the process of repeating learning, the prediction model is updated with the best trained learning model obtained so far for each epoch (learning) in which the best trained learning model is obtained. As a result, when learning is repeated until the number of epochs reaches a predetermined number, the best trained learning model until the number of epochs reaches a predetermined number is obtained as the prediction model. Machine learning is performed as described above. In machine learning, by verifying the trained learning model using verification data prepared separately from the training data, it is possible to discriminate learning (overlearning) specific to the training data, and overlearning of the prediction model is suppressed by using the trained learning model with the best accuracy index in the verification data as the prediction model.

[0019] Subsequently, as shown in Fig. 2(a), the prediction model is saved. In saving the prediction model, the prediction model (the learning model after the best training) is stored in the storage unit. Thus, a corrosion amount prediction device is constructed by the construction method according to one embodiment.

[0020] (Method for Predicting Corrosion Amount of Rusty Part of Object to be Predicted) In the method for predicting the corrosion amount of the rusty part of the object to be predicted according to one embodiment, the corrosion amount prediction device according to one embodiment is used to predict the maximum height Hmax (corrosion amount) of the rusty part of the object to be predicted on the electrodeposited coated steel sheet (object to be predicted) after corrosion. In this case, first, as shown in Fig. 5(a), an image of the rusty part is prepared. In preparing the image of the rusty part, in the image obtained by imaging the surface of the electrodeposited coated steel sheet after corrosion, the rusty part to be predicted is selected, and an image of the area including the selected rusty part is cut out. At this time, the area from which the image is cut out is a rectangle having the same shape as the area from which the image of the rusty part of the teacher data was cut out. Then, a process of converting the cut-out area image into an image with a specified pixel ratio (a pixels × b pixels) without changing the aspect ratio is performed, and the converted image is obtained as the image of the rusty part to be predicted and stored in the storage unit. Thereby, an image of the rusty part to be predicted on the electrodeposited coated steel sheet after corrosion is prepared.

[0021] [[ID=I9]] Next, as shown in Fig. 5(a), the corrosion amount is predicted. In predicting the corrosion amount, first, as shown in Fig. 5(b), an image of the rusty part is read. Specifically, the image of the rusty part to be predicted stored in the storage unit is read. Next, forward propagation is performed. Specifically, the prediction unit of the processing unit uses the prediction model (learning model) to obtain the maximum height Hmax (target variable) of the rusty part from the image of the rusty part to be predicted (explanatory variable), thereby predicting the maximum height Hmax [mm] of the rusty part to be predicted. Next, the prediction result is output. Specifically, the output unit outputs the predicted value of the maximum height Hmax of the rusty part.

[0022] Next, as shown in Fig. 5(a), the prediction result is saved. In saving the prediction result, the predicted value of the maximum height Hmax of the rusty part is stored in the storage unit. Thus, the maximum height Hmax of the rusty part to be predicted is predicted by the method for predicting the corrosion amount of the rusty part according to one embodiment.

[0023] (effect) In one embodiment of the corrosion amount prediction device, as described above, the maximum height Hmax of the rusted area can be predicted from an image of the rusted area of ​​the electrodeposited steel sheet (the object to be predicted) after corrosion, by using a prediction model constructed by machine learning with training data consisting of images of the rusted area and the amount of corrosion of the electrodeposited steel sheet after corrosion. Therefore, the progression of rust in the rusted area of ​​the electrodeposited steel sheet after corrosion can be easily detected from an image of the rusted area that has slightly occurred on the surface of the electrodeposited steel sheet, allowing for early detection of rust progression in the rusted area. Furthermore, the progression of rust in the rusted area of ​​the electrodeposited steel sheet after corrosion can be inspected without destructive testing or contact, and the inspection of rust progression in the rusted area can be performed quickly. Accordingly, according to the corrosion amount prediction device of the embodiment, similar to the above embodiment, the amount of corrosion, such as the maximum height of the rusted area, which is quantitative information that serves as an indicator of rust progression in the rusted area, can be predicted from an image of the rusted area of ​​the object to be predicted. In other words, the progression of rust in the rusted area can be evaluated more quantitatively from an image of the rusted area of ​​the object to be predicted. Furthermore, the progression of rust in the rusted areas of the target object can be easily detected from images of the slight rust that has formed on the surface of the object, allowing for early detection of rust progression. In addition, the progression of rust in the rusted areas of the target object can be inspected without destructive testing or contact, enabling rapid inspection of rust progression.

[0024] [Details of the embodiment] Next, the corrosion amount prediction device according to the embodiment will be described in more detail.

[0025] The corrosion amount prediction device is not particularly limited as long as it is a device that predicts the amount of corrosion in the rusted parts of the object to be predicted. The object to be predicted is not particularly limited as long as it is a component in which rust occurs due to the corrosion of the constituent material, and the amount of corrosion in the rusted parts can be predicted by obtaining the amount of corrosion from images of the rusted parts using a prediction model. Examples include steel plates, galvanized steel plates, electrodeposited steel plates, etc., and among these, components used in vehicles are preferred, and in particular components used on the underside of the vehicle floor are preferred. This is because rust on vehicle components, especially rust on components on the underside of the vehicle floor, greatly affects the durability and safety of the vehicle, but it is difficult to detect its progression and measure it quantitatively, and it is difficult to devise preventive measures, so the effect of the corrosion amount prediction device can be greatly obtained. Specifically, by using a prediction model, the amount of corrosion, which is a quantitative indicator of the progression of rust in the rusted parts, can be predicted from images of the rusted parts of the component, so the progression of rust in the rusted parts of the vehicle, especially components on the underside of the vehicle floor, can be detected at an early stage by rapid inspection, so appropriate maintenance and repairs can be carried out on the vehicle, and the durability of the vehicle can be improved. In other words, the corrosion prediction device makes it possible to predict the amount of corrosion from images of rusted areas, which was previously difficult, thereby enabling the management of vehicle components, especially those under the floor, and improving vehicle safety.

[0026] Galvanized steel sheet is a steel sheet comprising a steel sheet and a zinc protective film (plating film) applied to the surface of the steel sheet. Electrodeposited steel sheet is a steel sheet comprising a steel sheet, a zinc protective film applied to the surface of the steel sheet, and an electrodeposited coating applied to the surface of the zinc protective film.

[0027] The amount of corrosion in the rusted portion of the object to be predicted is not particularly limited as long as it is quantitative information that serves as an indicator of the progression of rust in the rusted portion of the object to be predicted, but examples include the maximum height of the rusted portion and the raised volume of the rusted portion. The maximum height of the rusted portion is the maximum value of the height of the rusted portion relative to the surface of the object to be predicted. For example, as in the embodiment described above, if the object to be predicted is an electrodeposited coated steel sheet, it is the maximum value of the height of the rusted portion relative to the surface of the electrodeposited coating. The raised volume of the rusted portion is the volume of the region above the surface of the object to be predicted in the raised portion of the rusted portion. For example, as in the embodiment described above, if the object to be predicted is an electrodeposited coated steel sheet, it is the volume of the region above the surface of the electrodeposited coating in the raised portion of the rusted portion.

[0028] The prediction unit of the corrosion amount prediction device is not particularly limited as long as it uses a prediction model to predict the amount of corrosion by obtaining the amount of corrosion from an image of the rusted part of the object to be predicted. The image of the rusted part of the object to be predicted is not particularly limited as long as it is an image from which the amount of corrosion of the rusted part can be obtained using the prediction model, but for example, as in the embodiment described above, an image in which the region containing the rusted part is cropped from an image of the surface of the object to be predicted.

[0029] The prediction model is not particularly limited as long as it is a learning model constructed by performing machine learning using training data consisting of images of the rusted parts of the object to be predicted and the amount of corrosion. For example, learning models constructed by performing machine learning using machine learning methods such as ResNet (Residual Neural Networks), Convolutional Neural Networks (CNNs), deep learning such as GoogleNet, SVM (Support Vector Machines), and Random Forests are examples. Among these, learning models constructed using deep learning are preferred, and learning models constructed using ResNet are particularly preferred. [Examples]

[0030] The corrosion amount prediction device according to the embodiment will be described in more detail below with reference to examples.

[0031] [Examples] As an example of a corrosion amount prediction device, an example of a corrosion amount prediction device according to the above embodiment was constructed using an example of a method for constructing a corrosion amount prediction device according to the embodiment. In this case, first, in preparing the training data, a corrosion test was conducted on a sample of electrodeposited coated steel plate simulating the underside of a vehicle floor to simulate corrosion including rust erosion, and a corroded electrodeposited coated steel plate (target object) was obtained. Then, from the corroded electrodeposited coated steel plate, training data of 196 sets of rusted area images (explanatory variables) and the maximum height Hmax [mm] of the rusted area (target variable) was prepared. Next, in machine learning, first, in the division of the training data, the 196 sets of training data were divided into 146 sets of training data and 50 sets of validation data. Next, learning was repeated on the training data including the 146 sets of training data and 50 sets of validation data until the number of epochs (number of learning iterations) reached 1000 (a predetermined number).

[0032] In this process, for each epoch (one learning cycle), the learning model was first trained by having it learn all sets of training data in a predetermined number of iterations. During training, all sets of training data were divided into multiple minibatches, and the learning model was trained with the training data belonging to each minibatch in sequence. In the learning of the training data belonging to each minibatch, the following steps were performed in this order: training data augmentation, forward propagation, error calculation, backpropagation, and weight updating. Furthermore, in the training data augmentation, multiple new images of rusted areas were generated from the rusted area image (reference image) of each training data belonging to each minibatch, using augmentation methods such as blurring, cropping, brightness change, horizontal flipping, vertical flipping, and sharpening, under the following conditions.

[0033] (Conditions for extension means) • Blurring: This is performed by randomly selecting target sets from the training data. It is performed under the conditions of kernel size 3 in the x direction, kernel size 3 in the y direction, standard deviation 3 in the x direction, and standard deviation 3 in the y direction. Smaller standard deviations in the x and y directions increase the impact on surrounding pixels and make the image smoother, so the strength of the blur is adjusted by the standard deviation. • Cropping: This process is performed by randomly selecting target sets from the training data. The image is cropped by deleting areas of 1 to 10 pixels from the vertical and horizontal edges of the reference image. The cropping area is set randomly. The aspect ratio of the cropped image is kept the same as that of the reference image. • Brightness change: This is performed on one out of every two sets of training data. The brightness of the reference image is changed within a range of 0.8 to 1.2 times. • Horizontal inversion: This is performed on one out of every two sets of training data. • Vertical inversion: This is performed on one out of every two sets of training data. • Sharpening: This process is performed by randomly selecting target sets from the training data. The edges of various color regions in the reference image are made twice as bright.

[0034] In this training process, the trained model was obtained by sequentially training the model with training data belonging to each of several mini-batches. Next, the trained model was validated using validation data from all sets. Then, the predictive model was updated. This process was repeated until the number of epochs reached 1000. Machine learning was performed in this manner. Subsequently, the predictive model (the best trained model) was stored in the memory unit.

[0035] [evaluation] An example of a method for predicting the amount of corrosion in rusted areas according to the above embodiment was implemented, and the prediction accuracy of the corrosion amount prediction device constructed in the embodiment was evaluated. In this case, when implementing the prediction method, in preparing the images of the rusted areas, images of 26 rusted areas on the electrodeposited coated steel plate (target object) of the vehicle after corrosion were prepared. Then, in predicting the amount of corrosion, for each image of the target rusted area, the prediction unit of the processing unit of the corrosion amount prediction device constructed in the embodiment used a prediction model to obtain the maximum height Hmax (dependent variable) of the rusted area from the image of the target rusted area (explanatory variable), thereby predicting the maximum height Hmax [mm] of the rusted area.

[0036] Furthermore, to evaluate the prediction accuracy, for each of the 26 rust areas to be predicted, the height H [mm] of the rust area was measured at each point in the xy plane parallel to the surface of the electrodeposited steel sheet in the rust area imaging region of the surface of the electrodeposited steel sheet where the image of each rust area was taken. The maximum value of the rust area height H was determined as the measured value [mm] of the maximum rust area height Hmax. Then, for the 26 rust areas to be predicted, the predicted value and the measured value of the maximum rust area height Hmax were compared, and the prediction error [mm] was calculated by subtracting the measured value from the predicted value. Figure 6 is a scatter plot showing the prediction error for the images of the 26 rust areas to be predicted. Furthermore, from the calculation results of all prediction errors, the following accuracy indices were calculated: R² (coefficient of determination), RMSE (root mean squared error), MAE (mean absolute error), sigma (standard deviation of absolute error), RMSE_ratio (RMSE of relative error), MAE_ratio (MAE of relative error), sigma_ratio (standard deviation of relative error), acc_thres0.025 (percentage of inference data with absolute error less than or equal to 0.025 mm), acc_thres0.05 (percentage of inference data with absolute error less than or equal to 0.05 mm), acc_thres0.1 (percentage of inference data with absolute error less than or equal to 0.1 mm), and acc_thres0.25 (percentage of inference data with absolute error less than or equal to 0.25 mm). These accuracy indices are shown in Table 1 below.

[0037] [Table 1]

[0038] Based on the results shown in Figure 6 and Table 1, it is considered that the corrosion amount prediction device constructed in the example can accurately predict the maximum height Hmax of the rusted area from images of the rusted area on the electrodeposited coated steel plate (target object) after corrosion of the vehicle. Specifically, since the MAE is approximately 0.1 mm, it is considered that the maximum height Hmax of the rusted area can be predicted with high accuracy with a prediction error of approximately 0.1 mm.

[0039] Although embodiments of the corrosion amount prediction device of the present invention have been described in detail above, the present invention is not limited to the above embodiments, and various design modifications can be made without departing from the spirit of the invention as described in the claims. [Explanation of Symbols]

[0040] 1: Corrosion amount prediction device, 2: Input unit, 4: Memory unit, 4a: Prediction model (learning model), 4b: Other data, 6: Processing unit, 6a: Prediction unit, 6b: Calculation unit, 8: Output unit

Claims

[Claim 1] A corrosion amount prediction device that predicts the amount of corrosion in the rusted part of an object to be predicted, The system includes a prediction unit that predicts the amount of corrosion of a rusted portion of an object by using a prediction model to obtain the amount of corrosion of the rusted portion from an image of the rusted portion of the object to be predicted. The prediction model is a learning model constructed by performing machine learning using training data which consists of a pair of images of the rusted portion of the object to be predicted and the amount of corrosion. The object to be predicted is an electrodeposited steel sheet after corrosion. The amount of corrosion in the rusted portion is the maximum height of the rusted portion. The corrosion amount prediction device is characterized in that the image of the rusted portion of the object to be predicted is a two-dimensional image of the region including the rusted portion that has been cut out from a two-dimensional image of the surface of the electrodeposited steel sheet after corrosion.

Images