Multi-stage cooperative machine learning prediction method for corrosion degree of coating material

Abstract

The invention discloses a multi-stage cooperative machine learning prediction method for the corrosion degree of a coating material. According to the first level of the method, feature data of a natural environment are extracted to serve as feature parameters, dimension reduction and noise reduction processing is conducted on the feature parameter data through a principal component analysis algorithm, the processed features serve as input parameters, and corresponding coating apparent performance data serve as output parameters; and establishing a machine learning prediction model between the environmental factors and the apparent performance of the coating. In the second level, an indoor accelerated experiment is utilized to construct a data set of the apparent performance and the corrosion degree, the apparent performance of the coating serves as an input parameter, the corresponding corrosion performance of the coating serves as an output parameter, and a machine learning prediction model between the apparent performance and the corrosion degree is established. And combining the two models to predict the corrosion degrees of the coating materials in different regions. The method has the advantages of being high in model generalization, capable of accurately judging the damage condition of the coating and the like, and effective guidance is provided for coating material selection design oriented to environmental characteristics.

Description

Technical Field

[0001] The present invention belongs to the technical field of predicting the corrosion degree of coatings during service, and particularly relates to a multi-stage collaborative machine learning prediction method for the corrosion degree of coating materials. Background Art

[0002] Due to their excellent properties and high cost performance, coating materials have been widely used in many fields, including automobiles, electronic devices, construction, packaging, chemical engineering, and bioengineering. However, in practical applications, coating materials are affected by environmental factors, mechanical stress, and chemical corrosion during production, storage, and use, and may exhibit deterioration phenomena such as loss of gloss, fading, cracking, and peeling. These deteriorations not only reduce the protective performance of the coatings but may also shorten the service life of the substrates. Therefore, the durability and service life prediction of coating materials have become key issues of concern in the industry.

[0003] Currently, the existing methods for predicting the aging degree of organic coating materials mainly include the following:

[0004] (1) Accelerated aging test method

[0005] This method accelerates the aging process of coating materials by simulating extreme conditions in the actual application environment (such as high temperature, high humidity, ultraviolet irradiation, salt spray corrosion, etc.). Based on the performance decay of the coatings under accelerated conditions, their service life under normal use conditions is estimated. The aging reaction rate constant K and the aging temperature T follow the Arrhenius equation, and the calculation formula is as follows:

[0006]

[0007] Among them, K is the aging reaction rate constant, A is the pre-exponential factor, Ea is the activation energy of the reaction, T is the absolute temperature, and R is the gas constant. However, this method has obvious defects: the aging mechanisms at different temperatures may be different, resulting in differences in the aging rate constants; at the same time, this method ignores the influence of environmental humidity, reducing the accuracy of the test results.

[0008] (2) On-site monitoring method

[0009] Electrochemical impedance spectroscopy (EIS) is a method for real-time monitoring of the impedance characteristics of coatings and predicting their corrosion behavior. The impedance prediction formula is as follows:

[0010]

[0011] Among them, Z(ω) is the impedance at frequency ω, Rs is the solution resistance, Rp is the polarization resistance, C is the capacitance, and j is the imaginary unit. However, this method also has some deficiencies: the monitoring equipment is expensive and requires professional operators; the monitoring data is vulnerable to environmental noise interference, which may lead to inaccurate predictions; in addition, long-term continuous monitoring is time-consuming and complex. Summary of the Invention

[0012] Aiming at the deficiencies of the prior art, the present invention proposes a method for predicting the corrosion degree of a coating based on actual environmental monitoring. This method first constructs an association model between environmental factors and changes in the apparent properties of materials by collecting environmental data in real time and combining machine learning algorithms. Subsequently, a mapping relationship between the apparent properties and the corrosion degree of the coating is further established, and the specific effects of various environmental factors on the material properties are analyzed in depth. Finally, a method for predicting the corrosion degree of the coating with higher accuracy and wider applicability is developed, effectively improving the accuracy and reliability of the prediction.

[0013] The above technical problems to be solved by the present invention can be achieved through the following technical solutions:

[0014] A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material, characterized in that

[0015] First, use environmental factor data to predict changes in the apparent properties of materials, and then further predict the corrosion degree of the materials based on these apparent properties; through this method, the corrosion behavior of the coating material under different environmental conditions can be evaluated more accurately, the prediction accuracy of the coating life and protection performance can be improved, and thus a scientific basis can be provided for the selection and application of materials.

[0016] Furthermore, the prediction method of the present invention is divided into two stages. In the first stage, by analyzing environmental factors in different regions, including temperature, humidity, air pressure, and solar sunshine hours, a relationship model between environmental data and changes in the apparent properties of the coating is established; the input of this relationship model is environmental data, and the output is the changes in the apparent properties of the coating, including glossiness, roughness, adhesion, and contact angle, so as to predict the influence of environmental conditions on the coating properties; through this relationship model, the influence of environmental factors on the apparent properties of the coating can be effectively quantified, laying a foundation for subsequent corrosion prediction;

[0017] In the second stage, according to the data of the changes in the apparent properties of the coating material obtained in the first stage, a prediction model between the apparent properties and the corrosion degree is further established; this prediction model uses the apparent property parameters as the input and outputs the corrosion degree of the coating; through this prediction model, the influence law of the apparent properties on the corrosion degree of the coating can be evaluated more accurately, and then the corrosion resistance of the coating under different environments can be deduced, thus providing support for the design and optimization of protective materials.

[0018] Further, the first stage includes the following steps:

[0019] 3.1 Select coating materials, conduct outdoor exposure experiments in different regions, and obtain the apparent performance parameters and corrosion degree parameters of the coating materials during the experimental period;

[0020] 3.2 Collect the corresponding environmental data in step (1), including temperature, humidity, sunshine hours, air pressure, wind speed, cloud cover, and rainfall;

[0021] 3.3 Extract the characteristic data from the environmental data obtained in step (2) as characteristic parameters, and use Spearman and principal component analysis algorithms to perform dimensionality reduction and noise reduction processing on the characteristic parameter data, where the characteristic data is the statistical value of different environmental data;

[0022] 3.4 Group the data from different regions into a test set and a validation set for constructing a coating apparent performance prediction model;

[0023] 3.5 Use one or several of recurrent neural network, self-supervised learning, gradient boosting tree, K-nearest neighbor algorithm, and random forest algorithm to construct a coating apparent performance prediction model with the environmental characteristics obtained after dimensionality reduction as input parameters and the coating apparent performance of the training set as output parameters;

[0024] 3.6 Use the characteristic environmental factors of the test set as input parameters to predict the material performance changes in different regions, obtain the apparent performance of its service coating, and at the same time, calculate the square value R of the correlation coefficient between it and the experimental value 2 , when R 2 ≥85%, the prediction result is credible.

[0025] Further, the second stage includes the following steps:

[0026] 4.1 Conduct corrosion aging tests on the coating materials, and obtain the apparent performance parameters and corrosion degree parameters of the coating materials during the aging process;

[0027] 4.2 Collect the corresponding apparent performance data in step 4.1, including gloss, adhesion, roughness, contact angle, corrosion degree parameters, and impedance values within 1 - 7 days of immersion;

[0028] 4.3 Extract the apparent performance data obtained in step 4.2, and preprocess the corrosion degree data and the apparent performance data; the preprocessing methods for the corrosion evaluation data and the apparent performance data include one or several of normalization, standardization, and taking logarithms;

[0029] 4.4 Group the apparent performance - corrosion degree of different samples, use part of the data as the training set for constructing a coating corrosion degree prediction model, and use the remaining data as the test set for verifying the coating corrosion degree prediction model;

[0030] 4.5 Take the apparent performance of the training set as the input parameter and the coating corrosion degree of the training set as the output parameter, and use one or several of the algorithms such as support vector machine, AdaBoost, K-nearest neighbor algorithm, and random forest to construct an apparent performance-corrosion degree prediction model;

[0031] 4.6 Take the apparent performance of the test set as the input parameter, predict the material corrosion degree, and at the same time, calculate the square value R of the correlation coefficient between it and the experimental value 2 , when R 2 ≥80%, the prediction result is credible;

[0032] 4.7 Take the prediction result of the environmental parameter-apparent performance model as the input parameter, and use the corrosion degree model in 4.5 to predict the coating performance, forming an overall coating corrosion degree prediction model of "environmental factor-surface performance-corrosion degree".

[0033] Furthermore, the coating material is a composite material of one or several of polyurethane, epoxy resin, and acrylic resin.

[0034] Furthermore, the apparent performance parameters of the coating material in step 3.1 include optical performance, mechanical performance, and surface performance.

[0035] Furthermore, the apparent performance parameters of the coating material in step 3.1 are roughness, gloss, adhesion, contact angle, and the corrosion performance parameters include the low-frequency impedance modulus value (|Z| 0.01Hz ) within 7 days of coating immersion and.

[0036] Furthermore, the experimental period in steps 3.1 to 3.2 is 1-5 years, that is, the performance of the coating material is recorded 1-5 years after sampling, and the environmental data is recorded every 1-24h.

[0037] Furthermore, the process of obtaining the statistical values of different environmental data in step 3.6 includes: using Python software to statistically analyze the environmental data of temperature, humidity, and sunshine hours, and obtaining the performance change data of the coating material within the experimental period and the statistical values under the corresponding temperature, humidity, and sunshine hours.

[0038] Furthermore, the corrosion aging experiment in step 4.1 is an indoor accelerated experiment.

[0039] The multi-stage collaborative machine learning prediction method for the corrosion degree of the coating material as described above specifically includes the following steps:

[0040] (1) Select the coating material and conduct outdoor aging tests in different regions. During the aging process, obtain the apparent performance parameters and corrosion degree of the coating material within the experimental period;

[0041] (2) Collect the corresponding environmental data in step (1), including temperature, humidity, sunshine hours, air pressure, wind speed, cloud cover, and rainfall;

[0042] (3) Extract the characteristic data from the environmental data obtained in step (2) as characteristic parameters, and use the Spearman and principal component analysis algorithms to perform dimensionality reduction and noise reduction processing on the characteristic parameter data, where the characteristic data is the statistical value of different environmental data;

[0043] (4) Group the environmental factors in different regions and the changes in the apparent properties of the material. The change data sets of different apparent properties are independent of each other. Divide the data set into a training set and a validation set

[0044] (5) Take the environmental factors obtained after dimensionality reduction as input parameters, and take the apparent properties of the coating in the training set as output parameters. Use Python to construct a semi-supervised machine learning algorithm to train the coating apparent property prediction model and construct the coating apparent property prediction model;

[0045] (6) Calculate the square value R2 of the correlation coefficient between the model prediction value and the experimental value. When R2 ≥ 85%, the prediction result is credible.

[0046] (7) Then, in order to study the relationship between the apparent properties of the coating and the corrosion degree of the coating, carry out an indoor aging test on the coating material, and obtain the apparent properties and corrosion degree of the coating material during the experimental period in the indoor aging process;

[0047] (8) Collect the corresponding corrosion degree data and apparent property data in step (7), including gloss, adhesion, roughness, and contact angle;

[0048] (9) Extract the corrosion degree data and apparent property data obtained in step (8), and perform preprocessing on the corrosion degree data and apparent property data;

[0049] (10) Group the apparent property - corrosion degree of different samples. Use some regions as the training set for the construction of the coating corrosion degree prediction model, and use the remaining regions as the test set for the verification of the coating corrosion degree prediction model;

[0050] (11) Take the apparent properties of the training set as input parameters, and take the corrosion property changes of the training set as output parameters. Use Python to construct a coating apparent property - corrosion degree prediction model;

[0051] (12) Take the apparent properties of the test set as input parameters, predict the corrosion degree of its coating, obtain its predicted corrosion degree, and calculate the square value R2 of the correlation coefficient between it and the experimental value. When R2 ≥ 80%, the prediction result is credible;

[0052] (13) Use the apparent performance prediction results obtained in (6) as input parameters, and use the coating apparent performance - old corrosion degree prediction model in (11) to predict the coating performance, forming a "environmental factor - apparent performance - corrosion degree" coating material corrosion degree prediction model.

[0053] (14) Use the environmental factors as input parameters and the corrosion degree as output parameters to predict the corrosion degree of the coating material. At the same time, calculate the square value R2 of the correlation coefficient between the predicted value and the experimental value. When R2 ≥ 80%, the prediction result is credible.

[0054] In the present invention, performance data of the coating material at different corrosion degrees are obtained through natural aging experiments at the test station, and environmental data during the aging process are recorded simultaneously, including temperature, humidity, sunshine hours, etc. By extracting natural environment characteristic data (such as sunshine hours, maximum temperature, minimum temperature, average temperature, annual rainfall, relative humidity, average air pressure, wind speed, cloud amount, etc.) as characteristic parameters, the data is subjected to dimensionality reduction and denoising processing using Spearman correlation coefficient analysis and principal component analysis algorithms, and key environmental factors are screened out. These key environmental factors are used as input parameters, and the apparent performance of the material (such as roughness index, glossiness, adhesion, and contact angle, etc.) is used as output parameters to construct an apparent performance prediction model. Subsequently, using the apparent performance parameters as input and the low-frequency impedance lg(|Z|0.01Hz) value within 7 days of the coating as output, machine learning is carried out in two steps using Python to establish a corrosion degree prediction model, and finally, accurate prediction of the service corrosion degree of the coating material in different regions is achieved.

[0055] In the above multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material:

[0056] Preferably, the coating material in step (1) is a composite material of one or more of polyurethane, epoxy resin, and acrylic resin.

[0057] Preferably, the different regions in step (1) include Malaysia, Jeddah, Saudi Arabia, Sri Lanka, Russia, and Jizan, Saudi Arabia.

[0058] Preferably, the aging test in step (1) is a natural aging experiment, and the aging test in step (7) is an indoor accelerated aging experiment.

[0059] Preferably, the apparent performance parameters of the coating material in step (1) include optical performance, mechanical performance, and surface performance.

[0060] Preferably, the performance parameters of the polymer material in step (1) are one or more of roughness, transparency, tensile strength, melting temperature, loss of gloss, glass transition temperature, and initial decomposition temperature.

[0061] The experimental period from step (1) to step (2) and the recording time of environmental data can be flexibly adjusted according to the actual situation. Specifically, the recording frequency of environmental data should match the length of the experimental period. When the experimental period is relatively long, the recording time interval can be appropriately extended to ensure that sufficient and representative data samples can be collected within the experimental period.

[0062] Preferably, the experimental period described in steps (1) to (2) is 1 - 3 years, that is, the performance of the coating material is recorded 1 - 3 years after the sample is put in, and the environmental data is recorded every 3 hours.

[0063] Preferably, the acquisition of environmental data in step (2) can be through on-site measurement in the test field, or through information published on the website to obtain local climate environmental data such as temperature, humidity, and sunshine hours.

[0064] Preferably, the process of obtaining statistical values of different environmental data in step (3) includes: using Python software to perform statistics on environmental data such as temperature, humidity, and sunshine hours, and obtaining the performance change data of the coating material during the experimental period and the statistical values corresponding to the temperature, humidity, and sunshine hours.

[0065] The time cumulative values of the performance change data of the coating material during the experimental period and the corresponding temperature, humidity, and sunshine hours are listed as follows: average temperature, maximum temperature, minimum temperature, average air pressure, average dew point temperature, average wind speed, annual average cloud cover, total rainfall, sunshine hours, average relative humidity, etc.

[0066] For example, in step (3), by extracting characteristic data of the natural environment (such as total rainfall, sunshine hours, annual average relative humidity, annual average solar sunshine hours, etc.) as characteristic parameters, the Spearman correlation coefficient analysis and principal component analysis algorithms are used to perform dimensionality reduction and noise reduction processing on the characteristic parameter data.

[0067] Taking the obtained several environmental data as characteristic data, and then as input parameters, taking the apparent performance of the corresponding material (such as the roughness, gloss, adhesion, and contact angle of the material, etc.) as output parameters, and then taking these output parameters as the input of the next model, and taking the corrosion performance parameter including the low-frequency impedance lg(|Z|0.01Hz) of the coating on the 7th day as the output of the next model, using Python software to perform machine learning on the relationship between the environment and the corrosion performance change in two steps, and constructing a corrosion degree prediction model for predicting the service corrosion degree of coating materials in different regions in the next step.

[0068] Preferably, the machine learning algorithm in step (5) is one or several of the algorithms such as support vector machine, AdaBoost, K-nearest neighbor algorithm, and random forest.

[0069] Furthermore, as a preferred embodiment of the present invention, a multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material provided by the present invention includes the following steps:

[0070] (1) Select the coating material as the test object to carry out the natural aging test. The material samples are defect-free and have high consistency. At the same time, set relevant properties (such as roughness index, glossiness, adhesion, and contact angle, etc.) as the life evaluation indicators, and test the initial properties to be investigated of the samples before the test, etc.;

[0071] (2) Carry out the natural aging test of the polymer material at multiple natural environment test stations, obtain the environmental data recorded every 1 h, including temperature, humidity, sunshine hours, etc., and at the same time detect its roughness index, glossiness, adhesion, and contact angle, etc.

[0072] (3) Use Python to statistically analyze the data such as temperature, humidity, and sunshine hours, and obtain the apparent properties and corrosion degree of the material. At the same time, since some environmental factors have correlations such as synergy and resistance, for the obtained data set, use the Spearman correlation coefficient analysis and principal component analysis algorithms to perform dimensionality reduction and noise reduction processing on this data set to complete the characterization of the environmental big data;

[0073] (4) Based on the fact that the environmental data after the characterization process and the data set of the material property changes after the test are relatively small, which belongs to the small sample problem, select the self-supervised learning algorithm and divide the training set and the test set;

[0074] (5) Based on Python, use the self-supervised learning algorithm to train the corrosion degree prediction model for the training set, so as to construct the prediction model. The input parameters of the training set are mainly the screened environmental data;

[0075] (6) Use the environmental data after the characterization process as the input to predict the change in service performance within its cycle, obtain the coating corrosion degree. At the same time, calculate the square value of the correlation coefficient (R2) between it and the experimental value. When R2≥90%, the prediction result is credible.

[0076] In a preferred embodiment of the present invention, taking the coating material sample as the research object, natural aging tests are carried out at natural environment test stations in Singapore, Guangxi, China, Nepal, India, Kalimantan, Indonesia, Thailand, Jeddah, Saudi Arabia, Cirebon, Indonesia, Egypt and Pakistan, etc. During the test process, climate environment data of each station are collected throughout the test cycle, including temperature, humidity, sunshine hours, etc. Based on Python, the data are processed to form statistical features, and the dimensionality reduction and denoising processing are carried out on the feature data set by using the Spearman correlation coefficient analysis and principal component analysis algorithms. Subsequently, taking the characteristic environmental factors as the input and the apparent properties (such as roughness, glossiness, etc.) of the coating material as the output, an apparent property prediction model is trained. Further, taking the apparent property parameters as the input and the low-frequency impedance lg(|Z|0.01Hz) of the coating on the 7th day as the output, a corrosion performance prediction model is constructed through a machine learning algorithm. Finally, a service corrosion degree prediction model is formed by combining the above two steps, which is used to analyze the performance change trend and corrosion degree in different regions. By combining domestic and foreign test data, this method can effectively predict the performance change and corrosion degree of the coating material of equipment products abroad, significantly reduce the test cost, shorten the development cycle, and improve the weather resistance quality, and has important practical application value.

[0077] Compared with the prior art, the present invention has the following advantages:

[0078] (1) Based on the actual environment aging test, the present invention generates environmental big data by real-time monitoring of environmental factors, and combines machine learning algorithms to construct a relationship model between environmental data and the apparent properties and corrosion degree of materials, deeply analyzes the influence of various factors on the coating material, and thus develops a more accurate service corrosion degree prediction method.

[0079] (2) The present invention uses machine learning to perform feature processing on environmental big data, fully excavates the influence of environmental factors on the corrosion degree of coating materials, and realizes the prediction of the service corrosion degree of coating materials in different regions. Its wide applicability can be extended to the corrosion prediction of various coating materials and performance indicators, provides guidance for the weather resistance and corrosion resistance design of equipment products, and supports the international development of products.

[0080] (3) The present invention first applies environmental big data to the prediction of the service corrosion degree of coating materials, deeply excavates the influence information of the environment on the corrosion of materials, and the correlation between the apparent properties of materials and the corrosion degree, and significantly improves the accuracy of model prediction.

[0081] (4) By combining domestic test data with foreign environmental data, the present invention can predict the performance change and service life of coating materials abroad, effectively reduce the test cost, shorten the development cycle, and improve the weather resistance and corrosion resistance quality of materials.

[0082] (5) The method of the present invention is characterized by being convenient, efficient and highly accurate, capable of significantly reducing the test workload, and applicable to guiding the improvement of the weather resistance and corrosion resistance of materials, as well as the design of the weather resistance and corrosion resistance of products. Description of the Drawings

[0083] Figure 1 It is a comparison between the true value and the predicted value of the model. Detailed Embodiments

[0084] The present invention will be described in detail below with reference to the drawings and embodiments. The following embodiments are illustrative and not restrictive, and the protection scope of the present invention cannot be limited by the following embodiments.

[0085] Embodiment 1

[0086] 1. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material provided in this embodiment includes the following steps:

[0087] 2. Conduct a natural aging test on a two-component epoxy polyurethane varnish sample in the site area. The color plate size is 150mm x 100mm x 1mm, and the paint film is prepared according to the "General Preparation Method of Paint Films"

[0088] (GB1727--2021), and there are no defects on the sample surface. Set the low-frequency impedance lg(|Z|0.01Hz) on the 7th day as the corrosion degree evaluation index.

[0089] 3. After the polyurethane sample has been naturally aged for 1 year, sampling tests are carried out for the corresponding roughness index, glossiness, adhesion and contact angle (shown in Table 1) in different natural environments.

[0090] Table 1 Performance data table of different sites

[0091]

[0092] 4. At the same time, through real-time monitoring, obtain the data of temperature, humidity, air pressure, wind speed, cloud cover and sunshine hours recorded every 1h in the test area, etc., and record the sunshine hours and dew point temperature every day.

[0093] 5. Use Python software to statistically analyze the data of temperature, humidity, air pressure, wind speed, cloud cover, sunshine hours and dew point temperature, etc., obtained, and obtain the apparent properties of the material (glossiness, adhesion, roughness and contact angle) in the test area and the corresponding environmental factors (annual average temperature, annual highest temperature, annual lowest temperature, annual average air pressure, annual average dew point temperature, annual average wind speed, annual average cloud cover, total rainfall, sunshine hours, annual average relative humidity). A mapping relationship is constructed between each apparent property and the corresponding environmental factor to complete the characterization process of environmental big data.

[0094] 7. Use the Spearman and principal component analysis algorithms to reduce the dimension and denoise the characteristic parameter data, and select the key environmental factors; group the environmental factors in different regions and the apparent property changes of the material, and the change data sets with different apparent properties are independent of each other. These grouped data will be used to construct the aging degree prediction model.

[0095] 8. Perform machine learning on the training set through Python software, and the algorithm used is the self-supervised learning algorithm, so as to construct a prediction model. The ratio of the verification set is 8:1, and other parameters adopt the system default parameters.

[0096] 9. Take the environmental factors obtained after dimension reduction as input parameters, and the material property changes as output parameters. Use Python software to construct a machine learning algorithm to train the environmental big data aging prediction model, and form an aging degree prediction model. As Figure 1 shown, compare the true value and predicted value of the adhesion force at each site, and calculate the square value R2 of the correlation coefficient between it and the experimental value at the same time. If it is greater than 85%, the prediction result is accurate.

[0097] 10. Conduct indoor acceleration experiments on the samples with different degrees to obtain the apparent properties and corrosion evaluation properties.

[0098] 11. Group the data set based on the apparent properties of the material and the corresponding corrosion building evaluation property changes, and the ratio of the training set to the verification set is 4:1.

[0099] 12. Take the apparent properties as input parameters, and the corrosion material property changes as output parameters. Use Python software to construct a machine learning algorithm to train the environmental big data aging prediction model, and form an aging degree prediction model. As Figure 1 shown, compare the true value and predicted value of the adhesion force at each site, and calculate the square value R2 of the correlation coefficient between it and the experimental value at the same time. R2 is 82.78%, which is greater than 80%, and the prediction result is accurate.

Claims

1. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material, characterized in that, First, use environmental factor data to predict the change in the apparent properties of the material, and then further predict the corrosion degree of the material based on these apparent properties. Through this method, the corrosion behavior of the coating material under different environmental conditions can be evaluated more accurately, the prediction accuracy of the coating life and protection performance can be improved, and thus a scientific basis can be provided for the selection and application of materials.

2. The multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 1, characterized in that: The prediction method is divided into two stages. In the first stage, by analyzing environmental factors in different regions, including temperature, humidity, air pressure, and solar sunshine hours, a relationship model between environmental data and the change in the apparent properties of the coating is established. The input of this relationship model is environmental data, and the output is the change in the apparent properties of the coating, including glossiness, roughness, adhesion, and contact angle, so as to predict the impact of environmental conditions on the coating performance. Through this relationship model, the impact of environmental factors on the apparent properties of the coating can be effectively quantified, laying a foundation for subsequent corrosion prediction. In the second stage, based on the change data of the apparent properties of the coating material obtained in the first stage, a prediction model between the apparent properties and the corrosion degree is further established. This prediction model uses the apparent property parameters as the input and outputs the corrosion degree of the coating. Through this prediction model, the influence law of the apparent properties on the corrosion degree of the coating can be evaluated more accurately, and then the corrosion resistance of the coating under different environments can be deduced, thus providing support for the design and optimization of protective materials.

3. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 2, characterized in that The first stage includes the following steps: 3.1 Select a coating material, conduct outdoor exposure experiments in different regions, and obtain the apparent property parameters and corrosion degree parameters of the coating material during the experimental period. 3.2 Collect the corresponding environmental data in step (1), including temperature, humidity, sunshine hours, air pressure, wind speed, cloud cover, and rainfall. 3.3 Extract the characteristic data in the environmental data obtained in step (2) as characteristic parameters, and use the Spearman and principal component analysis algorithms to perform dimensionality reduction and noise reduction processing on the characteristic parameter data, where the characteristic data is the statistical value of different environmental data. 3.4 Group the data from different regions into a test set and a validation set for the construction of the coating apparent property prediction model. 3.5 Use one or several of the recurrent neural network, self-supervised learning, gradient boosting tree, K-nearest neighbor algorithm, and random forest algorithm to construct a coating apparent property prediction model with the environmental characteristics obtained after dimensionality reduction as the input parameters and the coating apparent properties of the training set as the output parameters. 3.6 Take the characteristic environmental factors of the test set as input parameters, predict the changes in material properties in different regions, obtain the apparent properties of the service coatings, and at the same time, calculate the square value R of the correlation coefficient between the predicted results and the experimental values. 2 When R 2 ≥ 85%, the prediction results are credible.

4. According to a multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material as described in claim 2, the second stage includes the following steps: 4.1 Conduct a corrosion aging test on the coating material, and obtain the apparent property parameters and corrosion degree parameters of the coating material during the aging process. 4.2 Collect the corresponding apparent property data in step 4.1, including glossiness, adhesion, roughness, contact angle, corrosion degree parameters, and impedance values within 1 - 7 days of immersion. 4.3 Extract the apparent property data obtained in step 4.2, and preprocess the corrosion degree data and the apparent property data. The preprocessing methods for the corrosion evaluation data and the apparent property data include one or several of normalization, standardization, and taking logarithms. 4.4 Group the apparent properties - corrosion degrees of different samples, use part of the data as the training set for constructing the coating corrosion degree prediction model, and use the remaining data as the test set for validating the coating corrosion degree prediction model; 4.5 Use the apparent properties of the training set as input parameters, use the coating corrosion degree of the training set as output parameters, and construct an apparent property - corrosion degree prediction model using one or several of algorithms such as support vector machine, AdaBoost, K-nearest neighbor algorithm, and random forest; 4.6 Take the apparent performance of the test set as an input parameter to predict its material corrosion degree. At the same time, calculate the square value R of the correlation coefficient between the predicted result and the experimental value. 2 , when R 2 ≥80%, the prediction result is credible; 4.7 Use the prediction result of the environmental parameter - apparent property model as the input parameter, and use the corrosion degree model in 4.5 to predict the coating performance, forming an overall coating corrosion degree prediction model of "environmental factors - surface properties - corrosion degree".

5. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 1, characterized in that: The coating material is a composite material of one or more of polyurethane, epoxy resin, and acrylic resin.

6. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 3, characterized in that: The apparent property parameters of the coating material described in step 3.1 include optical properties, mechanical properties, and surface properties.

7. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 3, characterized in that: The apparent performance parameters of the coating material described in Step 3.1 are roughness, gloss, adhesion, and contact angle, and the corrosion performance parameters include the low-frequency impedance modulus (|Z| 0.01Hz ) within 7 days of coating immersion and.

8. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 3, characterized in that: In steps 3.1 to 3.2, the experimental period is 1 to 5 years, that is, the performance of the coating material is recorded 1 to 5 years after the samples are put in, and the environmental data is recorded every 1 to 24 hours.

9. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 3, characterized in that: The process of obtaining the statistical values of different environmental data in step 3.6 includes: using Python software to statistically analyze the environmental data of temperature, humidity, and sunshine hours, and obtaining the data of the performance change of the coating material during the experimental period and the statistical values under the corresponding temperature, humidity, and sunshine hours.

10. A multi-stage collaborative machine learning prediction method for the corrosion degree of a coating material according to claim 4, characterized in that: The corrosion aging experiment described in step 4.1 is an indoor accelerated experiment.

Images