Machine learning based actinium alloy corrosion behavior prediction method, device and storage medium

Abstract

The application discloses an actinium series alloy corrosion behavior prediction method and device based on machine learning and a storage medium, and comprises the following steps: acquiring a corrosion morphology identification model and a corrosion behavior prediction model based on machine learning; inputting each actinium series alloy corrosion image to be identified into the corrosion morphology identification model for identification processing; determining corrosion morphology identification time series data according to a processing result; inputting the corrosion morphology identification time series data into the corrosion behavior prediction model for prediction processing; and determining actinium series alloy corrosion behavior prediction data according to a processing result. The application can predict the corrosion behavior of the actinium series alloy according to the image to be identified of the actinium series alloy, provides a more comprehensive and reliable tool and technical means for solving the difficulty in studying the corrosion behavior of the actinium series alloy and analyzing the corrosion characteristics and rules of the actinium series alloy, and overcomes the defect that the accuracy is not high when the existing technology is used for studying the corrosion behavior of the actinium series alloy. The application is widely applied to the technical field of metal corrosion behavior prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of metal corrosion behavior prediction, and in particular to a method, device and storage medium for predicting the corrosion behavior of actinide alloys based on machine learning. Background Art

[0002] Actinide metals refer to the 15 elements in the periodic table with atomic numbers 89 to 103, including actinium, thorium, protactinium, uranium, neptunium, plutonium, americium, curium, berkelium, californium, einsteinium, fermium, mendelevium, nobelium, and lawrencium. Actinide metals are a key strategic resource, demonstrating unique advantages and playing an indispensable role in key areas such as nuclear energy development, space exploration, and medical applications.

[0003] Actinide alloys are alloys containing at least one actinide metal. Like other metals, actinide alloys are subject to the risk of corrosion due to chemical reactions with substances in the environment. While some research has been conducted on corrosion of common metal products, due to the strong radioactivity of actinide metals, the long physical experiment cycles and high costs make it difficult to use traditional metal corrosion research methods to investigate the corrosion behavior of actinide alloys, establish evaluation methods, and predict their corrosion behavior. Therefore, there is an urgent need to develop technologies for identifying and predicting the corrosion behavior of actinide alloys to address the difficulties in studying the corrosion behavior of actinide alloys and to provide important tools and technical means to analyze the corrosion characteristics and patterns of actinide alloys. Summary of the Invention

[0004] In view of the current lack of technology for identifying and predicting the corrosion behavior of actinide alloys, the purpose of the present invention is to provide a method, device and storage medium for predicting the corrosion behavior of actinide alloys based on machine learning.

[0005] In one aspect, an embodiment of the present invention includes a method for predicting the corrosion behavior of actinide alloys based on machine learning, the method comprising the following steps:

[0006] Acquire multiple actinide alloy corrosion images to be identified;

[0007] Obtaining a corrosion morphology recognition model and a corrosion behavior prediction model based on machine learning; the corrosion morphology recognition model and the corrosion behavior prediction model are trained on images of uranium-niobium alloy corrosion samples;

[0008] Inputting each of the actinide alloy corrosion images to be identified into the corrosion morphology recognition model for identification processing, and determining corrosion morphology recognition time series data according to the processing results of the corrosion morphology recognition model;

[0009] The corrosion morphology recognition time series data is input into the corrosion behavior prediction model for prediction processing, and the actinide alloy corrosion behavior prediction data is determined according to the processing result of the corrosion behavior prediction model.

[0010] Furthermore, the steps of inputting each of the actinide alloy corrosion images to be identified into the corrosion morphology recognition model for identification processing, and determining the corrosion morphology recognition time series data according to the processing results of the corrosion morphology recognition model include:

[0011] For any of the actinide alloy corrosion images to be identified, input the actinide alloy corrosion image to be identified into the corrosion morphology recognition model for identification processing, obtain corrosion area data output by the corrosion morphology recognition model, and mark the corrosion area data with the acquisition time of the actinide alloy corrosion image to be identified;

[0012] Sorting the corrosion area data according to their respective corresponding collection times;

[0013] The corrosion morphology recognition time series data is determined according to the sorting result.

[0014] Furthermore, determining the corrosion morphology recognition time series data according to the sorting result includes:

[0015] Using the sorted corrosion area data as the corrosion morphology recognition time series data;

[0016] or

[0017] Determining a plurality of corrosion rate data according to the sorted corrosion area data;

[0018] Using each of the corrosion rate data as the corrosion morphology identification time series data;

[0019] or

[0020] According to the metal corrosion rating standard, the sorted corrosion area data are converted into corresponding corrosion grade data;

[0021] The corrosion level data are used as the corrosion morphology recognition time series data.

[0022] Furthermore, the method for predicting the corrosion behavior of actinide alloys based on machine learning also includes:

[0023] performing main color analysis on each of the actinide alloy corrosion images to be identified, respectively, to obtain color distribution information of each of the actinide alloy corrosion images to be identified;

[0024] Marking the color distribution information according to the acquisition time of the actinide alloy corrosion image to be identified;

[0025] sorting the color distribution information according to the collection time corresponding to each of the color distribution information;

[0026] The corrosion color recognition time series data is determined according to the sorting results.

[0027] Furthermore, the method for predicting the corrosion behavior of actinide alloys based on machine learning also includes:

[0028] Establish corrosion morphology recognition model and corrosion behavior prediction model;

[0029] Acquire several images of uranium-niobium alloy corrosion samples;

[0030] labeling the uranium-niobium alloy corrosion sample image according to the pixel distribution corresponding to the corrosion area and the non-corrosion area in the uranium-niobium alloy corrosion sample image to obtain label data corresponding to the uranium-niobium alloy corrosion sample image;

[0031] The corrosion morphology recognition model and the corrosion behavior prediction model are trained using the uranium-niobium alloy corrosion sample images and the corresponding label data.

[0032] Furthermore, the establishment of the corrosion morphology recognition model and the corrosion behavior prediction model includes:

[0033] A DeeplabV3+ model based on MobileNetV2 was established as the corrosion morphology recognition model.

[0034] Furthermore, the establishment of the corrosion morphology recognition model and the corrosion behavior prediction model includes:

[0035] An 8th-order polynomial regression model was established as the corrosion behavior prediction model.

[0036] Furthermore, the use of the uranium-niobium alloy corrosion sample images and the corresponding label data to train the corrosion morphology recognition model and the corrosion behavior prediction model includes:

[0037] performing multi-dimensional image optimization on each of the uranium-niobium alloy corrosion sample images, thereby expanding the number of the uranium-niobium alloy corrosion sample images;

[0038] The corrosion morphology recognition model and the corrosion behavior prediction model are trained using the uranium-niobium alloy corrosion sample images after multi-dimensional image optimization and the corresponding label data.

[0039] On the other hand, an embodiment of the present invention also includes a computer device including a memory and a processor, the memory being used to store at least one program, and the processor being used to load at least one program to execute the machine learning-based actinide alloy corrosion behavior prediction method in the embodiment.

[0040] On the other hand, an embodiment of the present invention also includes a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to execute the method for predicting the corrosion behavior of actinide alloys based on machine learning in the embodiment.

[0041] The beneficial effect of the present invention is that the method for predicting the corrosion behavior of actinide alloys based on machine learning in the embodiment, by using a corrosion morphology recognition model and a corrosion behavior prediction model, first identifies the corrosion morphology and then predicts the corrosion behavior. It can predict the corrosion behavior of actinide alloys based on the image to be identified of the actinide alloy, and provides more comprehensive and reliable tools and technical means to solve the difficulties in studying the corrosion behavior of actinide alloys and analyze the corrosion characteristics and laws of actinide alloys. It overcomes the defect of low accuracy of the existing technology in studying the corrosion behavior of actinide alloys and significantly improves the processing efficiency of actinide alloy corrosion data. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the steps of the method for predicting the corrosion behavior of actinide alloys based on machine learning in an embodiment;

[0043] Figure 2 Schematic diagram of the overall process of the method for predicting the corrosion behavior of actinide alloys based on machine learning in the embodiment;

[0044] Figure 3 Schematic diagram of the corrosion states corresponding to different uranium-niobium alloy corrosion sample images in the embodiment;

[0045] Figure 4 Schematic diagram of corrosion behavior prediction data of actinide alloys in the form of corrosion area in the embodiment;

[0046] Figure 5 Schematic diagram of corrosion behavior prediction data of actinide alloys in the form of corrosion grades in the examples;

[0047] Figure 6 Schematic diagram of an actinide alloy corrosion image to be identified in an embodiment;

[0048] Figure 7 Schematic diagram of visualization results of color distribution information in corrosion color recognition time series data in an embodiment. DETAILED DESCRIPTION

[0049] At present, the corrosion behavior identification and evaluation of actinide alloys still relies on special personnel to carry out visual observation. This manual observation method is affected by the experience of the observer, which can easily lead to inconsistent detection results and detection bias. In addition, manual detection often obtains qualitative data, and quantitative data is less. In addition, the processing efficiency of the observer for a large amount of data is limited, which cannot meet the demand of modern production and life for rapid decision-making.

[0050] With the development of the application of artificial intelligence technology and the latest research results, machine learning methods can be used as a technical means to assist people in researching and solving corrosion problems. Based on the above principle, in the embodiment, a method for predicting the corrosion behavior of actinide alloys based on machine learning is provided. Referring to Figure 1 , the method for predicting the corrosion behavior of actinide alloys based on machine learning includes the following steps:

[0051] S1. Obtain a plurality of actinide alloy corrosion images to be identified;

[0052] S2. Obtain a corrosion morphology identification model and a corrosion behavior prediction model based on machine learning;

[0053] S3. Input each actinide alloy corrosion image to be identified into the corrosion morphology identification model for identification processing, and determine corrosion morphology identification time series data according to the processing result of the corrosion morphology identification model;

[0054] S4. Input the corrosion morphology identification time series data into the corrosion behavior prediction model for prediction processing, and determine the actinide alloy corrosion behavior prediction data according to the processing result of the corrosion behavior prediction model.

[0055] In the process of executing steps S1-S4, the corrosion morphology identification model and the corrosion behavior prediction model based on machine learning are needed. Among them, the plurality of actinide alloy corrosion images to be identified in step S1 can be the corrosion parts of the same actinide alloy (such as an actinide alloy sheet or an actinide alloy block), and the images are obtained at different times. These images will be processed by the corrosion morphology identification model and the corrosion behavior prediction model. Specifically, the corrosion morphology identification model is used to identify the information related to the corrosion morphology of the actinide alloy in the actinide alloy corrosion image to be identified, and the corrosion behavior prediction model is used to predict the corrosion behavior of the actinide alloy according to the information related to the corrosion morphology of the actinide alloy identified by the corrosion morphology identification model.

[0056] In the embodiment, the corrosion morphology identification model and the corrosion behavior prediction model can be trained and saved before steps S1-S4 are executed, and the trained corrosion morphology identification model and the corrosion behavior prediction model are called when steps S1-S4 are executed. Specifically, the training process of the corrosion morphology identification model and the corrosion behavior prediction model includes the following steps:

[0057] P1. Establish a corrosion morphology recognition model and a corrosion behavior prediction model;

[0058] P2. Acquire several images of uranium-niobium alloy corrosion samples;

[0059] P3. Based on the pixel distribution corresponding to the corroded and non-corroded areas in the uranium-niobium alloy corrosion sample image, obtain label data corresponding to the uranium-niobium alloy corrosion sample image;

[0060] P4. Use the uranium-niobium alloy corrosion sample images and corresponding label data to train the corrosion morphology recognition model and corrosion behavior prediction model.

[0061] In step P1, appropriate machine learning models are selected as the corrosion morphology recognition model and the corrosion behavior prediction model respectively.

[0062] (1) For corrosion morphology recognition model:

[0063] In this embodiment, the DeeplabV3+ model can be considered as the corrosion morphology recognition model. Specifically, the DeeplabV3+ model is optimized and explored, and the DeeplabV3+ model with Xception as the backbone network and the DeeplabV3+ model with MobileNetV2 as the backbone network are selected for prediction performance comparison and analysis, and finally the model with the best performance is selected as the corrosion morphology recognition model. To ensure that the experimental results are more comparable, in this embodiment, the experimental steps and experimental parameter settings are consistent, including loading the corresponding pre-training weights for model initialization, setting the same Epoch number in the comparative experiment, setting the same batch size in the freezing stage and the thawing stage, using the SGD optimizer and setting the same model maximum learning rate and minimum learning rate, etc. In this embodiment, multiple key indicators are used to comprehensively evaluate the prediction performance of the model, including training loss Train Loss, validation loss Val Loss, average precision mPrecision, average recall rate mRecall, average intersection-over-union ratio mIOU, etc. Among them, Train Loss and Val Loss are used to measure the difference between the prediction results of the model and the actual labels in the training set and validation set.

[0064] The performance of the DeeplabV3+ model with Xception backbone network and the DeeplabV3+ model with MobileNetV2 backbone network in the uranium niobium alloy corrosion image is shown in Table 1. As shown in Table 1, in the prediction task of the uranium niobium alloy corrosion image, the training loss and the validation loss of the DeeplabV3+ model show a stable downward trend, and finally converge to 0.05 to 0.06, indicating that the DeeplabV3+ model has learned the effective features of the uranium niobium alloy corrosion morphology image. By comprehensively analyzing and evaluating the actual performance of the DeeplabV3+ model on the uranium niobium alloy corrosion image in different dimensions, it can be seen from the experimental data mPrecision, mRecall and mIOU that the DeeplabV3+ model based on MobileNetV2 has high precision, can reduce the false prediction of the corrosion area, has high recall rate, can identify more corrosion areas, and has high intersection over union, indicating that the overlap degree of the predicted area and the real corrosion area is better, and the boundary prediction is more accurate. Therefore, compared with the DeeplabV3+ model based on Xception, the DeeplabV3+ model based on MobileNetV2 is the best model in terms of performance on the uranium niobium alloy corrosion image dataset. In this embodiment, the DeeplabV3+ model with MobileNetV2 backbone network is selected as the corrosion morphology recognition model.

[0065] Table 1 Performance of DeeplabV3+ model with different backbone networks

[0066] backbone network Train Loss Val Loss mPrecision mRecall mIOU MobileNetV2 0.058 0.060 97.96％ 98.09％ 96.16％ Xception 0.063 0.055 97.73％ 97.99％ 95.85％

[0067] After selecting the DeeplabV3+ model with a MobileNetV2 backbone network as the corrosion morphology recognition model, in step P1, the DeeplabV3+ model can be built using the Pytorch framework, combined with the lightweight MobileNetV2 neural network as the backbone feature extraction network. The DeeplabV3+ model's network architecture is primarily divided into an encoding and decoding portion. The encoding portion primarily comprises the MobileNetV2 backbone network and the ASPP module. Image data input to the corrosion morphology recognition model (e.g., uranium-niobium alloy corrosion images) is first fed into MobileNetV2 for multi-layer feature extraction, yielding backbone features and shallow features. The backbone features output by MobileNetV2 are then fed into the ASPP module, which comprises multiple parallel convolutional branches and a global average pooling branch. The convolutional branches include one 1×1 and three 3×3 convolutions with varying dilation rates (e.g., 1, 6, 12, and 18). All branches capture information at different scales and obtain global features. Finally, the feature maps from the five branches are fused and passed through a 1×1 convolutional layer as the final output of the ASPP module. In the decoding part, the feature map output of the ASPP module undergoes a bilinear interpolation and is then spliced ​​with the feature map of the shallow feature after a 1×1 convolution operation. The spliced ​​feature map undergoes a 3×3 convolution and a bilinear interpolation to obtain a predicted image with the same resolution as the original uranium-niobium alloy image. The predicted image is output as the output result of the corrosion morphology recognition model. The predicted image represents the corrosion area identified by the corrosion morphology recognition model from the image data (such as the uranium-niobium alloy corrosion image) input to the corrosion morphology recognition model.

[0068] (2) For corrosion behavior prediction model:

[0069] In this embodiment, a polynomial regression model may be considered as a corrosion behavior prediction model. In this embodiment, when building a polynomial regression model, optimization may be performed by sequential polynomial fitting to obtain a polynomial regression model for use as a corrosion behavior prediction model.

[0070] Specifically, we can build polynomial regression models from order 1 to order 9 and compare the coefficient of determination R of polynomial regression models at different orders. 2 and mean square error MSE, where R 2 Used to evaluate the model's fit to the data, R 2 The range is between 0 and 1. The closer the value is to 1, the better the model fit is. MSE is used to measure the difference between the predicted value and the true value. The smaller the MSE, the better the prediction performance of the model.

[0071] The fitting results of the polynomial regression models at different orders are shown in Table 2. From the comparison of the prediction effects of the polynomial regression models at different orders in Table 2, it can be seen that the nonlinear relationship between the corrosion rate of uranium-niobium alloy and time is best fitted by the polynomial regression model with an order of 8, and its mean square error (MSE) and determination coefficient (R) are 0. 2 The value of shows that the predicted value of the corrosion rate by the polynomial regression model is very close to the actual value. The model has high accuracy in predicting the corrosion data of uranium-niobium alloy and achieves the best fitting effect.

[0072] Table 2 MSE and R of polynomial regression models of different orders 2 value

[0073]

[0074]

[0075] In addition, in this embodiment, a random forest regression model (RF), a support vector regression model (SVR) and a gradient boosting regression model (GBR) were built for the uranium-niobium alloy corrosion data, and GridSearchCV was used for model optimization. Finally, the RF, SVR and GBR models with the best performance after optimization were selected for comparison with the 8th-order polynomial regression model (PR) to select the model with the best performance in the prediction of uranium-niobium alloy corrosion.

[0076] The comparison of the fitting results of the best performing RF, SVR, GBR models and the 8th order polynomial regression model is shown in Table 3. As shown in Table 3, the determination coefficient R of the 8th order polynomial regression model is 2 The mean square error (MSE) and the mean square error (MSE) are superior to those of the random forest regression model, support vector regression model, and gradient boosting regression model, indicating that the polynomial regression model has a more accurate prediction effect on the corrosion behavior data of uranium-niobium alloy. The model is more excellent in capturing the nonlinear relationship in the corrosion process, making it the optimal corrosion behavior prediction model for uranium-niobium alloy. If an 8th-order polynomial regression model is used as the corrosion behavior prediction model, then by inputting the time parameter into the corrosion behavior prediction model, it can accurately achieve digital prediction of corrosion behavior and digital simulation of the corrosion process, effectively reflecting the corrosion development state of uranium-niobium alloy during the time evolution process.

[0077] Table 3 MSE and R of RF, SVR, GBR and PR 2 value

[0078] MSE <![CDATA[R 2 ]]> RF 0.00002149 0.9964 SVR 0.00004199 0.9930 GBR 0.00004518 0.9925 PR 0.00000504 0.9992

[0079] Therefore, in this embodiment, an 8th-order polynomial regression model can be selected as the corrosion behavior prediction model.

[0080] In step P1, a DeeplabV3+ model with a MobileNetV2 backbone network is established as a corrosion morphology recognition model, and an 8th-order polynomial regression model is established as a corrosion behavior prediction model. Then, steps P2-P4 are executed to train the corrosion morphology recognition model and the corrosion behavior prediction model.

[0081] In step P2, a plurality of uranium-niobium alloy corrosion sample images are obtained. These uranium-niobium alloy corrosion sample images can be images taken at different times of corrosion sites of uranium-niobium alloy (can be the same piece or different pieces). The morphology of the uranium-niobium alloy corrosion sample images is as follows: Figure 3 As shown in FIG. 1 , the uranium-niobium alloy corrosion sample image can be labeled according to the corrosion state of the uranium-niobium alloy presented by the uranium-niobium alloy corrosion sample image. Figure 3 ,The uranium-niobium alloy corrosion sample images basically show four corrosion states, namely, the initial non-corrosion state (e.g. Figure 3 The first row in the figure shows a corrosion area of ​​0.96%, and the corresponding corrosion level is relatively mild, level 6. A small amount of pitting corrosion occurs (e.g. Figure 3 The second row in the middle has a corrosion area of ​​11.63%, and the corresponding corrosion level is medium to slight level 2). A few areas have no corrosion (e.g. Figure 3 The third row in the figure has a corrosion area of ​​26.84%, and the corresponding corrosion level is medium to severe level 1). The entire surface is covered with corrosion (e.g. Figure 3 In the 4th row, the corresponding corrosion area is at the level of 69.15%, and the corresponding corrosion level is severe level 0).

[0082] In step P2, for the multiple acquired uranium-niobium alloy corrosion sample images, a metal corrosion morphology extraction method can be first used, such as a color segmentation algorithm based on computer vision, to convert the uranium-niobium alloy corrosion sample images into the HSV color space, so as to better separate the corrosion morphology in the subsequent process. Secondly, based on the color characteristics of the uranium-niobium alloy corrosion morphology, a specific color threshold is set for the algorithm to achieve feature extraction of the corrosion area.

[0083] For each uranium niobium alloy corrosion sample image, step P2 can be performed to extract the corrosion area therein, which indicates the area where the corrosion site of the uranium niobium alloy in the uranium niobium alloy corrosion sample image is located. For the same uranium niobium alloy corrosion sample image, the pixel points corresponding to the corrosion area (the area where the corrosion site has occurred) and the pixel points corresponding to the non-corrosion area (the area where the corrosion site has not occurred) can be counted to obtain the area of the corrosion area and the area of the non-corrosion area, and the corrosion grade corresponding to the uranium niobium alloy corrosion sample image can be obtained according to the area of the corrosion area and the technical rating standard (for example, GB / T 6461-2002 “Metallic Coatings and Other Inorganic Coatings on Metal Substrates - Evaluation of Specimens and Test Pieces After Corrosion Testing”).

[0084] In step P3, for any uranium niobium alloy corrosion sample image, the corrosion area and the corrosion grade corresponding thereto can be used as the label data of the uranium niobium alloy corrosion sample image, and the uranium niobium alloy corrosion sample image can be labeled by using a labeling tool such as Labelme. Specifically, for any uranium niobium alloy corrosion sample image, the pixel points in the corrosion area can be set with a label with a class name “Corrosion”, and the pixel points in the non-corrosion area can be set with a label with a class name “Background”, and these labels constitute the label data corresponding to the uranium niobium alloy corrosion sample image.

[0085] In this embodiment, due to the radioactivity and scarcity of uranium niobium alloy, the number of uranium niobium alloy corrosion sample images directly obtained by photographing the uranium niobium alloy in step P2 is generally small. Therefore, when step P2 is performed, multi-dimensional image optimization can be performed on each uranium niobium alloy corrosion sample image to expand the number of uranium niobium alloy corrosion sample images.

[0086] For example, when step P2 is performed, the number of uranium niobium alloy corrosion sample images directly obtained by photographing the uranium niobium alloy is generally 60, and thus 60 corresponding label data files will be obtained by performing step P3. Such a small number of uranium niobium alloy corrosion sample images and label data are not conducive to the training of step P4.

[0087] In order to enable the trained corrosion morphology recognition model to have higher prediction accuracy and lower overfitting, a large amount of high-quality training data is usually required. In this embodiment, before step P4 is performed, multi-dimensional image optimization is performed on each uranium niobium alloy corrosion sample image to expand the number of uranium niobium alloy corrosion sample images.

[0088] Specifically, the 60 original uranium-niobium alloy corrosion sample images and their corresponding label files obtained by executing step P2 can be subjected to various transformations such as mirroring and random angle rotation to obtain more uranium-niobium alloy corrosion sample images. By using the uranium-niobium alloy corrosion sample images expanded by these methods to train the corrosion morphology recognition model, the adaptability of the corrosion morphology recognition model to rotation invariance can be enhanced; the original uranium-niobium alloy corrosion sample images can also be randomly adjusted to perform color dithering on the contrast, sharpness, and brightness of the images to obtain more uranium-niobium alloy corrosion sample images. By using the uranium-niobium alloy corrosion sample images expanded by these methods to train the corrosion morphology recognition model, the corrosion morphology recognition model can be improved. The generalization ability of the image under different color conditions; it is also possible to randomly crop different areas from the original uranium-niobium alloy corrosion sample image to obtain more uranium-niobium alloy corrosion sample images, and by using the uranium-niobium alloy corrosion sample images expanded by these methods to train the corrosion morphology recognition model, it is possible to improve the recognition accuracy of the corrosion morphology recognition model when some image information is lost; it is also possible to add random noise to the original uranium-niobium alloy corrosion sample image to obtain more uranium-niobium alloy corrosion sample images, and by using the uranium-niobium alloy corrosion sample images expanded by these methods to train the corrosion morphology recognition model, it is possible to help the corrosion morphology recognition model learn a wider range of features and improve the stability of the corrosion morphology recognition model in real environments.

[0089] Generally speaking, by performing multi-dimensional image optimization on each uranium-niobium alloy corrosion sample image, the original number of 60 uranium-niobium alloy corrosion sample images can be expanded and enhanced to the number of 960. The corresponding label data of the uranium-niobium alloy corrosion sample images is also expanded, enriching the diversity of uranium-niobium alloy corrosion sample images and obtaining a larger uranium-niobium alloy corrosion sample image library. This effectively alleviates the model overfitting problem caused by insufficient training data and provides an important data foundation for improving the accuracy and stability of the prediction model. In addition, the multi-dimensional image optimization algorithm can solve the problem of image annotation requiring a large amount of time, significantly reducing the cost of manual annotation, and generating a large amount of pixel-level label data for the corrosion morphology recognition model, helping the corrosion morphology recognition model to more accurately evaluate and predict the corrosion morphology of uranium-niobium alloys.

[0090] In step P4, the expanded uranium-niobium alloy corrosion sample images are divided into a data set with a training set and a validation set ratio of 9:1. For example, if there are a total of 960 uranium-niobium alloy corrosion sample images, 864 of them are randomly selected as the training set, and 96 are randomly selected as the validation set. When the training set is used to train the corrosion morphology recognition model, the uranium-niobium alloy corrosion sample images in the training set are used as the input of the corrosion morphology recognition model, and the label data corresponding to the uranium-niobium alloy corrosion sample images input into the corrosion morphology recognition model are used as the expected output of the corrosion morphology recognition model. By training the corrosion morphology recognition model, the corrosion morphology recognition model can acquire the ability to recognize the input image data, thereby obtaining information such as the location and size of the corrosion area and the corrosion grade.

[0091] Since uranium-niobium alloy is a typical actinide alloy, the ability to identify the corrosion area of ​​uranium-niobium alloy obtained by training the corrosion morphology recognition model can also be used to identify the information of the corrosion area of ​​other actinide alloys.

[0092] In step P4, each label data item can be arranged into a time series based on the capture time of each uranium-niobium alloy corrosion sample image corresponding to each label data item. This time series represents the temporal development of the corrosion region in the uranium-niobium alloy. The corrosion behavior prediction model can be trained using the first portion of this time series as input, and the second portion of this time series as the expected output of the corrosion behavior prediction model. This enables the corrosion behavior prediction model to predict the time series representing information related to the corrosion region and output a prediction result for the development of this time series.

[0093] By executing steps P1-P4, a digital prediction platform for the corrosion behavior of actinide alloys can be constructed using the trained corrosion morphology recognition model and corrosion behavior prediction model, and steps S1-S4 can be executed using the digital prediction platform for the corrosion behavior of actinide alloys.

[0094] In this embodiment, the overall process of the method for predicting the corrosion behavior of actinide alloys based on machine learning is as follows: Figure 2 shown.

[0095] Reference Figure 2 In step S1, the actinide alloy whose corrosion behavior needs to be predicted can be photographed continuously, and corresponding actinide alloy corrosion images to be identified can be obtained by photographing them at multiple moments.

[0096] In step S2, the corrosion morphology recognition model and the corrosion behavior prediction model obtained by training in steps P1-P4 are called.

[0097] In this embodiment, when executing step S3, that is, inputting each actinide alloy corrosion image to be identified into the corrosion morphology recognition model for identification processing, and determining the corrosion morphology recognition time series data based on the processing results of the corrosion morphology recognition model, the following steps can be specifically performed:

[0098] S301. For any actinide alloy corrosion image to be identified, input the actinide alloy corrosion image to be identified into the corrosion morphology recognition model for identification processing, obtain corrosion area data output by the corrosion morphology recognition model, and mark the corrosion area data with the acquisition time of the actinide alloy corrosion image to be identified;

[0099] S302. Sort the corrosion area data according to their corresponding collection time;

[0100] S303. Determine corrosion morphology recognition time series data based on the sorting results.

[0101] Assume that in step S1, the image image1 of the actinide alloy corrosion to be identified taken at time t1, the image image2 of the actinide alloy corrosion to be identified taken at time t2, and so on are obtained. n Actinide alloy corrosion image to be identified captured at any moment n If there are n actinide alloy corrosion images to be identified, then in step S301, the corrosion morphology recognition model can be used to identify the actinide alloy corrosion image to be identified image1, and obtain the corresponding corrosion area information (including corrosion area S1, corrosion level class1, etc.). Similarly, the corrosion morphology recognition model can be used to identify the actinide alloy corrosion image to be identified image2, and obtain the corresponding corrosion area information (including corrosion area S2, corrosion level class2, etc.). n Identify and obtain the corresponding corrosion area information (including corrosion area S n , corrosion grade class n wait).

[0102] The corrosion area information obtained in step S301 corresponds to t1, t2, ..., t n Therefore, in step 302, the corrosion area data are sorted according to their corresponding collection time, for example, the corrosion areas are sorted, so as to obtain the corrosion morphology recognition time series data S1, S2...S n ; Sort the corrosion levels to obtain the corrosion morphology recognition time series data class1, class2...class n ; Identify time series data S1, S2...S according to corrosion morphology n By calculation, the corrosion morphology recognition time series data v1, v2...v in the form of corrosion velocity can be obtained.n-1 , where v1=(S 2- S1) / (t 2- t1), v2=(S 3- S2) / (t 3- t2)……v n-1 =(S n- S n-1 ) / (t n- t n-1 ).

[0103] In step S303, S1, S2, ..., S3 can be selected according to the data type that the corrosion behavior prediction model can process. n Or class1, class2...class n or v1, v2...v n-1 Corrosion morphology recognition time series data in other forms.

[0104] In step S4, any corrosion morphology recognition time series data obtained in step S303 is input into the corrosion behavior prediction model, and the corrosion behavior prediction model performs prediction processing on the corrosion morphology recognition time series data. For example, the corrosion behavior prediction model is used for S1, S2...S n The corrosion morphology recognition time series data in the form of t n After time t n+1 The corrosion area S corresponding to one or more moments n+1 etc.; for class1, class2...class n The corrosion morphology recognition time series data in the form of t n After time t n+1 The corrosion level class corresponding to one or more moments n+1 etc.; for v1, v2...v n-1 The corrosion morphology recognition time series data in the form of t n After time t n+1 The corrosion rate v corresponding to one or more moments n etc. Output of corrosion behavior prediction model S n+1 、class n+1 and v n The data are prediction data of the corrosion behavior of actinide alloys, which represent the prediction results of the corrosion behavior of the actinide alloys in the actinide alloy corrosion to be identified image at a future moment.

[0105] The method for predicting the corrosion behavior of actinide alloys based on machine learning in this embodiment uses a corrosion morphology recognition model and a corrosion behavior prediction model to first identify the corrosion morphology and then predict the corrosion behavior. It can predict the corrosion behavior of actinide alloys based on the image to be identified of the actinide alloy. This provides more comprehensive and reliable tools and technical means to solve the difficulties in studying the corrosion behavior of actinide alloys and analyze the corrosion characteristics and laws of actinide alloys. It overcomes the defect of low accuracy of existing technologies in studying the corrosion behavior of actinide alloys and significantly improves the processing efficiency of actinide alloy corrosion data.

[0106] Moreover, by selecting the best-performing DeeplabV3+ model as the corrosion morphology recognition model, the automatic detection and identification of corrosion morphology features were successfully achieved, thereby improving the efficiency of corrosion morphology appearance analysis. By building and optimizing the random forest regression model, support vector regression model, gradient boosting regression model, and polynomial regression model based on uranium-niobium alloy corrosion data, the best-performing model was selected as the corrosion behavior prediction model. This corrosion behavior prediction model can capture the nonlinear relationship between the corrosion behavior of uranium-niobium alloy and time. By inputting time parameters, it can accurately realize the digital prediction of corrosion behavior and the digital simulation of the corrosion process, thereby realizing the quantitative calculation of the corrosion area of ​​uranium-niobium alloy. For example, By inputting the actinide alloy corrosion image to be identified, the output actinide alloy corrosion behavior prediction data can reflect the expansion and change trend of the corrosion area on the actinide alloy, realize the automatic assessment and prediction of the corrosion level of actinide alloys such as uranium-niobium alloy, and achieve rapid and accurate evaluation of the corrosion level of uranium-niobium alloy under data-driven, reducing the complexity of manual analysis and effectively improving detection efficiency; by performing multi-dimensional image optimization on each uranium-niobium alloy corrosion sample image when training the corrosion morphology recognition model and the corrosion behavior prediction model, the number of uranium-niobium alloy corrosion sample images can be expanded, which can solve the current problems of the extreme lack of research on the identification and prediction of actinide alloy corrosion behavior, as well as the scarcity of actinide alloy corrosion data.

[0107] In this embodiment, the corrosion area S obtained by performing steps S1-S4 once is n+1 The corrosion behavior prediction data of actinide alloys in the form of Figure 4 As shown. Figure 4 By executing steps S1-S4, the automatic detection and identification of corrosion morphology features were successfully achieved, which improved the efficiency of corrosion morphology appearance analysis. The matching degree between the predicted corrosion area and the actual corrosion area reached 96.16%.

[0108] In this embodiment, the corrosion level class obtained by performing steps S1-S4 once is n+1 The corrosion behavior prediction data of actinide alloys in the form of Figure 5 As shown. Figure 5By executing steps S1-S4, a curve chart showing the change of the corrosion level of the corrosion area over time is obtained. The chart reflects the expansion and change trend of the corrosion level of the actinide alloy, achieving the effect of monitoring the dynamic process of corrosion. The average prediction accuracy of the corrosion area reaches 97.96%, and the average pixel accuracy of the corrosion level prediction results of the actinide alloy reaches 98.09%. This realizes the rapid and accurate assessment of the corrosion level of uranium-niobium alloy under data-driven, reduces the complexity of manual analysis, and effectively improves detection efficiency and result consistency.

[0109] In this embodiment, referring to the principle of step P2, the image data input into the corrosion morphology recognition model is generally image data in the RGB color space, that is, when executing step S3, the metal corrosion morphology extraction method can also be used first, such as a color segmentation algorithm based on computer vision, to convert the actinide alloy corrosion image to be identified into the HSV color space, so as to better separate the corrosion morphology in the future. Secondly, according to the color characteristics of the actinide alloy corrosion morphology, a specific color threshold is set for the algorithm to achieve feature extraction of the corrosion area. However, the experimental results show that the traditional metal corrosion morphology extraction method cannot accurately extract the corrosion morphology of actinide alloys, and the segmentation effect of the corrosion area is not ideal. Further analysis of the traditional method based on the preset color threshold shows that this method is very sensitive to the illumination changes of the actinide alloy corrosion image to be identified. When the brightness and darkness of the actinide alloy corrosion image to be identified changes, it is easy to cause inaccurate segmentation effect, affecting the subsequent evaluation and prediction of the corrosion morphology; for example, the typical content of the actinide alloy corrosion image to be identified is as follows Figure 6 As shown in the figure, when the corrosion area and the background color are similar, it is impossible to accurately distinguish the boundary between the corrosion area and the non-corrosion area in the actinide alloy corrosion image to be identified, and the edge processing is relatively rough. This method relies on the threshold setting of the color space, which makes it difficult to uniformly set the color threshold in complex corrosion scenes, which is not conducive to achieving fast and accurate identification and prediction. These factors may bring difficulties to the execution of step S3.

[0110] In order to overcome the above difficulties, in this embodiment, reference can be made to Figure 2 , perform the following steps:

[0111] SA301. Perform main color analysis on each actinide alloy corrosion image to be identified, and obtain the color distribution information of each actinide alloy corrosion image to be identified;

[0112] SA302. Mark the color distribution information based on the acquisition time of the actinide alloy corrosion image to be identified;

[0113] SA303 sorted according to the corresponding acquisition time of each color distribution information;

[0114] SA304. Determine the corrosion color recognition time series data based on the sorting results.

[0115] In step SA301, the OpenCV library and the K-Means clustering algorithm can be used to perform main color analysis on the actinide alloy corrosion image to be identified. Specifically, OpenCV is used to read and convert the color format of the actinide alloy corrosion image to be identified, and the colors in the actinide alloy corrosion image to be identified are divided into multiple clusters (generally 3 clusters) through the K-Means clustering algorithm, where the center color of the cluster represents the main color of the cluster. Finally, the proportion of the 3 main colors in the corrosion image is calculated to obtain the color distribution information of each actinide alloy corrosion morphology image. In this embodiment, the color distribution information of each actinide alloy corrosion image to be identified can be represented as a set of vectors, and each element in the vector represents the proportion of the corresponding main color.

[0116] In steps SA302-SA304, each color distribution information (vector) can be mapped to the shooting time t1, t2, ..., t of the corresponding actinide alloy corrosion image to be identified. n Arrange them into a time series to obtain corrosion color recognition time series data.

[0117] After executing steps SA302-SA304, the color distribution information represented by each corrosion color recognition time series data can be visualized through a histogram and played in a time sequence, thereby intuitively displaying the color distribution of the actinide alloy corrosion image to be identified, achieving quantitative analysis of the color characteristics of the actinide alloy corrosion image, and helping researchers to intuitively understand the overall color tendency of the actinide alloy corrosion morphology image. Figure 7 The figure shows a visualization result of color distribution information in corrosion color recognition time series data.

[0118] In this embodiment, steps SA301-SA304 can replace step S3. That is, after executing steps S1-S2, steps SA301-SA304 are selected to be executed instead of step S3. After executing steps SA301-SA304, step S4 is executed. In step S4, the corrosion color recognition time series data is used as the corrosion morphology recognition time series data and input into the corrosion behavior prediction model for processing. By using the corrosion color recognition time series data as the corrosion morphology recognition time series data, the corrosion behavior prediction can be directly performed using the image feature information contained in the to-be-identified corrosion image of the actinide alloy. This avoids the effects of noise interference or precision degradation generated during the acquisition of the corrosion morphology recognition time series data in the form of corrosion area, corrosion level, or corrosion rate, and facilitates more accurate tracking and prediction of the corrosion behavior of the actinide alloy.

[0119] A computer program that executes the method for predicting the corrosion behavior of actinide alloys based on machine learning in this embodiment can be written and written into a computer device or storage medium. When the computer program is read out and run, the method for predicting the corrosion behavior of actinide alloys based on machine learning in this embodiment is executed, thereby achieving the same technical effect as the method for predicting the corrosion behavior of actinide alloys based on machine learning in the embodiment.

[0120] It should be noted that, unless otherwise specified, when a feature is referred to as being "fixed" or "connected" to another feature, it may be directly fixed or connected to the other feature, or it may be indirectly fixed or connected to the other feature. In addition, the descriptions of up, down, left, right, etc. used in this disclosure are only relative to the relative positional relationships of the components of the present disclosure in the accompanying drawings. The singular forms of "a", "" and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. In addition, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as those generally understood by those skilled in the art. The terms used in the specification of this embodiment are only for describing specific embodiments and are not intended to limit the invention. The term "and / or" used in this embodiment includes any combination of one or more related listed items.

[0121] It should be understood that, although the present disclosure may adopt the term first, second, third etc. to describe various elements, these elements should not be limited to these terms.These terms are only used to distinguish the elements of the same type from each other.For example, without departing from the scope of the present disclosure, the first element may also be referred to as the second element, and similarly, the second element may also be referred to as the first element.The use of any and all examples or exemplary language ("for example", "such as" etc.) provided by the present embodiment is only intended to better illustrate embodiments of the present invention, and unless otherwise required, the scope of the present invention will not be limited.

[0122] It should be appreciated that embodiments of the present invention can be implemented or practiced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The methods can be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and figures described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. In addition, the program can be run on a programmed application-specific integrated circuit for this purpose.

[0123] In addition, the operations of the processes described in this embodiment may be performed in any suitable order, unless otherwise indicated in this embodiment or otherwise clearly contradicted by the context. The processes described in this embodiment (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. A computer program includes multiple instructions that can be executed by one or more processors.

[0124] Furthermore, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, RAM, ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or portions thereof, can be transmitted over a wired or wireless network. When such media includes instructions or programs that implement the above steps in conjunction with a microprocessor or other data processor, the invention of this embodiment includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.

[0125] The computer program can be applied to input data to perform the functions of the present embodiment, thereby converting the input data to generate output data that is stored in a non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the present invention, the converted data represents a physical and tangible object, including a specific visual depiction of the physical and tangible object produced on the display.

[0126] The above are merely preferred embodiments of the present invention. The present invention is not limited to the aforementioned embodiments. As long as the technical effects of the present invention are achieved by the same means, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention. Within the scope of protection of the present invention, various modifications and variations of the technical solutions and / or implementation methods may be made.

Claims

1. A method for predicting the corrosion behavior of actinide alloys based on machine learning, characterized in that: The method comprises: Acquire multiple actinide alloy corrosion images to be identified; Obtaining a corrosion morphology recognition model and a corrosion behavior prediction model based on machine learning; the corrosion morphology recognition model and the corrosion behavior prediction model are trained on images of uranium-niobium alloy corrosion samples; Inputting each of the actinide alloy corrosion images to be identified into the corrosion morphology recognition model for identification processing, and determining corrosion morphology recognition time series data according to the processing results of the corrosion morphology recognition model; Inputting the corrosion morphology recognition time series data into the corrosion behavior prediction model for prediction processing, and determining the actinide alloy corrosion behavior prediction data according to the processing result of the corrosion behavior prediction model; performing main color analysis on each of the actinide alloy corrosion images to be identified, respectively, to obtain color distribution information of each of the actinide alloy corrosion images to be identified; Marking the color distribution information according to the acquisition time of the actinide alloy corrosion image to be identified; sorting the color distribution information according to the collection time corresponding to each of the color distribution information; Determine corrosion color recognition time series data according to the sorting results; The step of inputting each of the actinide alloy corrosion images to be identified into the corrosion morphology recognition model for identification processing, and determining the corrosion morphology recognition time series data according to the processing results of the corrosion morphology recognition model, comprises: For any of the actinide alloy corrosion images to be identified, input the actinide alloy corrosion image to be identified into the corrosion morphology recognition model for identification processing, obtain corrosion area data output by the corrosion morphology recognition model, and mark the corrosion area data with the acquisition time of the actinide alloy corrosion image to be identified; Sorting the corrosion area data according to their respective corresponding collection times; Determining the corrosion morphology recognition time series data according to the sorting result; Determining the corrosion morphology identification time series data according to the sorting results includes: Using the sorted corrosion area data as the corrosion morphology recognition time series data; or Determining a plurality of corrosion rate data according to the sorted corrosion area data; Using each of the corrosion rate data as the corrosion morphology identification time series data; or According to the metal corrosion rating standard, the sorted corrosion area data are converted into corresponding corrosion grade data; The corrosion level data are used as the corrosion morphology recognition time series data.

2. The method for predicting corrosion behavior of actinide alloys based on machine learning according to claim 1, characterized in that: The method for predicting the corrosion behavior of actinide alloys based on machine learning also includes: Establish corrosion morphology recognition model and corrosion behavior prediction model; Acquire several images of uranium-niobium alloy corrosion samples; labeling the uranium-niobium alloy corrosion sample image according to the pixel distribution corresponding to the corrosion area and the non-corrosion area in the uranium-niobium alloy corrosion sample image to obtain label data corresponding to the uranium-niobium alloy corrosion sample image; The corrosion morphology recognition model and the corrosion behavior prediction model are trained using the uranium-niobium alloy corrosion sample images and the corresponding label data.

3. The method for predicting corrosion behavior of actinide alloys based on machine learning according to claim 2, characterized in that: The establishment of the corrosion morphology recognition model and the corrosion behavior prediction model includes: A DeeplabV3+ model based on MobileNetV2 was established as the corrosion morphology recognition model.

4. The method for predicting corrosion behavior of actinide alloys based on machine learning according to claim 2, characterized in that: The establishment of the corrosion morphology recognition model and the corrosion behavior prediction model includes: An 8th-order polynomial regression model was established as the corrosion behavior prediction model.

5. The method for predicting corrosion behavior of actinide alloys based on machine learning according to any one of claims 2 to 4, characterized in that: The method of using the uranium-niobium alloy corrosion sample images and the corresponding label data to train the corrosion morphology recognition model and the corrosion behavior prediction model includes: performing multi-dimensional image optimization on each of the uranium-niobium alloy corrosion sample images, thereby expanding the number of the uranium-niobium alloy corrosion sample images; The corrosion morphology recognition model and the corrosion behavior prediction model are trained using the uranium-niobium alloy corrosion sample images after multi-dimensional image optimization and the corresponding label data.

6. A computer device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load at least one program to execute the method for predicting corrosion behavior of actinide alloys based on machine learning as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to execute the method for predicting the corrosion behavior of actinide alloys based on machine learning when executed by the processor.

Images