Rebar corrosion inhibitor performance prediction method based on functional group classification and machine learning

Abstract

The invention belongs to the field of corrosion inhibitor performance prediction, and particularly relates to a reinforcing steel bar corrosion inhibitor performance prediction method based on functional group classification and machine learning. Acquiring index data highly related to the corrosion inhibitor, such as soaking time, corrosion inhibitor content, transfer resistance, corrosion inhibition efficiency and the like, and forming an original data feature set; secondly, preprocessing the collected original data; then key functional groups and environmental condition characteristics influencing the corrosion inhibition efficiency are identified based on importance ranking; and finally, training the machine learning regression model by using the training set, and evaluating the performance of the model through the error index of the test set to obtain a prediction model capable of predicting the corrosion inhibition efficiency under different corrosion inhibitor molecular structures. The method has the advantages of being high in prediction accuracy, high in interpretability and good in expansibility, and efficient technical support is provided for research, development, screening and application of the steel bar corrosion inhibitor.

Description

Technical Field

[0001] The invention belongs to the technical field of rust inhibitor performance prediction, and specifically relates to a steel bar rust inhibitor performance prediction method based on functional group classification and machine learning. Background Art

[0002] Reinforced concrete is widely used in the fields of construction, transportation, marine engineering, etc. due to its superior mechanical properties and cost advantages. However, in chloride erosion, carbonization or other multi-factor coupled corrosive environments, the steel bars in concrete are very prone to corrosion, causing concrete cracks and even structural damage. There are currently two types of steel bar rust prevention measures. One is to improve the protective ability of concrete itself, such as using high-performance concrete; the second is "additional measures", such as electrochemical desalination, film covering protection, cathodic protection, adding steel bar rust inhibitors, etc. Steel bar rust inhibitors are economical, practical, and easy to operate, and are widely used in steel bar rust prevention. There are many types of rust inhibitors, and the rust prevention principles of different rust inhibitors are different. Their working principles are closely related to the active functional groups they contain. The core difference lies in the different mechanisms of the active functional groups contained in the metal surface to form a protective film or undergo a complex reaction. Existing research on the design of rust inhibitors mostly relies on empirical mix proportions or small-scale experimental screening, and it is difficult to timely and systematically compare key indicators such as the rust prevention efficiency of diversified mix proportions.

[0003] With the rapid development of big data and machine learning technology, the field of materials has begun to try to use data-driven methods to predict and optimize material properties. However, the current multi-functional group design of steel bar rust inhibitors still lacks mature and effective data accumulation and modeling methods; in addition, there are large differences in the dimensions, types and quantities of various active functional groups. If there is a lack of systematic data classification and feature extraction rules, machine learning models often find it difficult to obtain high-precision predictions. Based on this, it is necessary to propose a technical solution that can effectively classify and encode the characteristics of the active functional groups of rust inhibitors, and use machine learning to fully explore the relationship between the functional group type of steel bar rust inhibitors and the rust inhibition efficiency, in order to reduce R&D costs and accelerate material performance evaluation and optimization design. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for predicting the performance of steel bar rust inhibitors based on functional group classification and machine learning. By systematically classifying and encoding active functional groups and combining multidimensional indicator data (such as immersion time, rust inhibitor content, transfer resistance, rust inhibition efficiency, etc.), a machine learning model is used to achieve rapid and accurate prediction of the performance of steel bar rust inhibitors, thereby overcoming the problems of time-consuming, experience-dependent, and insufficient mechanism analysis in the existing screening and performance evaluation of steel bar rust inhibitor formulas.

[0005] The present invention provides the following technical solutions:

[0006] A method for predicting the performance of a steel bar rust inhibitor based on functional group classification and machine learning comprises the following steps:

[0007] (1) Functional group classification and data collection:

[0008] The active groups in the molecular structure of steel bar rust inhibitors were systematically classified according to different functional group types, the functional group types and quantities of each rust inhibitor sample were collected, and the corresponding immersion time, rust inhibitor content, transfer resistance, and rust inhibition efficiency data were extracted to form the original data feature set;

[0009] (2) Data preprocessing:

[0010] The collected raw data is preprocessed by removing all zero features, eliminating outliers, and standardizing, and the functional group classification information and quantity are converted into feature parameters that can be identified by machine learning;

[0011] (3) Feature analysis:

[0012] Conduct feature importance analysis, identify key functional groups and environmental condition characteristics that affect rust inhibition efficiency, and build a simplified model;

[0013] (4) Machine learning model training and evaluation:

[0014] The machine learning regression model was trained using the training set, and the model performance was evaluated using the error index of the test set to obtain a prediction model that can predict the rust inhibition efficiency under different rust inhibitor molecular structures.

[0015] (5) Prediction of steel bar rust inhibitor performance:

[0016] For the steel bar rust inhibitor to be predicted, its characteristics are preprocessed and then input into the prediction model established in step (4), and the corresponding prediction results are output.

[0017] Furthermore, in step (1), the active groups are classified into eight categories, namely: amino groups, aromatic amines, alcohols, alcoholamines, nitrogen-containing heterocyclic groups, carboxylic acids, phosphates, esters and fatty acid esters;

[0018] There are twenty types of functional groups, namely: hydroxyl, carboxyl, amino, amide, imino, nitro, ether, ester, carbonyl, thioketone, cyano, imidazoline, imidazole, triazole, pyridine, thiol, sulfonic acid, phosphoric acid, halogen atom, and phenyl.

[0019] Furthermore, in step (2), the feature columns that are all zero or seriously missing in the original data feature set are deleted, individual extreme outliers are excluded, and invalid features are removed; the classification information of the active groups and functional groups is numerically encoded; and the continuous numerical features are standardized so that the mean is 0 and the standard deviation is 1.

[0020] Furthermore, in step (3), the CatBoost model is used to rank the feature importance, and the correlation heat map analysis is used to evaluate the nonlinear correlation between the active groups and functional groups and between the active groups and the rust inhibition efficiency index. Based on the feature importance ranking and correlation analysis results, the top 12 most influential features are retained to participate in subsequent machine learning training and prediction tasks.

[0021] Furthermore, in step (3), the steel bar rust inhibitor performance prediction method based on functional group classification and machine learning also performed PCA feature dimensionality reduction analysis and interpretability analysis.

[0022] Furthermore, in step (3), the steel bar rust inhibitor performance prediction method based on functional group classification and machine learning also uses a three-dimensional scatter plot to preliminarily observe the sample distribution of different functional group combinations. In the three-dimensional scatter plot, a 1% dynamic adjustment random fine-tuning offset is used according to the axis range to avoid excessive overlap of sample data points of multiple functional groups.

[0023] Furthermore, in step (4), the data is divided into a training set and a test set in a certain ratio, and a CatBoost machine learning regression model suitable for tabular data and nonlinear features is modeled using the training set. The model is trained and validated, and the network model hyperparameters are adjusted according to the error index of the test set. The model configuration is further optimized using cross-validation or grid search methods to obtain a prediction model that can predict the rust inhibition efficiency under different rust inhibitor molecular structures.

[0024] Furthermore, in step (4), the performance of the machine learning model is intuitively evaluated by drawing a regression fitting graph of the true value and the predicted value and a comparison graph of the prediction results of the test set; and the error indicators mean square error, root mean square error, mean absolute error, mean absolute percentage error and determination coefficient are combined to evaluate the fitting ability of the prediction model for the rust inhibition efficiency under different rust inhibitor molecular structures.

[0025] The principle of the present invention is that the present invention establishes a data set based on the classification rules of the active functional groups of the rust inhibitor, combines the machine learning method to carry out regression training on the rust inhibition efficiency, and realizes the accurate prediction of the rust inhibition performance of the steel bar rust inhibitor.

[0026] Compared with the prior art, the present invention has the following significant advantages:

[0027] (1) The present invention groups active functional groups according to the proposed classification rules, and the classification level is clear and reasonable, which is convenient for data collation and modeling analysis. Combined with machine learning algorithms, the complex nonlinear relationship between multiple functional groups and rust inhibition efficiency can be captured, and the prediction accuracy is significantly better than traditional empirical methods or single factor experiments;

[0028] (2) When new rust inhibitors or other active groups are introduced, feature extraction and model update can be performed according to the same classification principle, which is scalable. Use highly explanatory visualization tools such as SHAP to analyze the contribution of each functional group category to rust inhibition efficiency, providing R&D personnel with intuitive mechanism analysis and improvement directions;

[0029] (3) In engineering applications, a large number of candidate mixes can be quickly virtually screened, reducing repeated experiments and lowering the research and development cost of steel bar rust inhibitors. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. Among them:

[0031] Figure 1 Schematic diagram of classification of active functional groups of the present invention;

[0032] Figure 2 It is a schematic diagram of a three-dimensional scatter plot of functional group data distribution of the present invention;

[0033] Figure 3 It is a three-dimensional PCA analysis diagram of feature dimension reduction of the present invention;

[0034] Figure 4 It is the feature correlation matrix heat map of the present invention;

[0035] Figure 5 This is a graph of the importance of features of the present invention;

[0036] Figure 6 It is the SHAP summary diagram of the present invention;

[0037] Figure 7 It is the regression fitting diagram of the true value and the predicted value of the present invention;

[0038] Figure 8 This is a comparison chart of the test set prediction results of the present invention. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0040] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. In this embodiment, 330 groups of steel bar rust inhibitor sample information in the prior art are collected, and the molecular structure of each rust inhibitor contains at least one active group.

[0041] A method for predicting the performance of a steel bar rust inhibitor based on functional group classification and machine learning comprises the following steps:

[0042] (1) Functional group classification and data collection:

[0043] According to the classification rules of active functional groups in the molecular structure of steel bar rust inhibitor, the number was recorded, and the immersion time (h), rust inhibitor content (%), rust inhibition efficiency η (%) and transfer resistance Rct (kΩ·cm 2 ) and other data to form the original data feature set; the original data information summary of 330 sets of steel bar rust inhibitor sample data is shown in Table 1;

[0044] Table 1 Summary of original data of 330 groups of steel bar rust inhibitor samples

[0045]

[0046]

[0047] In Table 1, the mean, standard deviation and five-number summary are used to describe the original data set of the steel bar rust inhibitor sample data; the minimum value, 0.25 (first quartile), 0.5 (second quartile), 0.75 (third quartile) and maximum value in the title row of Table 1 are the five key values ​​summarized by the five-number summary. They are important components of descriptive statistics and help to quickly understand the characteristics of the data set.

[0048] As shown in Table 1, Figure 1 As shown in the figure, the rule for dividing active functional groups is to divide each active group into eight categories and record the number of each functional group type; the eight categories of active groups are: 1-amino, 2-aromatic amines, 3-alcohols, 4-alcoholamines, 5-nitrogen-containing heterocyclics (such as pyridine, imidazole, etc.), 6-carboxylic acids, 7-phosphates, 8-esters and fatty acid esters; there are 20 types of functional groups, namely: hydroxyl (-OH), carboxyl (-COOH), amino (-NH2), amide (-CONH2), imino (=NH), nitro (-NO2), ether (-O-), ester (-COOR), carbonyl (-C=O), thioketone, cyano (-CN), imidazoline, imidazole, triazole, pyridyl (-C5H5N), thiol (-SH), sulfonic acid (-SO3H), phosphate (-PO4), halogen atom (-X), phenyl. Figure 1 The numbers in the table represent the number of functional groups in the eight types of active groups.

[0049] (2) Data preprocessing:

[0050] The collected raw data is preprocessed by removing all zero features, eliminating outliers, and standardizing, and the functional group classification information and quantity are converted into feature parameters that can be identified by machine learning;

[0051] Remove the columns that are all 0 in all samples to exclude extreme abnormal data caused by individual statistical errors; standardize continuous features such as "immersion time", "rust inhibitor content", and "transfer resistance" to make the mean 0 and the standard deviation 1 to eliminate the impact of dimensional differences on model training; use numerical coding for different functional group categories and include them in the feature matrix X of machine learning (functional group categories will be part of the feature matrix X after coding); form the target value rust inhibition efficiency η (%) into a vector y to prepare for subsequent model training and prediction. For example, for Table 1, the columns that are all 0 in all samples include nitro (-NO2) and thiol (-SH), which should be deleted; the 3σ criterion shows that there are extreme abnormal data columns caused by individual statistical errors in columns such as "immersion time" and "rust inhibition efficiency". Numerical coding is performed according to the 8 types of active groups. For example, "amino" is regarded as feature 1, "aromatic amine" is regarded as feature 2, and so on. The corresponding quantities are also included in the feature matrix. Figure 1 The first column is the 8 types of active groups after numerical coding. Figure 1 This is only an example of the classification rule of active groups. In other embodiments, other classification rules may be used to classify and numerically encode active groups.

[0052] (3) Feature analysis:

[0053] Based on PCA (principal component analysis) feature dimensionality reduction analysis, feature correlation analysis, importance ranking and interpretability analysis, key functional groups and environmental condition characteristics that affect rust inhibition efficiency are identified;

[0054] Use the three-dimensional scatter plot (immersion time, rust inhibitor content, rust inhibition efficiency) to preliminarily observe the distribution of samples with different functional group combinations. Figure 2 is a three-dimensional scatter plot of the functional group data distribution. Figure 2 In the example, a 1% dynamic adjustment random fine-tuning (jitter) offset is used according to the axis range to avoid excessive overlap of sample data points of multiple functional groups; Figure 2 It can be seen that the functional group data are mostly distributed in the area of ​​low "rust inhibitor content", short "immersion time" and high "rust inhibition efficiency", indicating that most steel bar rust inhibitors are used in small amounts but have high efficiency. Their functional groups can quickly generate protective films or react in a short period of time, and can play a greater role in rust inhibition; among them, the data points of the functional groups "hydroxyl" and "amino" are widely scattered, with both high efficiency and low efficiency, which may indicate that the functional group itself is not the only dominant factor, and it is also necessary to combine "rust inhibitor content", "immersion time" or other functional group conditions to exert a stable rust inhibition effect; the functional groups "hydroxyl", "carboxyl" and "amino" are easy to form clusters, indicating that they have similar effects and have a certain synergistic effect;

[0055] like Figure 3As shown in the figure, PCA (n_components = 3) is used to reduce the dimension of the feature matrix, generate three-dimensional principal component coordinates, and use the rust inhibition efficiency value as the color mapping to intuitively understand the distribution characteristics of the sample in the principal component coordinate system from a macro level. Figure 3 It can be found that the data point cloud has a large dispersion and a wide color range, indicating that the samples have significant differences in the main feature dimensions, and the difference in rust prevention efficiency may be due to the influence of multiple factors.

[0056] The characteristic correlation analysis is presented through the correlation heat map to evaluate the nonlinear correlation between each functional group and the rust inhibition efficiency index. Figure 4 The correlation between the sample data features in Table 1 is shown. For example, Figure 4 It can be seen that "thioketone" has a high positive correlation with "amino", "imino" and "ether" functional groups, indicating that these functional groups may often exist at the same time; "sulfonic acid group" is negatively correlated with "rust inhibition efficiency", indicating that the two may have an inverse relationship;

[0057] Extract the feature importance ranking based on the CatBoost model to identify which key functional groups and environmental characteristics most significantly affect the rust inhibition efficiency. The results are as follows: Figure 5 As shown, the importance of each feature is: rust inhibitor content (%) > immersion time (h) > transfer resistance Rct (kΩ·cm 2 )>carboxyl (-COOH)>amino (-NH2)>hydroxyl (-OH)>phenyl>phosphate (-PO4)>imino (=NH)>imidazoline>thioketone>imidazole>carbonyl (-C=O)>amide (-CONH2)>sulfonic acid (-SO3H)>ether (-O-)>halogen atom (-X)>pyridyl (-C5H5N)>triazole>cyano (-CN)>ester (-COOR); The results show that active functional groups such as carboxyl (-COOH), amino (-NH2), hydroxyl (-OH) and phenyl contribute more to the rust inhibition efficiency;

[0058] like Figure 6 As shown in the figure, SHAP is also used to draw an interactive dependency diagram to observe the positive / negative impact of different features on the model prediction output, providing R&D personnel with an explainable mechanism analysis and rust inhibitor design basis to guide subsequent design; SHAP is the abbreviation of "SHapley Additive exPlanations", which is a method for explaining machine learning model predictions. Figure 6As shown in the figure, the features such as "rust inhibitor content", "transfer resistance", "carboxyl", "immersion time", "amino", "hydroxyl" and "phenyl" are located at the top, indicating that they have a greater impact on the model output (prediction of rust inhibition efficiency); among them, the "rust inhibitor content" feature has both low value (blue) points distributed in the positive SHAP area and some high value (red) points distributed in the negative SHAP area on the same row, and the scattered distribution is relatively dispersed, indicating that the relationship between this feature and the target value is not a simple linear monotonic, and may also be affected by other features;

[0059] In order to reduce the feature dimension, only the top 12 most important features are retained for subsequent machine learning training and prediction tasks according to the feature importance sorting results, reducing redundant features to build a simplified model and improve the model prediction speed and accuracy. PCA analysis, interpretability analysis, correlation analysis and three-dimensional scatter plots are all feature analysis to understand the influence of the main features, assist in feature screening, and assist in rust inhibitor design.

[0060] (4) Machine learning model training and evaluation:

[0061] The machine learning regression model was trained using the training set, and the model performance was evaluated using the error index of the test set to obtain a prediction model that can predict the rust inhibition efficiency under different rust inhibitor molecular structures.

[0062] The processed data is divided into training set and test set in a ratio of 80%:20%. The CatBoost machine learning regression model suitable for tabular data and nonlinear features is built using the training set, and the error indicators (MSE, MAE, R 2 , MAPE) to adjust the network model hyperparameters (such as tree depth, learning rate, number of iterations, etc.), use cross-validation or grid search method to further optimize the model configuration, and finally obtain a prediction model (regression prediction model) that can predict the rust inhibition efficiency under different rust inhibitor molecular structures: depth = 8, learning_rate = 0.05, iterations = 1000, l2_leaf_reg = 10, random_state = 42, verbose = 0 (tree depth is 8, learning rate is 0.05, iterations are 1000 times, L2 leaf node regularization item is 10, random state is 42, and the training process does not output detailed information);

[0063] Grid Search and 5-fold cross validation were used to further adjust model hyperparameters, iterated multiple times on the training set and evaluated on the test set;

[0064] By drawing the regression fitting graph of the true value and the predicted value and the comparison graph of the prediction results of the test set, the performance of the machine learning model can be intuitively evaluated. Figure 7 , the test set prediction results are compared in Figure 8 The error evaluation indicators are mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R 2 );

[0065] From the training set and test set error results in Table 2, we can see that R 2 They are all greater than 0.90, indicating that the established machine learning model has a good fitting ability for the rust inhibition efficiency and is suitable for carrying out regression training and prediction tasks of steel bar rust inhibitor performance based on functional group classification.

[0066] Table 2 Comparison of errors between training set and test set

[0067] Error indicators MSE RMSE MAE MAPE <![CDATA[R 2 <!-- 6 -->]]> Training set 1.57264654 1.254052 0.964139 0.037705 0.998414 Test Set 200.597024 14.16323 7.692063 9.13E+13 0.90132

[0068] (5) Prediction of steel bar rust inhibitor performance:

[0069] For the steel bar rust inhibitor to be predicted, the sample features of the steel bar rust inhibitor are preprocessed according to step (2) and then input into the trained regression prediction model to output the corresponding prediction results.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are within the scope of protection of the pending claims of the present invention.

Claims

1. A method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning, characterized in that: The following steps are involved: (1) Functional group classification and data collection: The active groups in the molecular structure of steel bar rust inhibitors were systematically classified according to different functional group types, the functional group types and quantities of each rust inhibitor sample were collected, and the corresponding immersion time, rust inhibitor content, transfer resistance, and rust inhibition efficiency data were extracted to form the original data feature set; (2) Data preprocessing: The collected raw data is preprocessed by removing all zero features, eliminating outliers, and standardizing, and the functional group classification information and quantity are converted into feature parameters that can be identified by machine learning; (3) Feature analysis: Conduct feature importance analysis, identify key functional groups and environmental condition characteristics that affect rust inhibition efficiency, and build a simplified model; (4) Machine learning model training and evaluation: The machine learning regression model was trained using the training set, and the model performance was evaluated using the error index of the test set to obtain a prediction model that can predict the rust inhibition efficiency under different rust inhibitor molecular structures. (5) Prediction of steel bar rust inhibitor performance: For the steel bar rust inhibitor to be predicted, its characteristics are preprocessed and then input into the prediction model established in step (4), and the corresponding prediction results are output.

2. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (1), the active groups are divided into eight categories, namely: amino, aromatic amine, alcohol, alcohol amine, nitrogen-containing heterocyclic, carboxylic acid, phosphate, ester and fatty acid ester; There are twenty types of functional groups, namely: hydroxyl, carboxyl, amino, amide, imino, nitro, ether, ester, carbonyl, thioketone, cyano, imidazoline, imidazole, triazole, pyridine, thiol, sulfonic acid, phosphoric acid, halogen atom, and phenyl.

3. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (2), the feature columns that are all zero or seriously missing in the original data feature set are deleted, individual extreme outliers are excluded, and invalid features are removed; the classification information of the active groups and functional groups is numerically encoded; and the continuous numerical features are standardized so that the mean is 0 and the standard deviation is 1.

4. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (3), the feature importance is ranked by using the CatBoost model, and the nonlinear correlation between each active group and functional group and between each other and the rust inhibition efficiency index is evaluated by correlation heat map analysis; According to the feature importance ranking and correlation analysis results, the top 12 most influential features are retained to participate in subsequent machine learning training and prediction tasks.

5. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (3), feature correlation analysis, PCA feature dimensionality reduction analysis and interpretability analysis were also performed.

6. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (3), a three-dimensional scatter plot is also used to preliminarily observe the sample distribution of different functional group combinations. In the three-dimensional scatter plot, a 1% dynamic adjustment random fine-tuning offset is used according to the axis range to avoid excessive overlap of sample data points of multiple functional groups.

7. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (4), the data is divided into a training set and a test set in a certain ratio, and a CatBoost machine learning regression model suitable for tabular data and nonlinear features is modeled using the training set. The model is trained and verified, and the network model hyperparameters are adjusted according to the error index of the test set. The model configuration is further optimized using cross-validation or grid search methods to obtain a prediction model that can predict the rust inhibition efficiency under different rust inhibitor molecular structures.

8. The method for predicting the performance of steel bar rust inhibitor based on functional group classification and machine learning according to claim 1, characterized in that: In step (4), the performance of the machine learning model is intuitively evaluated by drawing a regression fitting graph of the true value and the predicted value and a comparison graph of the test set prediction results; and the error indicators mean square error, root mean square error, mean absolute error, mean absolute percentage error and determination coefficient are combined to evaluate the fitting ability of the prediction model for the rust inhibition efficiency under different rust inhibitor molecular structures.

Images