Institute of Advanced Materials and Technology, University of Science and Technology Beijing, Beijing 100083, China
2)
AECC Key Laboratory of Advanced Corrosion and Protection of Aeronautical Materials, AECC Beijing Institute of Aeronautical Materials, Beijing 100095, China
3)
Shunde Innovation School, University of Science and Technology Beijing, Foshan 528399, China
Traditional 3Ni weathering steel cannot completely meet the requirements for offshore engineering development, resulting in the design of novel 3Ni steel with the addition of microalloy elements such as Mn or Nb for strength enhancement becoming a trend. The stress-assisted corrosion behavior of a novel designed high-strength 3Ni steel was investigated in the current study using the corrosion big data method. The information on the corrosion process was recorded using the galvanic corrosion current monitoring method. The gradient boosting decision tree (GBDT) machine learning method was used to mine the corrosion mechanism, and the importance of the structure factor was investigated. Field exposure tests were conducted to verify the calculated results using the GBDT method. Results indicated that the GBDT method can be effectively used to study the influence of structural factors on the corrosion process of 3Ni steel. Different mechanisms for the addition of Mn and Cu to the stress-assisted corrosion of 3Ni steel suggested that Mn and Cu have no obvious effect on the corrosion rate of non-stressed 3Ni steel during the early stage of corrosion. When the corrosion reached a stable state, the increase in Mn element content increased the corrosion rate of 3Ni steel, while Cu reduced this rate. In the presence of stress, the increase in Mn element content and Cu addition can inhibit the corrosion process. The corrosion law of outdoor-exposed 3Ni steel is consistent with the law based on corrosion big data technology, verifying the reliability of the big data evaluation method and data prediction model selection.
Material corrosion and degradation are affected by the complex coupling of numerous factors, such as material composition, preparation process, experimental conditions, and service environment, and the process is highly complex and nonlinear with time [1]. Therefore, the corrosion data comprising the information describing the corrosion degree and the corrosion influence parameters have high-dimensional characteristics, and the high-dimensional properties are related to each other; thus, describing the internal relationship through the traditional multi-parameter physical model is often difficult [2]. Considering this problem, some researchers have used artificial neural networks, support vector machines, random forests, and their improved optimization algorithms to focus on the analysis and mining of corrosion data for various materials under laboratory test conditions; this approach aims to establish a prediction model between material composition, experimental conditions, and corrosion failure degree and initially realize the intelligent screening and optimization of material composition design [3–6].
Machine learning methods have been used in many fields, such as the design of novel materials, evaluation of the corrosion modes, and prediction of the residual corrosion life. The research team at the University of California, Berkeley, established a prediction model of corrosion current with time and solution environment of carbon steel and low alloy materials based on the polarization curve test data through an artificial neural network algorithm [7]. They further predicted the crack growth rate of materials, such as stainless steel, based on this method and excavated that temperature and conductivity are the most important factors affecting the growth rate of corrosion cracks [8]. The research team of Monash University in Australia prepared 53 kinds of materials by adding different types and contents of rare metals to magnesium-based materials and used these data to study the change in corrosion rate and hardness with alloy composition and established an artificial neural network prediction model for the corrosion rate and hardness models [9]. The team from the University of Cádiz in Spain used support vector machine, artificial neural network, classification tree, and K nearest neighbor algorithms to conduct a series of studies on the pitting corrosion prediction of stainless steel, predicted the pitting corrosion behavior of stainless steel under different laboratory conditions, and optimized the support vector machine model based on different feature selection methods and kernel functions to classify the pitting corrosion behavior of stainless steel [10]. Furthermore, Koo et al. [11] proposed a cascade support vector machine model for predicting the effect of residual stress on the stress corrosion of nuclear materials.
Feature fusion and dimension reduction are crucial data preprocessing methods. Various feature fusion and dimension reduction methods have been used on the basis of gray correlation analysis, principal component analysis, Pearson correlation, Spearman correlation, factor analysis, maximum information coefficient, genetic algorithm, local linear embedding, Laplace mapping, random forest, and other linear, local linear, or nonlinear methods [12]. Dimensionality reduction methods for high-dimensional complex data affecting environmental failure of materials have also been employed by considering the statistical characteristics of data, trend changes, spatial structure, causal logic and other aspects, material composition, and research on correlation and causal screening [13–16]. Combined with the simulation of material environmental failure prediction, the dimensionality reduction capability, accuracy, and knowledge mining capability of the dimensionality reduction method of correlation and causality key parameter screening can be compared and evaluated. The key parameter feature fusion method of material environmental failure based on correlation and causality analysis can be analyzed in accordance with the principle of “maximum correlation and minimum redundancy.” The above research proved that data dimensionality reduction and key feature parameter extraction play an important role in improving the accuracy of the material environmental failure evaluation prediction model.
Stress is quite an important factor that affects the corrosion process and modes of steels [17–18]. Stress-induced corrosion or stress corrosion cracking often occurs due to the exposure of components to a wide variety of internal and external stresses when in service. The strength of traditional 3Ni weathering steel cannot fully meet the requirement of offshore engineering development, and the design of novel 3Ni steel for strength enhancement with the addition of microalloy elements, such as Mn or Nb, has become a trend [19]. However, the high strength of 3Ni steel introduces high risks for stress-assisted corrosion or stress corrosion cracking [20]. Therefore, studying the mechanism of the corrosion behavior for high-strength 3Ni steel and proposing ways for its improvement are necessary. Moreover, the addition of Cu is helpful in improving the corrosion resistance of 3Ni steel. The formation of CuO in the rust layer can inhibit the anodic reaction process and promote uniform corrosion by preventing the initiation of localized corrosion. The existence state of Cu in 3Ni steel will affect the corrosion resistance. The corrosion rate of 3Ni steel with the Cu precipitation state is lower than that of 3Ni steel with the Cu solution state. The Cu-rich precipitated phase is helpful to increase the CuO proportion in the rust layer of 3Ni steel at the early corrosion stage and promote the enrichment of Cu at the matrix interface.
In this work, a machine learning method was used to extract the stress assisted corrosion behavior of a novel designed high-strength 3Ni steels, and the applied stress was considered as a variable. First, the information on the corrosion process was recorded using the galvanic corrosion current monitoring method. Second, the gradient boosting decision tree (GBDT) method was used to mine the corrosion mechanism, and the importance of the structure factor was studied. Field exposure tests were then performed to verify the results calculated by the GBDT method.
2.
Experimental
2.1
Materials and specimens
Three kinds of high-strength 3Ni steel were designed in the current study with the addition of Mn and Cu, and their chemical compositions are shown in Table 1. The steels were smelted in a vacuum furnace and rolled by a thermos mechanical control process. Schematics of the tuning fork specimen are depicted in Fig. 1. Fig. 1(a) presents the dimensions of the tuning fork, and the 10 mm × 10 mm surface was used as the working area. Fig. 1(b) presents the dimensions of the cathode graphite. The loaded and unloaded tuning forks were mounted together with the cathode graphite to form a corrosion sensor, as shown in Fig. 1(c). The deflection of the loaded fork was controlled to 1.5 mm by turning the screws to generate stress for investigation. The mechanical properties of the steel are listed in Table 2.
Table
1.
Chemical compositions of the tested steels wt%
Deform 3D V6.0 software was used to simulate the generated stress of the tuning fork. A tetrahedral meshing method with an element value of 12000 was used to generate the mesh. The materials were referred to the system library steel AISI 316L SS. The number of simulation steps was 20, 40, 60, 80, and 100. The constant die displacement was set to 0.03 mm. The strain and stress distribution of the tuning fork were recorded.
2.3
Data acquisition and evaluation
The loaded and unloaded corrosion sensors were tested in the dry and wet alternating corrosion test chamber, and the galvanic corrosion current between the working electrode and the cathode graphene was recorded by a corrosion current monitoring system (PCM-100W, TYD). The sampling frequency was 0.0167 Hz (1 data point per minute). The galvanic corrosion current was continuously acquired and evaluated by the corrosion big data clock and the cumulative electric quantity method. The corrosion big data clock diagram is a method of recording the collected corrosion current data according to the clock rotation mode, and the schematic is shown in Fig. 2. The corrosion big data clock diagram comprises a series of concentric rings, the number of rings represents the test days, and each ring records the 1440 galvanic corrosion data collected by the sensor in one day. Simultaneously, the color change of the corrosion big data clock chart corresponds to the corrosion current intensity, which can intuitively reflect the corrosion degree of the entire service cycle of material degradation. The cumulative electric quantity method is expressed as the integration of the relative corrosion current intensity value over time. A high amount of cumulative electric quantity leads to considerably serious corrosion. Thus, the difference in corrosion resistance of different steels is evaluated. The total integral of the relative corrosion current intensity value over time is expressed as follows:
Fig.
2.
Schematic representation of the corrosion big data clock evaluation method.
where Qi is the cumulative electric quantity of the corrosion sensor, and the unit is Coulomb (C); in is the relative corrosion current intensity at n (n = 1, 2, ..., T); ∆t is the time interval of corrosion current intensity acquisition. The Qi value is superimposed and calculated when the relative corrosion current intensity is collected, and the calculated result is plotted as a cumulative corrosion curve.
2.4
Machining learning method
A total of 181372 data collected by the sensor are used as the sample set of machine learning, and each data sample includes one time-factor, two environmental factors, and eight material factors. Among these, the environmental factors are temperature and humidity, and the material factors include Mn content, Cu content, original austenite grain size (PAGS), lath width (Lath), small-angle grain boundary ratio (LAGB), ratio of Σ3 grain boundaries (Σ3), residual austenite ratio (RA), and core mean misalignment (KAM). The 11 aforementioned variable data are used as input, the current obtained by the sensor measurement is the output, and the data training set in stress and stress-free states is shown in Table 3, where each row of data has a unique data encoding (ID).
Table
3.
Input feature values for machine learning
The gray correlation analysis method was used for feature fusion and dimension reduction. The GBDT model was utilized to explore the correlation between the corrosion rate of 3Ni steel and environmental and material factors. In the GBDT model, time, environment, and material factors are used as input. A typical clustering and regression tree (CART) is trained by the residual of the previous tree in each iteration through a series of weak prediction models, mainly for the CART iteration tree algorithm to learn data samples, and the regression results of all trees are finally accumulated as output. Through continuous optimization and adjustment of the algorithm, the final number of decision trees, the maximum depth, and the number of bins are all 20 in the current study. The idea of the GBDT algorithm is shown in Fig. 3.
Fig.
3.
Schematic of the gradient boosting decision tree algorithm.
Three parallel specimens with dimensions of 100 mm × 50 mm × 4 mm were prepared for field exposure test. Each piece of steel was weighed, and the length, width, and height dimensions were measured. The exposure test site was the Wenchang atmospheric station (China), and the exposure test was divided into the following three cycles: half a year, one year, and two years. The Wenchang atmospheric station has the characteristics of tropical and subtropical climates, the perennial temperature is maintained between 23.4–24.4°C, the annual average relative humidity is 87%, the annual average sunshine is 1953.8 h, and the annual average rainfall is 1721 mm. The annual average chloride ion deposition rate is 72 mg/(m2⋅d) and is characterized by high temperature, high humidity, and high irradiation.
3.
Results and discussion
3.1
Stress and strain distribution
Fig. 4 shows the simulation of the stress–strain distribution of the tuning fork figure. The stress on the working surface is concentrated, and the stress value at the center position is the largest. The center stresses for 0.82Mn, 1.36Mn, and 1.36MnCu steels are 978, 1090, and 940 MPa, respectively, which exceed the yield stress of the material. The strains at the top of the three tuning fork specimens are the largest, and the value are 8.1%, 7.9%, and 8.2%, respectively.
Fig.
4.
Stress–strain distribution of tuning fork specimens: the stress distributions of (a1) 0.82Mn, (a2) 1.36Mn, and (a3) 1.36MnCu steels; the strain distributions of (b1) 0.82Mn, (b2) 1.36Mn, and (b3) 1.36MnCu steels.
Fig. 5 presents the instantaneous current clock diagram of the loaded and unloaded corrosion sensors in the dry and wet alternating corrosion test chamber. Among them, blue indicates that the relative corrosion current intensity value and the corrosion rate are both low, while red indicates that the relative corrosion current intensity value and the corrosion rate are both high. The instantaneous current clock chart color of unloaded and loaded steel is radial, which is mainly due to the low and high corrosion rates during the drying and wetting states, respectively. The corrosion rate of 3Ni steel showed dynamic changes throughout the test. However, differences were observed in the corrosion process of unloaded and loaded 3Ni steel. Fig. 5(a)–(c) shows that under stress-free conditions, the corrosion current of the three 3Ni steels demonstrates a trend of initial increase and then decrease, which reflects the evolution of the corrosion resistance of the rust layer. The corrosion rate of 3Ni steel is high during the early stage of corrosion. As the corrosion process progresses, the protective rust layer gradually forms and reaches stability, which increases the corrosion resistance of 3Ni steel, thereby reducing its corrosion rate. Additionally, Mn and Cu did not have a significant effect on the time required for the stabilization of the rust layer. In addition, the corrosion clock diagram of 0.82Mn steel is dark blue overall, followed by 1.36MnCu and 1.36Mn steels is blue-green, indicating that the average corrosion rate order of the three steel is: 0.82Mn < 1.36MnCu < 1.36Mn.
Fig.
5.
(a–c) Corrosion clock diagram and (d) accumulated corrosion quantity for the unloaded specimens: (a) 0.82Mn, (b) 1.36Mn, and (c) 1.36MnCu.
The color change of the corrosion clock graph for the loaded specimen is quite different from the unloaded specimen. According to different colors, the corrosion current of the three kinds of 3Ni steel shows a trend of first increasing and then decreasing and then continuing to increase. Among these, 0.82Mn and 1.36Mn steels are more evident than 1.36MnCu steel. This finding shows that the rust layer of steel cannot effectively reduce the corrosion rate of stressed 3Ni steel despite reaching its stability. In addition, the color comparison of the clock graph revealed that the average corrosion current value order is 0.86Mn > 1.36Mn > 1.36 MnCu.
The cumulative electric quantity can still intuitively reflect that the corrosion of 3Ni steel during the test cycle is a dynamic process. As shown in Figs. 5(d), under stress-free conditions, different 3Ni steels are divided into two stages in the pre-corrosion and post-corrosion stages during the test of 32000 min according to the slope change of the cumulative corrosion curve. In the early stage of corrosion, no consistent law of the corrosion rate in the comparison of three 3Ni steels was observed. When the corrosion process and rust layer stabilized in the late corrosion stage, the corrosion rate of the three 3Ni steels remained as follows: 0.82Mn < 1.36MnCu < 1.36Mn. In Fig. 6(d), according to the slope of the curve, the corrosion process is also divided into two stages: pre-corrosion and post-corrosion. In the early stage of corrosion, the corrosion rate order of tche three kinds 3Ni steel is 1.36Mn > 0.82Mn > 1.36MnCu. In the later stage of corrosion, the corrosion rate of the three kinds of 3Ni steel increased, but 0.82Mn steel increased the fastest. After the corrosion stage was stabilized, the corrosion rate of 0.82Mn steel was the largest, followed by 1.36Mn steel, and the corrosion rate of 1.36MnCu steel always remained the lowest. This finding shows that the addition of Mn and Cu can reduce the corrosion rate of 3Ni steel in the presence of stress.
Fig.
6.
(a–c) Corrosion clock diagram and (d) accumulated corrosion quantity for the loaded specimens: (a) 0.82Mn, (b) 1.36Mn, and (c) 1.36MnCu.
The GBDT model was used to select a part of the values of the galvanic corrosion current measured by the test steel of the loaded and unloaded specimens randomly, and the results are shown in Fig. 7. The figure reveals that the predicted and true values with or without stress coincide, which shows the reliability of the GBDT model. The GBDT model was used to fit the pre-corrosion and post-corrosion currents of loaded and unloaded steels, as shown in Fig. 8. The function y = x is used to characterize the fitting effect, that is, the fitting result is satisfactory when the true value is close to the predicted value. The results of the predicted values corresponding to the actual current values are distributed around the function; among which, the fitted values in the stress-free state are close to linearity, the fitted variances are 0.922 and 0.962, and the scatter distribution dispersion is relatively large in the stressful state. However, the values of the fitted variances are ideal, namely 0.893 and 0.929.
Fig.
7.
Training and prediction values of the corrosion current (I) based on the GBDT mode: (a) unloaded and (b) loaded 3Ni steels.
Fig.
8.
Corrosion current fitting based on the GBDT model: (a) unloaded 3Ni steel at the initial corrosion stage, (b) loaded 3Ni steel at the initial corrosion stage, (c) unloaded 3Ni steel at the later corrosion stage, and (d) loaded 3Ni steel at the later corrosion stage.
Fig. 9 shows the weight analysis of the influence of environmental, composition, and microstructure factors on the corrosion rate of 3Ni steel under a stress-free state calculated using the GBDT model. Fig. 9(a) reveals that the influence weight of temperature, relative humidity, and time is substantially larger than the composition and structure parameters in the early corrosion stage of unloaded 3Ni steel. The order of influence is as follows: temperature > relative humidity > exposure time. However, temperature is positively correlated with corrosion current, while relative humidity and exposure time are negatively correlated with corrosion current (Fig. 9(c)). This finding indicates that the corrosion rate in the early corrosion stage of 3Ni steel is controlled by the temperature, relative humidity, and time of the environment. The increase in temperature can promote the corrosion process, but that in relative humidity and exposure time reduces the corrosion rate. However, in the later corrosion stage of 3Ni steel, the influence of environmental parameters weakens, and that of microstructure and composition factors on the corrosion current value increases, as shown in Fig. 9(b). The weight of the influence includes ΚΑΜ, PAGS, ΡΑ, Mn content, Lath, LAGB, Σ3 grain boundary ratio, temperature, time, relative humidity, and Cu content. The correlation between the above factors and the corrosion current intensity is different. Fig. 9(d) shows a high promoting effect of Mn content on corrosion rate in the later stage of corrosion, and the addition of Cu element can inhibit corrosion. Besides, during the later stage, the predominant position of temperature and humidity decreases while the importance of structure factors increases. It is because that the formation of the rust layer can further protect the corrosion of steels from temperature and humidity to some degree.
Fig.
9.
(a, b) Variable importance factors and (c, d) weight coefficients of the unloaded specimen based on the GBDT model.
Fig. 10 shows the weight analysis of the influence of environmental, composition, and microstructure factors on the corrosion rate of the loaded 3Ni steel calculated by using the GBDT model. Fig. 10(a) and (b) reveals that the presence of stress assisted the influence of composition and microstructure factors on the corrosion rate of 3Ni steel compared with the no-stress specimen. In the early corrosion stage (Fig. 10(c)),in addition to the relative humidity and temperature of the environment, the KMA and LAGB were positively correlated with the corrosion current, while Mn and Cu contents were negatively correlated with the corrosion current, indicating that the addition of two elements could inhibit the corrosion process of 3Ni steel in the presence of stress. The influence weight of various factors on the corrosion rate of 3Ni steel changed with the evolution of corrosion time. In the later corrosion period (Fig. 10(d)), the LAGB, PAGS, and RA of 3Ni steel were positively correlated with the corrosion current, while the ratio of Σ3 grain boundaries, Mn content and Cu content could suppress the corrosion current in the presence of stress. This suppression can be attributed to the reduction in the residual austenite proportion and increase of the ratio of Σ3 grain boundaries due to the additional Mn and Cu, contributing to the improved stress corrosion cracking (SCC) sensitivity of 3Ni steel.
Fig.
10.
(a, b) Importance factors of the corrosion process of stressed 3Ni steel based on the GBDT model and (c, d) their influencing weights.
The analysis of corrosion big data revealed that the increase in Mn content raises the uniform corrosion rate of 3Ni steel but possibly reduces its SCC sensitivity, while the increase in Cu content can improve the corrosion resistance of 3Ni steel. The results of the outdoor exposure tests were verified. Table 4 shows the corrosion rates of three different 3Ni steels exposed to the Wenchang atmospheric corrosion site for 6, 12, and 24 months, and the corrosion trend is shown in Fig. 11. The figure reveals that the corrosion rate of the test steel decreases with the evolution of exposure time, and the increase of Mn content raises the corrosion rate of 3Ni steel. Meanwhile, the addition of Cu content is conducive to reducing the corrosion degree of 3Ni steel. Fig. 12 presents the cross-sectional morphology of the rust layer and the EDS mapping result of the specimens after exposure for 24 months. The average thickness of the rust layer for the three steels was 232.6, 206.5, and 138.3 μm for 0.82Mn, 1.36Mn, and 1.36MnCu steel, respectively. The EDS mapping of the rust layer depicts that Fe, Mn, Ni, O, and Cu were evenly distributed, while a certain aggregation of chloride ions was observed in the inner and outer layers. However, only a small amount of chloride ions existed in the inner layer for 1.36 MnCu steel, which explains that the Cu element can inhibit the penetration of chloride ions into the inner rust layer to a certain extent.
Table
4.
Corrosion rates of the three test steels exposed in the Wenchang atmospheric site (China) for different months mm·a−1
Fig.
12.
Cross-section morphology of three specimens after exposure for 24 months and the corresponding EDS mapping: (a) 0.82Mn, (b) 1.36Mn, and (c) 1.36MnCu.
The morphology of the U-shaped arc top observed after rust removal is shown in Fig. 13 to compare the differences in the stress-assisted corrosion resistance of the three steels. The addition of appropriate Mn and Cu can inhibit crack initiation and propagation based on the length and the number of microcracks. Quite a few cracks were found in 1.36Mn steel, and almost no microcracks can be detected in the 1.36MnCu steel, which implies that the increase in appropriate Mn and Cu contents in 3Ni steel helps improve its SCC resistance. Additionally, corrosion pits can be detected in the arc top, which indicates that the mechanism of the SCC may be controlled by anodic dissolution.
Fig.
13.
Corrosion morphologies of U-shaped bent specimens exposed for one year after rust removal: (a) 0.82Mn, (b) 1.36Mn, and (c) 1.36MnCu.
Overall, the effects of Mn and Cu elements on the uniform and stress-assisted corrosion of 3Ni steel proven by outdoor exposure tests are consistent with the calculation results of the GBDT model. The results depict that Mn is not conducive to the uniform corrosion of 3Ni steel but is beneficial to improve its SCC resistance. The appropriate Cu content in 3Ni steel can improve the stress-assisted corrosion resistance of 3Ni steel.
4.
Conclusions
Self-developed stress-free and stress-based corrosion sensors were used in the current study to realize the corrosion big data collection and analyze the dynamic mechanism of the corrosion process in the marine atmospheric environment of the South China Sea by using big data evaluation methods and machine learning, combined with the verification of field exposure tests. The following results are presented.
(1) The corrosion current of unstressed 3Ni steel first increases and then decreases with time, while the corrosion current of stressed 3Ni steel shows a trend of first increasing and then decreasing and then continuing to increase. Mn and Cu have minimal effect on the time required for the stability of the unstressed and stressed 3Ni steel rust layer, and the formation of the rust layer can effectively inhibit the corrosion process of stressless 3Ni steel. However, this rust layer cannot reduce the corrosion rate of stressed 3Ni steel after a long cycle.
(2) Mn and Cu have no effect on the corrosion rate of unstressed 3Ni steel in the early stage of corrosion. When the corrosion reaches a stable state, the increase in the Mn element content can raise the corrosion rate of 3Ni steel, while Cu reduces the corrosion rate of 3Ni steel.
(3) Mn has little effect on the corrosion rate of stressed 3Ni steel in the early corrosion stage but will inhibit the later corrosion process, and Cu addition can reduce the corrosion rate of the entire corrosion stage of 3Ni steel.
(4) The influence weight of temperature, relative humidity, and time in the early corrosion stage of stress-free 3Ni steel is substantially larger than that of composition and microstructure parameters, and the influence weight of structural and component factors on the corrosion current value in the later stage of corrosion increases. The presence of stress enhanced the influence of composition and microstructure factors on the entire corrosion process of 3Ni steel, in which Mn and Cu were negatively correlated with the corrosion current, and the increase in the Mn element content and Cu addition could inhibit the corrosion process in the presence of stress.
(5) The corrosion law of outdoor exposure 3Ni steel is consistent with the law based on corrosion big data technology, which verifies the reliability of the big data evaluation method and data prediction model selection.
Acknowledgements
This work was financially supported by the National Natural Science Foundation of China (No. 52203376) and the National Key Research and Development Program of China (No. 2023YFB3813200).
Conflict of Interest
Xiaogang Li is an editorial board member for this journal and was not involved in the editorial review or the decision to publish this article. All authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
X.G. Li, D.W. Zhang, Z.Y. Liu, Z. Li, C.W. Du, and C.F. Dong, Materials science: Share corrosion data, Nature, 527(2015), No. 7579, p. 441. DOI: 10.1038/527441a
[2]
X.J. Yang, Y. Yang, M.H. Sun, et al., A new understanding of the effect of Cr on the corrosion resistance evolution of weathering steel based on big data technology, J. Mater. Sci. Technol., 104(2022), p. 67. DOI: 10.1016/j.jmst.2021.05.086
[3]
Z.B. Pei, X.Q. Cheng, X.J. Yang, et al., Understanding environmental impacts on initial atmospheric corrosion based on corrosion monitoring sensors, J. Mater. Sci. Technol., 64(2021), p. 214. DOI: 10.1016/j.jmst.2020.01.023
[4]
X.J. Yang, J.K. Yang, Y. Yang, et al., Data-mining and atmospheric corrosion resistance evaluation of Sn- and Sb-additional low alloy steel based on big data technology, Int. J. Miner. Metall. Mater., 29(2022), No. 4, p. 825. DOI: 10.1007/s12613-022-2457-9
[5]
Y.J. Zhi, Z.H. Jin, L. Lu, et al., Improving atmospheric corrosion prediction through key environmental factor identification by random forest-based model, Corros. Sci., 178(2021), art. No. 109084. DOI: 10.1016/j.corsci.2020.109084
[6]
Z.B. Pei, D.W. Zhang, Y.J. Zhi, et al., Towards understanding and prediction of atmospheric corrosion of an Fe/Cu corrosion sensor via machine learning, Corros. Sci., 170(2020), art. No. 108697. DOI: 10.1016/j.corsci.2020.108697
[7]
D.D. Macdonald, Y.K. Zhu, J. Yang, et al., Corrosion of rebar in concrete. Part IV. On the theoretical basis of the chloride threshold, Corros. Sci., 185(2021), art. No. 109460. DOI: 10.1016/j.corsci.2021.109460
[8]
Y.K. Zhu, D.D. Macdonald, J. Qiu, and M. Urquidi-Macdonald, Corrosion of rebar in concrete. Part III: Artificial Neural Network analysis of chloride threshold data, Corros. Sci., 185(2021), art. No. 109438. DOI: 10.1016/j.corsci.2021.109438
[9]
D.R. Feenstra, A. Molotnikov, and N. Birbilis, Utilisation of artificial neural networks to rationalise processing windows in directed energy deposition applications, Mater. Des., 198(2021), art. No. 109342. DOI: 10.1016/j.matdes.2020.109342
[10]
M.J. Jiménez-Come, E. Muñoz, R. García, et al., Pitting corrosion behaviour of austenitic stainless steel using artificial intelligence techniques, J. Appl. Log., 10(2012), No. 4, p. 291. DOI: 10.1016/j.jal.2012.07.005
[11]
B. Koo, S. La, N.W. Cho, and Y. Yu, Using support vector machines to classify building elements for checking the semantic integrity of building information models, Autom. Constr., 98(2019), p. 183. DOI: 10.1016/j.autcon.2018.11.015
[12]
H.Y. Wu, H.G. Lei, and Y.F. Chen, Grey relational analysis of static tensile properties of structural steel subjected to urban industrial atmospheric corrosion and accelerated corrosion, Constr. Build. Mater., 315(2022), art. No. 125706. DOI: 10.1016/j.conbuildmat.2021.125706
[13]
J.M. Yao, W. Liang, and J.Y. Xiong, Novel intelligent diagnosis method of oil and gas pipeline defects with transfer deep learning and feature fusion, Int. J. Press. Vessels Pip., 200(2022), art. No. 104781. DOI: 10.1016/j.ijpvp.2022.104781
[14]
H.D. Fu, H.T. Zhang, C.S. Wang, W. Yong, and J.X. Xie, Recent progress in the machine learning-assisted rational design of alloys, Int. J. Miner. Metall. Mater., 29(2022), No. 4, p. 635. DOI: 10.1007/s12613-022-2458-8
[15]
H.T. Zhang, H.D. Fu, Y.H. Shen, and J.X. Xie, Rapid design of secondary deformation-aging parameters for ultra-low Co content Cu–Ni–Co–Si–X alloy via Bayesian optimization machine learning, Int. J. Miner. Metall. Mater., 29(2022), No. 6, p. 1197. DOI: 10.1007/s12613-022-2479-3
[16]
G.F. Pan, F.Y. Wang, C.L. Shang, et al., Advances in machine learning- and artificial intelligence-assisted material design of steels, Int. J. Miner. Metall. Mater., 30(2023), No. 6, p. 1003. DOI: 10.1007/s12613-022-2595-0
[17]
X.J. Yang, M.H. Liu, Z.Y. Liu, C.W. Du, and X.G. Li, Failure analysis of a 304 stainless steel heat exchanger in liquid sulfur recovery units, Eng. Fail. Anal., 116(2020), art. No. 104729. DOI: 10.1016/j.engfailanal.2020.104729
[18]
X.J. Yang, J.M. Shao, Z.Y. Liu, et al., Stress-assisted microbiologically influenced corrosion mechanism of 2205 duplex stainless steel caused by sulfate-reducing bacteria, Corros. Sci., 173(2020), art. No. 108746. DOI: 10.1016/j.corsci.2020.108746
[19]
J.H. Jia, X.Q. Cheng, X.J. Yang, X.G. Li, and W. Li, A study for corrosion behavior of a new-type weathering steel used in harsh marine environment, Constr. Build. Mater., 259(2020), art. No. 119760. DOI: 10.1016/j.conbuildmat.2020.119760
[20]
J.H. Jia, Z.Y. Liu, X.G. Li, C.W. Du, and W. Li, Comparative study on the stress corrosion cracking of a new Ni-advanced high strength steel prepared by TMCP, direct quenching, and quenching & tempering, Mater. Sci. Eng. A, 825(2021), art. No. 141854. DOI: 10.1016/j.msea.2021.141854
Yuanzheng Guo, Dongfeng He, Xiaolong Li, et al. End-point Temperature Prediction of Converter Steelmaking Based on Key Feature Amplification and Grey Wolf Algorithm Improved Affinity Propagation Clustering. Metallurgical and Materials Transactions B, 2025, 56(2): 2047.
DOI:10.1007/s11663-025-03472-4
2.
Xiaojia Yang, Qing Li, Sen Liu, et al. Atmospheric corrosion prediction of carbon steel and weathering steel based on big data technology. Corrosion Communications, 2025.
DOI:10.1016/j.corcom.2023.06.002
3.
Christian Mathew, Emmanuel Adu-Gyamfi. A review on AI-driven environmental-assisted stress corrosion cracking properties of conventional and advanced manufactured alloys. Corrosion Engineering, Science and Technology: The International Journal of Corrosion Processes and Corrosion Control, 2025, 60(2): 145.
DOI:10.1177/1478422X241276727
4.
Xiaojia Yang, Qing Li, Renzheng Zhu, et al. Study the outdoor corrosion performance of Sn- and Mo-additional low alloy steel based on big data technology. Journal of Materials Research and Technology, 2025, 35: 1950.
DOI:10.1016/j.jmrt.2025.01.105
5.
Yanxin Qiao, Tianyu Wang, Zhilin Chen, et al. Corrosion techniques and strategies for used fuel containers with copper corrosion barriers under deep geological disposal conditions: A review. International Journal of Minerals, Metallurgy and Materials, 2024, 31(12): 2582.
DOI:10.1007/s12613-024-2949-x
6.
Alexey G. Zinyagin, Alexander V. Muntin, Vadim S. Tynchenko, et al. Recurrent Neural Network (RNN)-Based Approach to Predict Mean Flow Stress in Industrial Rolling. Metals, 2024, 14(12): 1329.
DOI:10.3390/met14121329
7.
Xiangyu Xu, Lu Zhang, Zifei Wang, et al. Critical precipitation behavior of MnTe inclusions in resulfurized steels during solidification. International Journal of Minerals, Metallurgy and Materials, 2024, 31(8): 1849.
DOI:10.1007/s12613-023-2757-8
8.
Meiqi Chi. Optimization of Prediction Model Based on Gradient Boosting Decision Tree. 2024 International Conference on Integrated Intelligence and Communication Systems (ICIICS),
DOI:10.1109/ICIICS63763.2024.10860220
Search
DownLoad:
