Investigation of Service Life Prediction Models for Metallic Organic Coatings Using Full-Range Frequency EIS Data
Various service life prediction models of organic coatings were analyzed based on the acquirement of the measurement of Electrochemical Impedance Spectroscopy (EIS) from indoor accelerated tests. First, some theoretical formulas on corrosion lifetime predictions of coatings were introduced, followed by the comparative assessment of four practical prediction models in view of prediction accuracy in application. The prediction from impedance data at single low frequency |Z| 0.1 Hz , the classical
... degradation kinetics, and proposed improved degradation kinetics model, as well as a self-organized neural network prediction based on sample detection, were focused in this paper. The standard AF1410 plates employed as the metallic substrates were coated with sprayed zinc layer, epoxy-ester primer and polyurethane enamel layer. The accelerated experiments which mimicked coastal areas of China were carried out with the specimens after surface treatment. The assessment of results showed that the proposed improved degradation kinetics model and neural network classification model based on the full range of frequency data obviously have higher prediction accuracies than the traditional degradation kinetics model, and the prediction precision of the sample detection-based neural network classification was the highest among these models. The study gives some insights for coating degradation lifetime prediction which may be useful and supportive for practical applications. Spectroscopy (EIS), are widely used to study the damage behavior and corrosion protection capability of the coating systems and a comprehensive review of EIS can be seen in the literature written by G. Bierwagen  . EIS is suitable for the research of polymer-coated metals, which obtains a complete view of the process of the paint coating degradation and precise quantitative data regarding the behavior of organic coatings. Furthermore, it is a fast and effective method for classifying the paint coatings. The impedance measurements need to be carried out in a wide frequency range and the impedance data of the coatings usually have to be analyzed with equivalent circuits  . However, sometimes it is difficult to get satisfactory models to simulate the degradation behaviors for complicated coating systems. Also, during the measurements in low frequency range, signal drift and nature of data scattering always exists [9, 10] . What is more, there are a few drawbacks in the application of EIS to study the coating corrosion. First, a complete transfer function from available studies of reaction mechanisms cannot be acquired due to the overlapping of time constants. The decision about the selection of appropriate equivalent circuits can only be made with a knowledge of the processes and with the help of other techniques. The second drawback is the realistic limitation on the poor reproducibility of the test data. A variation in magnitude of up to three orders between replicate measurements is reported as a result of the heterogeneity of the coating  . Compared with equivalent circuit models, more studies for coating systems are needed to reveal the relationships among parameters from impedance spectroscopy and performance as well as the failure process of coatings. Zuo et al.  proposed that phase angles in the middle and high frequency range can be used as quick measures to assess the coating performance. Mahdavian and Attar  also verified that phase angle at high frequencies has a very good agreement with impedance data. Mansfeld and Tsai  showed that the minimum phase angle and its frequency are dependent on the delaminated interface between the applied coating and metals. At present, there are many research results on the corrosion mechanism of coatings [15, 16] . These studies reveal that water vapor in hot and humid environments can penetrate into the interface between the coating and the substrate by adsorption and diffusion, resulting in bubbling and cracking or peeling of the coating. In the salt fog environment, Chloride ion induces the corrosion of metal matrix. The corrosion product causes the coating to deform, bulge, and eventually separate from the metal matrix. Under the action of ultraviolet light, the functional group of the coating molecule degrades, or the chain breaks, changing the structural composition of the coating, and finally the coating performance is degraded. In general, the coating degradation process is decomposed into three stages, which can be named as early, middle and late stages, respectively. In the early stage, coatings have good protective properties; while in the middle stage, coatings are permeated by corrosive medium and the protective performance of the coatings decreases; in the late stage, coatings completely fail and the substrate is corroded  . In recent years, researchers also use artificial intelligence classification methods to study the coating corrosion process [18, 19] . On the other hand, studies on the service life prediction models of protective organic coatings are few. It was reported by the Japanese scholar Yamamoto Takashi that, over 26,838 literatures on coatings found that there were 90 papers referring to coating life, but only 3 papers involving coating life prediction equations  . The research of life prediction shows that the mechanism of metal materials and coating damage in long service is very complex, because there are so many factors that affect the life of coatings, such as temperature, humidity, pH and corrosive media. To date, there is no convincing method to accurately describe the quantitative relationship between these influencing factors and protective properties of coatings. In the initial stage of material design, experimental means is still the most effective research method. It is of great significance to predict the material life using limited test data. For the prediction method, degradation dynamics model and neural network technology have been widely studied and applied in recent years. Cai et al.  studied the degradation process of polyamide epoxy varnish for aluminum alloy in UV and salt spray combined environmental test. Their research suggested that the electrochemical impedance modulus at low frequency is suitable Metals 2017, 7, 274 3 of 16 to construct degradation kinetics. On the basis of the degradation kinetic equation, Mark Evans  added exponential parameters to improve the prediction accuracy, and the extrapolated distribution obtained by this new approach was much closer to the distribution for the naturally weathered data. In 1999, C. C. Lee and F. Mansfield  applied Kohonen neural network to analyze experimental impedance spectra of steel coated with different paints used in naval construction, and they classified the corrosion process into three stages, namely, 'good', 'intermediate' and 'poor'. Zhao Xia et al.  analyzed the electrochemical impedance spectroscopy (EIS) of the wetting-drying cycle in the coating failure process by self-organizing feature mapping network (SOM), and the coating cycles were also divided into three categories. Wang et al.  applied the BP (Back Propagation) neural network to predict the atmospheric corrosion process of aluminum alloy, and the relationship between neural network training accuracy and prediction accuracy was studied. In addition, there is another interesting approach for ranking and evaluating organic paint coatings via calculating the area ratio surrounded by the data in the electrochemical impedance spectroscopy  . Based on the previous studies, an improved degradation kinetics model was proposed in this paper and the Kohonen neural network classification model was also established for organic coating sprayed on AF1410 high strength steel in UV, thermal shock, low temperature fatigue and salt spray combined environmental tests. The results of the morphological analysis were compared with the predicted results from these models to prove that both of the models are more accurate than the traditional degradation model. Materials and Methods Preparation of Specimens The AF1410 high strength steel was used as the test material, and dimensions of the test specimens are 110 mm × 30 mm × 3 mm, as showed in Figure 1 . The number of the specimens were three, which were labeled Plate 1#, Plate 2# and Plate 3#. These specimens were cut by CNC punching (AECC Beijing Institute of Aeronautical Materials, Beijing, China). The use of alternative processes to milling can shorten the time needed to produce the pieces and the specimens have good tensile strength and elongation hardness  .