Electrochemical and Mechanical Properties of Super-hydrophobic Aluminum Substrates Modified with Nano-silica and Fluorosilane

Surface & Coatings Technology 206 (2012) 3700–3713

Contents lists available at SciVerse ScienceDirect

Surface & Coatings Technology

j ourna l homepage: www.e lsev ie r .com/ locate /sur fcoat

Short Communication

Electrochemical and mechanical properties of superhydrophobic aluminumsubstrates modified with nano-silica and fluorosilane

Xianming Shia,b,c,⁎,1, Tuan Anh Nguyenb,1, Zhiyong Suod, Jianlin Wua, Jing Gonga, Recep Avcid

a School of Civil Engineering and Architecture, Wuhan Polytechnic University, Wuhan, 430023, Chinab Corrosion and Sustainable Infrastructure Laboratory, Western Transportation Institute, PO Box 174250, Montana State University, Bozeman, MT 59717-4250, USAc Civil Engineering Department, 205 Cobleigh Hall, Montana State University, Bozeman, MT 59717-2220, USAd Imaging and Chemical Analysis Laboratory, Department of Physics, Montana State University, Bozeman, MT 59717, USA

⁎ Corresponding author at: Corrosion and SustainaWestern Transportation Institute, PO Box 174250, MontMT 59717-4250, USA. Tel.: +1 406 994 6486; fax: +1

E-mail address: [email protected] (X. Sh1 Co-first author who made equal contribution to this

0257-8972/$ – see front matter. Published by Elsevier Bdoi:10.1016/j.surfcoat.2012.02.058

a b s t r a c t

Article history:Received 10 January 2012Accepted in revised form 15 February 2012Available online 10 March 2012

Keywords:SuperhydrophobicityNano-silicaAluminumEISNanoindentationNeural networks

Nano-silica particles were deposited on acid-etched hydrophilic aluminum (Al) substrates by immersion inwell-dispersed nano-silica aqueous suspension and tetramethylamonium hydroxide, followed by a heattreatment. The surface was then further treated by a reaction with fluorosilane. The hydrophobicity, surfacemorphology, and mechanical properties of the coated Al substrates were investigated, along with their elec-trochemical properties over time of exposure to two NaCl solutions (0.3% and 3% by weight). All the coated Alsurfaces exhibited a water contact angle of 155–158°, i.e., superhydrophobicity. The use of nano-silica sus-pension significantly enhanced the hydrophobicity of the coated Al. Artificial neural networks were used toprovide quantitative understanding in how the microstructure of the treated Al surface contributed to itssuperhydrophobicity and electrochemical properties. When Ra, total (nano-roughness+micro-roughness) ex-ceeds 450 nm, WCA is greater than 154°, independent of the nano/micro-roughness ratio (RRNM). FESEM andAFM images of these surfaces suggest that a rough two-length-scale hierarchical structure coupled with thelow surface energy of fluorosilane topcoat led to the superhydrophobicity of the formed coatings. The coatingprepared with the 0.2% nano-silica suspension (vs. other concentrations) featured the highest Young's mod-ulus and the best corrosion protection to the Al substrate in both NaCl solutions.

Published by Elsevier B.V.

1. Introduction

Superhydrophobic surfaces, defined as surfaces with a watercontact angle (WCA) greater than 150°, have recently received sig-nificant attention because of their importance in many technologicalapplications. For example, contamination, corrosion, snow accumu-lation, and several other undesirable effects developed on outdoorsurfaces can be inhibited by the use of coatings that exhibit en-hanced water repellency [1–6]. Manca et al. [1] reported the use offunctionalized silica nanoparticles and a sol–gel process to fabricatea durable self-cleaning and antireflective coating on the glass sub-strate. He et al. [2] reported that a superhydrophobic surface treat-ment of anodized aluminum by myristic acid to significantlyimprove its corrosion resistance in sterile seawater, at least overthe 24 h of immersion. Kato et al. [6] found that by altering the che-mistry and morphology of a surface (e.g., introducing hydrophilic

ble Infrastructure Laboratory,ana State University, Bozeman,406 994 1697.i).work as the first author.

.V.

channels to a superhydrophobic surface) one could change itssnow adhesion and snow-sliding performance. Sarkar and Farzaneh[7] reported that ice spontaneously de-bonded from the surfacetreated with a superhydrophobic coating whereas ice growing on abare aluminum surface had adhesion strength of 369±89 kPa. Kuli-nich and Farzaneh [8,9] also suggested the superhydrophobic coat-ings with low WCA hysteresis to feature very low ice adhesionstrength. Li et al. [10] presented a new type of polydimethylsiloxane(PDMS)/nano-silica hybrid superhydrophobic coating to reduce iceaccumulation on glass insulators. They found that both multi-scalemicrostructure and low surface energy contributed to the superhy-drophobicity of the treated glass surface. Genzer and Efimenko[11] reviewed the design of superhydrophobicity (i.e., extremenon-wettability) on physically smooth, physically rough, or respon-sive surfaces and discussed its implications to the fabrication ofmarine antifouling surfaces.

It is well known that hydrophobic properties may be enhanced byan increased surface roughness [12,13], and superhydrophobic sur-faces require appropriate surface roughness and low surface energy[14]. As such, techniques to fabricate superhydrophobic surfaces canbe generally divided into two categories: making a rough surfacefrom a low-surface-energy material, or modifying a rough surfacewith a low-surface-energy material. In the first category, numerous

Fig. 1. A typical multi-layer feed-forward ANN architecture [25].

3701X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

methods using low-surface-energymaterials such as fluorocarbons andsilicones have been reported [14]. The second category usually issuitable for various hydrophilic substrates, which are required tofeature a high roughness prior to its modification by a low-surface-energy material (e.g., silanes). There are many ways to make roughsurfaces, including mechanical stretching, laser/plasma/chemicaletching, lithography, sol–gel processing and solution casting, layer-by-layer and colloidal assembling, electrical/chemical reaction anddeposition, electrospinning and chemical vapor deposition. Notethat as the roughness of a surface increases its water contact angletends to increase or decrease depending on whether the surface ishydrophobic or hydrophilic [12]. Furthermore, Zhu et al. [15] sug-gested that both the nanostructure and microstructure of a surfacecontribute to its superhydrophobicity. Hong et al. [16] concludedthat the hydrophobicity of many naturally occurring surfaces “arisesfrom their hierarchical surface roughness at two different lengthscales (specifically, at ca. 3–10 μm and about 100 nm length scales)”.

Nanoparticles have been recently used to achieve superhydro-phobicity via surface roughening. In one method, nanoparticleswere deposited on smooth or microroughened substrates to inducesubmicron-scale surface roughness on substrates. The roughenedsubstrates were then chemically treated or covered with low-surface-energy coatings to enhance the hydrophobic character ofthe produced surfaces [17–19]. In another method, nanoparticleswere mixed with polymer solutions and then deposited on smoothsurfaces such as glass and silicon [20–23].

Aluminum (Al) is remarkable for its high corrosion resistance andlow density, and Al alloys feature high specific strength, excellentheat and electrical conductivities, and low-specific weight. Structural

Fig. 2. WCA of Al substrates with o

components made from Al and its alloys are vital to the aerospace,building, transportation, energy, and many other industries. In thiscontext, recent years have seen great interest and increased explora-tion in the fabrication of superhydrophobic surfaces on Al substrates.One interesting application of such technology is the anti-icing coatingon aircraft wings or on electric transmission lines. Onda et al. [24]produced one of the first synthetic superhydrophobic surfaces(wetting angles >140°) by anodically oxidizing an Al surface andtreating the oxidized surface with a fluorosilane. They suggestedthat the hydrophobicity was due to the fractal nature of the roughsurface. Tadanaga et al. [25] produced a superhydrophobic surfacewith a special heat treatment of an Al oxide surface film formedby a chemical precursor method. When heat treated, the Al oxidecrystallized to form a topography that resembled tall hills anddeep valleys. This special surface became superhydrophobic afterbeing coated with a monolayer of fluoroalkyltrichlorosilane mole-cules. Guo et al. [26] reported the fabrication of superhydrophobicsurfaces on Al alloy by wet chemical etching (in NaOH solution)coupled with modification by organic agents such as:(i) poly(dimethysiloxane) vinyl terminated (PDMSVT) and 1 wt.%curing agent as the crosslinker, (ii) perfluorononane (C9F20), and(iii) perfluoropolyether (PFPE), respectively. These modified filmson Al alloy were obtained by spin-coating and subsequent heatcuring. Zhang et al. [27] reported that a superhydrophobic surfacetreatment of porous anodic alumina (PAO)/Al substrate with aZnAl-layered double hydroxides (LDHs)-laurate film to “providean effective barrier to aggressive species” and to significantlyimprove its corrosion resistance in 3.5% NaCl solution, at leastover the 21-d immersion.

Klein et al. [28] showed that the superhydrophobic surface on Alcan be achieved with very small dosage of nano-particles. After etch-ing in acid solution, Al substrates were dip-coated in dilute suspen-sions of nano-silica particles and then heat treated to partially sinterthe particles to the surface. A reaction of the resulting surface witha solution containing fluorosilane molecules led to the ultimate fabri-cation of superhydrophobic surfaces. Wetting measurements showedthat the contact angle between the surface and water droplets in-creased with decreasing area coverage of nano-silica spheres. None-theless, the electrochemical and mechanical properties of suchsuperhydrophobic Al surfaces have not been extensively investigatedthus far, despite the important implications of these properties in de-fining the durability and reliability of the surfaces.

In this context, this work is dedicated to exploring the electro-chemical and mechanical properties of superhydrophobic Al sub-strates modified with nano-silica and fluorosilane. In addition to

r without nano-silica coatings.

3702 X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

WCA measurements, scanning electron microscopy (SEM) andatomic force microscopy (AFM) are used to characterize the lotus-like hierarchical structure of the coating surfaces. Electrochemicalimpedance spectroscopy (EIS) and potentiodynamic weak polari-zation (PWP) are used to characterize the properties of these coat-ings in NaCl solutions. An AFM-based nanoindentation method isalso utilized to quantify the Young's modulus of the coatings atthe nanometer scale. With the aid of neural network modeling,this work also aims to provide quantitative understanding inhow the microstructure of the treated Al surface contributed toits superhydrophobicity.

2. Experimental

2.1. Coating preparation

The silica slurry (Snowtex-OL, Nissan Chemicals, Tokyo, 20 wt.%,particle size 45±5 nm, pH 3±1) was diluted with deionized waterto a concentration of 0.025, 0.05, 0.1, and 0.2 wt.% respectively. Foreach solution, its pH was then adjusted to 6.0 with tetramethylamo-nium hydroxide (TMAOH) to produce well-dispersed slurry. Theseconcentrations were selected to study the effect of surface roughnesson the hydrophobicity of the coating.

Al substrates (99.9%, Metals Supermarkets Co. Ltd., thickness0.05 in., unpolished) were cut into 1×1 in.2 coupons. The Al couponswere cleaned by ultra-sonication first in acetone, then in a cleaningsolution (90% of 98% H2SO4, 10% of 30% H2O2) for 5 min, and finallyin a 1% HF solution for 5 min, after which they were rinsed with deio-nized water. Thereafter, these etched substrates were immediatelyimmersed into one of the four aforementioned silica slurries for20 min.

The coated substrates were then heat treated at 200 °C for120 min. After heating, the Al coupons were placed in deionizedwater for 20 min to ensure that the surfaces were sufficiently hydrat-ed to allow a reaction with the fluoroalkyltrichlorosilanes [26].

To obtain a superhydrophobic surface, the silica-coated Al cou-pons were dried in air at room temperature and then placed in a mix-ture of 0.4 ml 1H,1H,2H,2H perfluorodecyltrichlorosilane (FDTS,Lancaster Synthesis, Windham, NH), 3 ml chloroform, and 30 ml hex-adecane for 24 h in a closed container, after which they were rinsedwith chloroform, then dried in air. Note that FDTS can form a mono-layer on the substrate, which serves to reduce capillary forces andvan der Waals interactions on the surface and to greatly lower its fric-tion coefficient [29].

a b

Fig. 3. Shape of water droplets on: a) as-received Al substrate; b) acid-etched Al sub

2.2. Superhydrophobicity evaluation

The superhydrophobicity of surfaces was evaluated by measuringthe contact angle formed between water drops and the surface of themodified samples using a contact angle measuring system (VCA 2500XE, ASC products). Specifically, the drops of water (nanopore water)were mounted on three different areas of the surface with a micro-syringe (needle diameter of 0.4 mm). The reported results are themean value of at minimum three measurements on different partsof the film. Photos of water drops on samples were obtainedfrom the monitor, based on which the WCA was calculated by thevendor-provided software.

2.3. Morphological studies

Images of each surface of interest were obtained with a field emis-sion scanning electron microscope (FESEM, Zeiss Supra 55VP) and anatomic force microscope (AFM, Dimension 3100 from Veeco, SantaBarbara, CA).

2.4. Surface roughness measurements

Numerous studies have confirmed that lotus leave features a com-bination of micrometer-scale and nanometer-scale roughness and alow surface energy, which are collectively responsible for its high ap-parent WCA (greater than 150°), low sliding angle, and self-cleaningeffect [30–32]. As such, the roughness of surfaces in this work wascharacterizedwith two different tools, to shed light on themicrostruc-ture of the surface respectively. For both, the parameter Ra was used todescribe the surface roughness, which is a commonway of quantifyingthe height variations of a given surface. Ra is the arithmetic mean ofthe absolute value of profile derivation from the mean within sam-pling length.

For micro-roughness of surfaces, a hand-held roughness tester(Model TR200, Time Group Inc., Beijing, China) with cut-off length of0.25 mm to 2.5 mm was used. For this study, the portable tester wasused to measure at minimum three 0.8×5 mm2 areas, from which anaverage Ra value was calculated. For nano-roughness of surfaces, theAFM was used to image at minimum three 10×10 μm2 areas, fromwhich an average Ra value was calculated.

2.5. Electrochemical measurements

Electrochemical measurements were conducted using a con-ventional three-electrode system. The coated Al coupon served as

strate coated first with 0.05% nano-SiO2 suspension and then with fluorosilane.

a

b

Fig. 4. Roughness (Ra) values of Al substrates with or without nano-silica coatings: a) micro-roughness over 0.8×5 mm2 areas; b) nano-roughness over 10×10 μm2 areas.

Fig. 5. Predicted WCA of coated Al surfaces as a function of Ra, total and RRNM.

3703X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

the working electrode, while the counter electrode and the referenceelectrode used were a platinum wire mesh and a saturated calomelelectrode (SCE) respectively. The corrosive solutions included0.3 wt.% and 3 wt.% aqueous NaCl solutions. Two methods wereused to assess the anticorrosion performance of the coatings: EISand PWP. The EIS measurements were carried using a GamryECM8 Multiplexer. The Al coupon was polarized at ±10 mV aroundits open circuit potential (OCP) by an alternating current (AC) signalwith its frequency ranging between 10 kHz and 10 mHz (10 pointsper decade). For PWP, the Al coupon was polarized at ±30 mVaround its OCP by a direct current (DC) signal at a scan rate of0.2 mV/s. Polarization resistance (RP) is defined by the slope of thepotential-current density plot at the corrosion potential.

2.6. Young's modulus measurement

The mechanical properties of superhydrophobic coatings weremeasured using an AFM based nanoindentation method [33]. For

each sample, 256 pairs of force vs. displacement curves wereobtained over the surface from a 5×5 μm2 area by subdividing thearea into 16×16 equal-sized pixels and acquiring one pair of curves

10 µm 10 µm

a b

Fig. 6. SEM micrographs of Al substrate: a) without coating; and b) coated with a nano-silica suspension at 0.1%, followed by the fluorosilane coating. Magnification: ×500.

3704 X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

from the center of each pixel. Each pair of curves includes a loadingand an unloading curve. The stiffness of each sample was evaluatedby fitting the corresponding loading curves to the Hertzian modelwhich approximates the tip geometry as a hemisphere [34]. Thecurves are fitted in such a way that only the initial part of the in-dentation (up to 1000 nN loading force) was selected for fitting sothat the indentation depth is limited to be no more than 100 nm,which is comparable to the diameter of the hemispherical tip. Asthe coating thickness is usually more than 20 μm, such an indenta-tion depth is small enough to minimize any potential effect of theAl substrate on the measured force vs. displacement curves.

2.7. Neural network modeling

It was envisioned that there would be complex cause-and-effectrelationships between the properties of treated Al substrates (e.g., hydro-phobicity and electrochemical parameters) and their surfacemorphologyaltered by the nano- and micro-modification. As such, artificial neuralnetworks (ANNs) were chosen to be the tool to quantify suchrelationships and to establish predictive models. ANNs present anon-parametric, self-adaptive approach to information processingand provide universal approximation functions flexible in tacklingcomplex, non-linear problems where conventional modelingtechniques (e.g., multiple regression) fail. This is generallyaccomplished without knowing the form of the predictive relation-ship a priori. The ANNs can produce generalizations from experienceeven if the data are incomplete or noisy, as long as over-fitting isavoided with expert intervention.

Multi-layer feed-forward ANNs were adopted for modeling, ofwhich a typical architecture is shown in Fig. 1 [35]. Every node simu-lates the function of an artificial neuron, of which the inputs arelinearly summated utilizing connection weights and bias terms andthen transformed via a non-linear transfer function. The nodes inthe input and output layers consist of influential variables and targetvariable(s), respectively. The number of hidden layers and hiddennodes is determined on a trial-and-error basis and generally dependson the complexity of relationship(s). Signals are propagated from theinput layer through the hidden layer(s) to the output layer, and eachnode in a layer is connected in the forward direction to every nodein the next layer.

In this study, a modified backpropagation (BP) algorithm wasemployed for the training of ANNs, in which a sigmoid function inEq. (1) was used as the non-linear transfer function. All the data for

input and output were normalized based on Eq. (2), where Xi andNXi are the ith value of factor X before and after the normalization,and Xmin and Xmax are the minimum and maximum value of factorX, respectively.

f xð Þ ¼ 1þ e−x� �−1 ð1Þ

NXi ¼Xi−Xmin þ 0:1ð Þ

Xmax−Xmin þ 0:1ð Þ ð2Þ

The network learns by adjusting itsweights and bias terms to reducethe difference between its output for each input pattern with a targetoutput for that pattern. All the connection weights and bias terms fornodes in different layers are initially randomized and then iterativelyadjusted based on certain learning rules. For each given sample, theinputs are forwarded through the network until they reach the layerproducing output values, which are then compared with the targetvalues. Errors are computed for the output nodes and propagatedback to the connections stemming from the input layer. The weightsare systematically modified to reduce the error at the nodes, first inthe output layer and then in the hidden layer(s). The changes inweightsinvolve a learning rate and amomentum factor (0.9 and 0.3 respectivelyfor this work) and are in proportion to the negative derivative of theerror term. It may take a few thousands of rounds, repeating the feed-forward and error back-propagation, before the predicted output getsvery close to the target value. The learning process continues withmultiple samples until the prediction error across all training samples(indicated by the sum of the mean squared error—SMSE) in the outputlayer converged to an acceptably low level. Thereafter, knowledge ofthe network remains encoded in the refined connection weights andbias terms that can be used to recall any trained sample or, if wellgeneralized, to predict unknown input–output pairs. A thoroughtreatment of the BP ANNs is beyond the scope of this paper, andthe detailed description of error propagation algorithm is providedelsewhere [36].

3. Results and discussion

3.1. Water contact angle of Al substrates

The WCA measurements were performed for various Al substrateswith orwithout the coatings. As shown in Fig. 2, the acid-etched Al sub-strates first treated with nano-silica suspension then with fluorosilane

3705X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

all showed a WCA in the range of 155–158°, i.e., superhydrophobicity.There is a slight increase in the WCA as the nano-silica particleconcentration in the suspension increased from 0.025% to 0.1%, butfurther increase to 0.2% in particle concentration did not show anybenefit in enhancing the WCA. In contrast, the as-received Al sub-strate featured a WCA of approximately 56°, i.e., being hydrophilic(WCA less than 90°). Such benefits of the coatings were furtherconfirmed by Fig. 3, where the as-received Al substrate showshydrophilicity and the coated Al substrate showed outstandinghydrophobicity.

It is interesting to note that the hydrophobicity of the coated Alsubstrates was a joint outcome of various treatments steps. For in-stance, in the absence of the nano-silica suspension, the Al substrateswith the fluorosilane coating featured a WCA of approximately 116°and 73° respectively, depending on whether the Al substrate was

Fig. 7. AFM images (10×10 μm2) of: a)–d) surfaces 1–4 [Al substrate coated with a nafluorosilane coating].

acid-etched or not. In the absence of the fluorosilane coating, theacid-etched Al substrates treated with a 0.2% nano-silica suspensionshowed a WCA of approximately 30°. As such, we can conclude thatboth the acid-etching and the fluorosilane coating were essential forachieving hydrophobicity of Al surfaces and the use of nano-silicasuspension significantly enhanced the hydrophobicity of the coatedAl substrates.

3.2. Surface morphology of Al substrates

In light of the WCA data, we hypothesize that the two-layer coat-ing system is designed such that the nano-silica particles in the firstlayer not only form chemical bonding to the functional groups inthe acid-etched Al substrate but also form surface profile contributing

no-silica suspension at 0.025%, 0.05%, 0.1%, and 0.2% respectively, followed by the

Fig. 7. (continued).

3706 X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

to the ultimate superhydrophobicity. To this end, the surface profileof various Al substrates was investigated at two different lengthscales, with one Ra to characterize the micro-roughness measuredwith a roughness tester over 0.8×5 mm2 areas (Fig. 4a) and theother Ra to characterize the nano-roughness measured with AFMover 10×10 μm2 areas (Fig. 4b). The data suggest that the nano-silica suspensions enhanced the surface micro-roughness of acid-etched Al substrates, regardless of the nano-silica concentration inthe suspension. The Al substrates treated with 0.1% and 0.05% nano-silica suspension featured the highest micro-roughness and nano-roughness, respectively. This differs from the findings of a study byManoudis et al. [30], where the surface nano-roughness of coatedglass substrates increased with the concentration of nano-silica parti-cles dispersed in a polymeric solution.

To investigate the role of surface profile in defining theWCAof coatedAl substrates, a predictive model was established to correlate twoinfluential variables (total roughness—Ra, total, nano/micro-roughnessratio—RRNM) and one target variable (WCA). Data from 10 coated Al sub-strates were used for training of the ANN model, whereas data from 2coated Al substrates were used for testing of the trained model. A 2-3-1BP ANN model was trained to allow for a reasonable training error anda reasonable testing error, i.e., SMSE both at 0.02. The modelpredictions are illustrated in Fig. 5, which reveals that when Ra, total isless than 450 nm, their WCA generally increases with Ra, total. Thisis similar to the findings of a study by Manoudis et al. [30],where the WCA increased with the nano-roughness of glass treatedwith a PMMA-SiO2 nanocomposite coating. Fig. 5 also shows that whenRa, total exceeds 450 nm,WCA is greater than 154°, independent of RRNM.

Fig. 8. A pair of representative force vs. displacement curves obtained by indenting thesuperhydrophobic coating prepared with 0.05% nano-SiO2 suspension using an AFMtip. (For interpretation of the references to color in this figure legend, the reader is re-ferred to the web version of this article).

Fig. 10. Schematic drawing of the equivalent circuit for fitting the EIS data. R0 is asso-ciated with the electrolyte resistance. C1 is capacitance of the coating. R1 (pore resis-tance) is the resistance of ion conducting paths that develop in the coating. C2 is thecapacitance of the double layer. R2 is the charge transfer resistance at the Al-coating in-terface. W is the Warburg diffusion impedance element [39].

3707X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

To shed more light on the surface morphology of Al substrateswith or without coatings, top-view FESEM and AFM images of thesesurfaces were taken, as shown in Figs. 6 and 7 respectively. These fig-ures reveal that at micron and submicron scales aggregates of nano-silica particles formed on the surface of acid-treated Al surfaces,which were still “visible” after the fluorosilane treatment. In light ofthe WCA data discussed earlier, such a rough “two-length-scale hier-archical structure” [30] coupled with the low surface energy of fluor-osilane topcoat led to the superhydrophobicity of the formedcoatings. Note that the submicron structures shown in the AFM im-ages (Fig. 7) resemble the rough surface hierarchical structure oflotus leaf at the micron scale [32].

The thickness of various superhydrophobic coatings was alsoevaluated by examining the cross-section of the coating/Al interfaceunder FESEM. The superhydrophobic coatings formed with 0.025%,0.05%, 0.1% and 0.2% nano-silica suspensions featured a typicalthickness of 45 μm, 84 μm, 83 μm, and 34 μm, respectively. Relativeto the two thicker superhydrophobic coatings (formed by 0.05% and

Fig. 9. Young's modulus of: a)–d

0.1% nano-silica suspensions respectively), the two thinner coatings(formed by 0.025% and 0.2% nano-silica suspensions respectively)also featured smoother surface profile at the submicron scale, as indi-cated by the AFM images (Fig. 7).

3.3. Young's modulus of hydrophobic coatings

Fig. 8 presents representative force-displacement curves for asuperhydrophobic coating, which were obtained using the AFM-based nanoindentation technique. The upper curve (black line)and the lower curve (red line) in the diagram correspond to theloading and the unloading processes of the indentation, respective-ly. A hysteresis region between the loading and unloading curveswas observed for all the tested samples, indicating a plastic defor-mation of the coatings upon the indentation. The nearly linearslope of the unloading curves (e.g., the red line in Fig. 8) indicatesthat the tip indentation resulted in very little elastic deformationand thus the loading curve was fitted to obtain the mechanicalproperties of the coating.

In this work, the coating is approximated as a homogenous filmwith identical mechanical properties along Z direction over thewhole thickness of the coatings. Such a simplified model allowsextracting the Young's modulus by fitting the initial part of the load-ing curve to the Hertzain model. The fitting is limited to the initial100 nm starting from the point of contact in order to maintain thehemisphere approximation of the tip geometry required by theHertzain model. As such, the representative values of the Young'smodulus of the outer coating layer are shown in Fig. 9 for coatings

) surfaces 1–4 respectively.

Fig. 11. EIS Nyquist diagrams of Al substrate in (left) 0.3% NaCl and (right) 3% NaCl solutions respectively, by time of immersion: a) without coating; b)–d) surfaces 1–4 respectively.

3708 X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

prepared at various concentrations of silica particles. Since the par-ticle size of SiO2 particles was about 45±5 nm and the highest con-centration used was only 0.2%, the coverage of the aluminumsurface by silica particles is rather low. Hence at the nanometerscale the AFM tip had rare chance to hit the particles directly,which means the contribution from the silica particles to the mea-sured Young's modulus is negligible. All the tested hydrophobiccoatings showed average Young's modulus values ranging from 31to 107 MPa, which are comparable to the documented values of20–560 MPa for organic coatings obtained using similar methods[37,38]. Fig. 9 suggests that the superhydrophobic coatings prepared

with the 0.2% nano-silica suspension (followed by the fluorosilanecoating) feature the highest Young's modulus.

3.3.1. Corrosion resistance of hydrophobic coatingsSuperhydrophobic coatings have been used for corrosion protec-

tion of metals. Zhu and Jin [39] reported that the superhydrophobicfilms preparing by silicon sol and fluoride latex substantiallyimproved the corrosion resistance of phosphating films on steelsubstrate in 5 wt.% NaCl solution. Barkhudarov et al. [40] reportedthe preparation of a rough, organosilica aerogel-like thin film(~500 nm) onto Al surfaces and demonstrated that a superhydrophobic

Fig. 11. (continued).

3709X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

coatingwith aWCAof >160°was able to greatly decrease the corrosionrate of Al in 5 wt.% NaCl D2O solution by preventing the infiltration ofwater into the highly porous superhydrophobic film. Liu et al. [41] dem-onstrated that a superhydrophobic surfacewith aWCA of 152°was ableto significantly decrease the marine bacterium Vibrio natriegens adhe-sion and thus greatly decrease the corrosion rate of Al in the NaCl-containing bacterial culture medium.

Etching is a straightforward and effective way to make roughsurfaces. Qian and Shen [42] described a simple surface rougheningmethod by chemical etching on polycrystalline metals such as Al.After treatment with fluoroalkylsilane, the etched metallic surfacesexhibited superhydrophobicity. In our work, chemical etching ofAl substrate was also performed prior to the fluorination. As dis-cussed earlier, both the acid-etching and the fluorosilane coatingwere essential for achieving hydrophobicity of Al surfaces. Similarresults have been reported previously [42], suggesting that the for-mation of the micro-pits by acid etching may have enhanced the

Table 1Example corrosion data of Al substrates as a function of exposure conditions and surface co

At 120 h in 3% NaCl

Surface 0 Surface 1 Surface 2 Surface 3 Surfac

Ecorr (mV, SCE) −1222 −749 −752 −855 −747Rp (Ω.cm2) 1.12E+04 1.73E+04 6.80E+04 2.27E+04 1.62ER1 (Ω.cm2) NA 253 208 169 284C1 (F.cm−2) NA 8.25E−06 3.36E−06 5.68E−06 2.64ER2 (Ω.cm2) NA 4.49E+03 1.33E+04 8.70E+03 3.95EC2 (F.cm−2) NA 3.00E−05 8.59E−06 1.75E−05 4.37EW (S.s1/2) NA 6.66E−04 2.06E−04 6.09E−04 8.16E

hydrophobicity of Al surface. This, however, presents a concern overthe corrosion resistance of the coated Al.

To investigate the durability of superhydrophobicity in corrosiveenvironments, the corrosion resistance of various Al substrates wastested with or without the coatings, by exposing them to 0.3 wt.%and 3 wt.% NaCl solutions for 140 and 120 h respectively. The corro-sion potential (Ecorr) and polarization resistance (RP) were estimatedfrom the periodically measured PWP curves of coated Al specimens.Furthermore, periodical EIS measurements provided information onelectrolyte-coating-Al interfaces.

The electrochemical impedance of coated metals has been widelystudied. The interpretation of impedance data from failed coatings canbe very complicated. For simplicity, this study used the equivalent cir-cuit shown in Fig. 10 to characterize the interfaces of interest and tobest fit the measured EIS data. The EIS parameters R1 and C1 are the re-sistance and capacitance of coating, characteristic of its pore networkstructure, whereas R2, C2 and W are the corrosion resistance of the Al,

ating.

At 140 h in 0.3% NaCl

e 4 Surface 0 Surface 1 Surface 2 Surface 3 Surface 4

−810 −654 −667 −641 −675+05 2.08E+05 1.51E+04 8.88E+05 4.06E+06 1.26E+06

NA 895 3483 7822 6295−06 NA 1.58E−06 1.26E−07 3.03E−07 1.47E−07+04 NA 4.99E+04 7.74E+04 4.22E+04 1.37E+05−06 NA 1.54E−06 2.94E−07 4.15E−07 2.78E−07−05 NA 1.99E−05 4.59E−06 1.58E−06 6.11E−06

3710 X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

the double layer capacitance on the Al surface, and the Warburg diffu-sion impedance respectively. Most EIS Nyquist diagrams of coated Alsubstrates (as shown in Fig. 11) featured two capacitive loops attrib-utable to the resistances and capacitances of the degrading coatingover time. For some coated Al specimens (e.g., those with 0.025%nano-silica suspension) after a certain time of exposure, a diffusionline was observed in the second loop at low frequencies, indicatingthe penetration of water into the coating and the formation of anew electrolyte/Al interface under the coating.

Table 1 presents some example corrosion data of Al substratesobtained by electrochemical measurements, as a function of exposureconditions and surface coating. Note that surface 0 is the Al substratewithout the acid-etching or nano-silica treatment but with the fluor-osilane topcoat. Surfaces 1–4 are the acid-etched Al substrate coatedwith a nano-silica suspension at 0.025%, 0.05%, 0.1%, and 0.2% respec-tively, followed by the fluorosilane coating. The RP, R1 and R2 data allsuggest that 3% NaCl solution was generally more corrosive than the0.3% NaCl solution to the coated Al substrates. The R2 (charge trans-fer resistance) and C2 (double layer capacitance) data suggest thatthe hydrophobic coatings prepared with 0.2% nano-silica suspensionprovided the best corrosion protection to the Al substrate in bothNaCl solutions. After 120 h in 3% NaCl solution, the hydrophobiccoating prepared with 0.2% nano-silica suspension featured thehighest R1 (pore resistance) and lowest C1 (coating capacitance),suggesting more compact, impermeable microstructure. In bothNaCl solutions, all the superhydrophobic coatings increased theEcorr of Al substrates. Note that the superhydrophobic coatings pre-pared with 0.025% nano-silica suspension reduced the polarizationresistance of Al substrate, i.e., increased its corrosion rate. This high-lights the need to have sufficient concentration of nanoparticles inthe suspension, in order to protect the acid-etched Al substrate;otherwise it can negatively affect the bonding of the topcoat tothe substrate. It was also observed that the protective property ofmost superhydrophobic coatings increased over the time of expo-sure in 0.3% NaCl solution, likely due to the formation of corrosionproducts that served to inhibit the propagation of corrosion.

Table 2 presents the initial surface profile of coated Al substratesas well as their electrochemical parameters obtained from the period-ical measurements over the time of exposure to the two NaCl solu-tions. These data are organized in such a manner to enable analysis

Table 2Initial surface profile and periodical electrochemical parameters of coated Al substrates.

Ra, total (nm) RRNM Ecorr (mV, SCE) RP (KΩ. cm2) R1 (Ω. cm

456 0.27 −721 3.5 407437 0.62 −715 32.0 1163669 0.18 −707 24.8 843468 0.17 −701 15.9 1478456 0.27 −721 3.0 241437 0.62 −739 6.3 256669 0.18 −813 19.4 181468 0.17 −761 52.3 229456 0.27 −749 17.3 253437 0.62 −752 68.0 208669 0.18 −855 22.7 169468 0.17 −747 161.5 284456 0.27 −622 136.2 8115437 0.62 −680 466.8 6359669 0.18 −682 632.4 5037468 0.17 −698 772.0 7950456 0.27 −630 6.8 704437 0.62 −655 92.9 4237669 0.18 −667 243.1 4740468 0.17 −685 332.7 6637456 0.27 −654 15.1 895437 0.62 −667 888.2 3483669 0.18 −641 4058.0 7822468 0.17 −675 1264.0 6295

of their internal correlations. For instance, the data were used toexplore inherent correlations between various EIS parameters(Fig. 12). Fig. 12a illustrates that there was a strong negative expo-nential relationship between R1 (pore resistance) and C1 (coatingcapacitance), as both of them are related to the permeability ofthe coating. Fig. 12b illustrates that there was a weak negative ex-ponential relationship between R2 (charge transfer resistance) andC2 (double layer capacitance), as both of them are related to thecorrosion resistance at the electrolyte/Al interface. Fig. 12c and dillustrate that there were strong positive relationships betweenW and C2 and between C2 and C1, respectively.

To investigate the role of initial surface profile and instantaneouselectrochemical condition in defining the corrosion resistance ofcoated Al substrates, three predictive models were established tocorrelate three influential variables (Ra, total, RRNM, Ecorr) and onetarget variable (RP, R1 or R2). These models were established usingthe data from Table 2, of which data from 23 coated Al substrateswere used to train the ANNs and data from the remaining 1 coatedAl substrate were used to test the trained models.

The topological structure of all three ANN models was deter-mined to be 3-4-1, on the basis of trial-and-error. The modeled RP,R1 or R2 values were plotted against the corresponding valuesobtained from EIS measurements and the following relations wereobserved for them respectively: for RP, y=0.932x (R2=0.95); forR1, y=0.951x (R2=0.96); and R2, y=0.902x (R2=0.85). Further-more, all the training and testing errors (SMSE) were less than0.08. These imply that the established ANN models have good“memory” and the trained matrices of interconnected weights andbias (“fabric” of the neural networks) reflect the hidden functionalrelationships well. As such, the models were used to predict theRP, R1 or R2 of coated Al substrates as a function of initial surfaceprofile and instantaneous Ecorr.

In the following sections, the model predictions are presented inthe form of three-dimensional response surfaces. Since ANNs arenot suitable for extrapolation, predictions were made with two influ-ential variables varying in the range of training data and the 3rd oneset at the median level.

Figs. 13a and 14a reveal that for coated Al surfaces with a moderateEcorr, the highest RP and R1 generally coincided with those featuring aninitial surface profile with low roughness (Ra, totalb500 nm) and a

2) C1 (F.cm−2) R2 (Ω.cm2) C2 (F.cm−2) W (S.s1/2)

0.00000160 2567 0.00000906 0.00091530.00000027 16,640 0.00000079 0.00026530.00000039 12,420 0.00000082 0.00009740.00000035 20,290 0.00000072 0.00016510.00000569 3589 0.00002582 0.00227000.00000305 7436 0.00000845 0.00048800.00000576 5044 0.00001496 0.00052510.00000254 11,700 0.00000416 0.00021770.00000825 4489 0.00002998 0.00066560.00000336 13,330 0.00000859 0.00020580.00000568 8698 0.00001754 0.00060900.00000264 39,530 0.00000437 0.00008160.00000008 17,360 0.00000033 0.00000130.00000001 13,180 0.00000002 0.00000120.00000015 26,330 0.00000021 0.00000140.00000005 141,000 0.00000011 0.00000290.00000131 3342 0.00000124 0.00022500.00000007 33,230 0.00000020 0.00000660.00000023 188,700 0.00000050 0.00001110.00000013 68,360 0.00000038 0.00002110.00000158 49,880 0.00000154 0.00001990.00000013 77,360 0.00000029 0.00000460.00000030 42,150 0.00000042 0.00000160.00000015 136,500 0.00000028 0.0000061

y = 0.0018x-1.136

R² = 0.856

0.000000

0.000003

0.000006

0.000009

0 2000 4000 6000 8000 10000

C2

(F.c

m-2

)C

2 (F

.cm

-2)

C1

(F.c

m-2

)

C1 (F.cm-2)

C1 (F.cm-2)

W (

S.s

0.5 )

EIS Data: Pore Capacitance vs. Resistance

y = 0.0252x -1

R² = 0.409

0.00000

0.00001

0.00002

0.00003

0 40000 80000 120000 160000 200000

R2 ( Ω Ω.cm2)

R1 ( Ω Ω.cm2)

EIS Data: Double Layer Capacitance vs. Resistanceb

y = 161.78x1.1124

R² = 0.7797

0.0000

0.0006

0.0012

0.0018

0.0024

0.00000 0.00001 0.00002 0.00003

EIS Data: Diffusion vs. Double Layer Capacitancec

y = 3.4369x - 8E-07R² = 0.91

0.00000

0.00001

0.00002

0.00003

0.000000 0.000003 0.000006 0.000009

EIS Data: Double Layer vs. Pore Capacitanced

a

Fig. 12. Correlations between various EIS parameters: (a) C1 vs. R1; (b) C2 vs. R2; (c) Wvs. C2; (d) C2 vs. C1.

Fig. 13. Predicted RP of coated Al surfaces as a function of: (a) Ra, total and RRNM, (b) Ecorrand RRNM.

Fig. 14. Predicted R1 of coated Al surfaces as a function of: (a) Ra, total and RRNM, (b) Ecorrand RRNM.

3711X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

RRNM less than 0.25 or more than 0.45. According to Fig. 13b, for coatedAl surfaces with a moderate Ra, total, the highest RP coincided with thosefeaturing Ecorr higher than−740 mV/SCE or lower than−820 mV/SCEand RRNM less than 0.3. According to Fig. 14b, for coated Al surfaceswitha moderate Ra, total, the highest R1 coincided with those featuring Ecorrhigher than −740 mV/SCE and RRNM less than 0.25 or higher than 0.4.Fig. 15a reveals that for coated Al surfaces with a moderate Ecorr, thehighest R2 generally coincided with those featuring an initial surfaceprofile with low roughness (Ra, totalb500 nm) and a RRNM less than0.35. Fig. 15b reveals that for coated Al surfaces with amoderate Ra, total,

the highestR2was foundwith those featuring Ecorr in the range of−650to −740 mV/SCE and RRNM preferably less than 0.5.

To obtain best possible corrosion resistance of the superhy-drophobic coatings, it is desirable to achieve high RP, R1 and R2 ofcoated Al substrates concurrently. In light of the model predictionsillustrated in Figs. 13–15, the initial surface profile of coated Alsubstrate should feature low roughness (Ra, totalb500 nm) and a

Fig. 15. Predicted R2 of coated Al surfaces as a function of: (a) Ra, total and RRNM, (b) Ecorrand RRNM.

Fig. 17. Shape of water droplets on a hydrophobic coating prepared with 0.05% nano-silica suspension: a) before, and b) after: 120 h in 3% NaCl solution.

3712 X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

RRNM less than 0.25 and the coated Al substrate should maintain aEcorr in the range of −650 to −740 mV/SCE.

3.4. Water contact angle of Al substrates after chloride exposures

After the exposure to the NaCl solutions, the WCA measurementswere performed again for various coated Al substrates. The datasuggest that after 140 h in 0.3% NaCl solution or 120 h in 3% NaCl,the coatings lost their hydrophobicity, as shown in Fig. 16. The dra-matic change in hydrophobicity was further confirmed by Fig. 17,where the coated Al substrate showed outstanding hydrophobicityprior to the chloride exposure but became hydrophilic afterwards.We hypothesize that this occurred as the NaCl solutions compro-mised the integrity of the fluorosilane topcoat and as the corrosionproducts forming under the coating altered the microstructure ofthe surface layer. Fig. 16 also indicates that for the hydrophobiccoatings prepared with relatively high nano-silica concentrations(0.1% and 0.2%) the residual effects of hydrophobicity were moreapparent. These laboratory data suggest that aging in the field envi-ronment presents a durability concern for such superhydrophobic

Fig. 16. WCA of coated Al substrates coatings after 140 h in 0.3% NaCl solution or after120 h in 3% NaCl solution, respectively: Comparison by superposition.

surfaces. As the air-borne chloride in marine environments is inconcentrations typically much lower than 0.3% NaCl, these surfacesare expected to lose their superhydrophobicity over a time scale ofthousands of hours, or a few years. Furthermore, aging of the top-coat by UV light is another concern. Future work will focus onsuperhydrophobic coatings that incorporate additives to slowdown their aging and to preserve their superhydrophobicity, in-cluding: corrosion-inhibiting additives or pigments with specificlight absorbance and heat-transfer properties (e.g., organic carbonblack, Fe2O3\Cr2O3, MnO2, and FeMnCuOx of micron and sub-micron dimensions).

4. Conclusions

A variety of superhydrophobic coatings were successfully pre-pared on Al substrates, through acid etching followed by modifica-tions with nano-silica (with 0.025%, 0.05%, 0.1%, 0.2% suspensions)and fluorosilane. The hydrophobicity, surface morphology, andmechanical properties of the coated Al substrates were investigated,along with their electrochemical properties over the time of expo-sure to two NaCl solutions (0.3% and 3% by weight). The mainfindings from this study are provided as follows.

1. The acid-etched Al substrates first treated with nano-silica suspen-sion then with fluorosilane all showed a WCA in the range of155–158°, i.e., superhydrophobicity. There is a slight increase inthe WCA as the nano-silica particle concentration in the suspen-sion increased from 0.025% to 0.1%, but further increase to 0.2%

3713X. Shi et al. / Surface & Coatings Technology 206 (2012) 3700–3713

in particle concentration did not show any benefit in enhancingthe WCA. The Al substrates treated with 0.1% and 0.05% nano-silica suspension featured the highest micro-roughness andnano-roughness, respectively.

2. From a modeling perspective, ANNs were used to achieve betterunderstanding of the complex cause-and-effect relationships in-herent in the superhydrophobic Al surfaces and have demonstratedgreat potential in finding meaningful, logical results from the ex-perimental data. The model predictions reveal that when Ra, total(nano-roughness+micro-roughness) is less than 450 nm, theWCA of coated Al substrates generally increases with Ra, total.When Ra, total exceeds 450 nm, WCA is greater than 154°, indepen-dent of the nano/micro-roughness ratio RRNM. Top-view FESEMandAFM images of these surfaces suggest that a rough two-length-scale hierarchical structure coupled with the low surface energyof fluorosilane topcoat led to the superhydrophobicity of theformed coatings.

3. The data from AFM-based nanoindentation of the coated Al surfacessuggest that the superhydrophobic coatings prepared with the 0.2%nano-silica suspension (followed by the fluorosilane coating) fea-tured the highest Young'smodulus. The electrochemical data suggestthat the same coatings provided the best corrosion protection to theAl substrate in both NaCl solutions, likely linked to their more com-pact, impermeable microstructure. In both NaCl solutions, all thesuperhydrophobic coatings increased the Ecorr of Al substrates. Inlight of the ANN model predictions, in order to achieve the bestpossible corrosion resistance for coated Al substrate, its initial sur-face profile should feature low roughness (Ra, totalb500 nm) and aRRNM less than 0.25.

Acknowledgments

The authors gratefully acknowledge financial support by theChuTian Scholar Visiting Professorship Fund provided by theHubei Department of Education, China, as well as financial supportby the U.S. DOT Research and Innovative Technology Administration.The authors also acknowledge the assistance of Prathish Kumar inconducting some of the later-stage validation experiments for thiswork.

References

[1] M. Manca, A. Cannavale, L. De Marco, A.S. Aric , R. Cingolani, G. Gigli, Langmuir25 (11) (2009) 6357.

[2] T. He, Y. Wang, Y. Zhang, Q. Lv, T. Xu, T. Liu, Corros. Sci. 51 (2009) 1757.[3] M. Qu, B. Zhang, S. Song, L. Chen, J. Zhang, X. Cao, Adv. Funct. Mater. 17 (2007) 593.[4] A. Nakajima, K. Hashimoto, T. Watanabe, Monatsh. Chem. 132 (2001) 31.[5] H.Y. Erbil, A.L. Demirel, Y. Avci, O. Mert, Science 299 (2003) 1377.[6] T. Kato, A. Nakajima, H. Irie, Z. Kato, K. Uematsu, T. Watanabe, K. Hashimoto,

J. Mater. Sci. 39 (2004) 547.[7] D.K. Sarkar, M. Farzaneh, J. Adhes. Sci. Technol. 23 (2009) 1215.[8] S.A. Kulinich, M. Farzaneh, Appl. Surf. Sci. 255 (2009) 8153.[9] S.A. Kulinich, M. Farzaneh, Langmuir 16 (2009) 8854.

[10] J. Li, Y. Zhao, J. Hu, L. Shu, X. Shi, J. Adhes. Sci. Technol. (in press), doi:10.1163/016942411X574826.

[11] J. Genzer, K. Efimenko, Biofouling 22 (5) (2006) 339.[12] H.M. Shang, Y.Wang, K. Takahashi, G.Z. Cao, D. Li, Y.N. Xia, J. Mater. Sci. 40 (2005) 3587.[13] R.N. Wenzel, J. Phys. Colloid Chem. 53 (1949) 1466.[14] M. Ma, R.M. Hill, Curr. Opin. Colloid Interface Sci. 11 (2006) 193.[15] M. Zhu, W. Zuo, H. Yu, W. Yang, Y. Chen, J. Mater. Sci. 41 (2006) 3793.[16] J. Hong,W.K. Bae, H. Lee, S. Oh, K. Char, F. Caruso, J. Cho, Adv.Mater. 19 (2007) 4364.[17] L. Zhai, F.C. Cebeci, R.E. Cohen, M.F. Rubner, Nano Lett. 4 (2004) 1349.[18] J. Bravo, L. Zhai, Z. Wu, R.E. Cohen, M.F. Rubner, Langmuir 23 (2007) 7293.[19] H.-M. Bok, T.-Y. Shin, S. Park, Chem. Mater. 20 (2008) 2247.[20] C.-T. Hsieh, J.-M. Chen, R.-R. Kuo, T.-S. Lin, C.-F. Wu, Appl. Surf. Sci. 240 (2005) 318.[21] E. Chibowski, L. Hołysz, K. Terpilowski, M. Jurak, Colloids Surf., A 291 (2006) 181.[22] M.Y. Yüce, A.L. Demirel, F. Menzel, Langmuir 21 (2005) 5073.[23] M.Y. Yüce, A.L. Demirel, Eur. Phys. J. B 64 (3–4) (2008) 493.[24] T. Onda, S. Shibuichi, N. Satoh, K. Tsuji, Langmuir 12 (9) (1996) 2125.[25] K. Tadanaga, N. Katata, T. Minami, J. Am. Ceram. Soc. 80 (1997) 1040.[26] Z. Guo, F. Zhou, J. Hao, W. Liu, J. Colloid Interface Sci. 303 (2006) 298.[27] F. Zhang, L. Zhao, H. Chen, S. Xu, D.G. Evans, X. Duan, Angew. Chem. Int. Ed. 47 (2008)

2466.[28] R.J. Klein, P.M. Biesheuvel, B.C. Yu, C.D. Meinhart, F.F. Lange, Z. Met. 94 (4) (2003) 377.[29] M. Cichomski, J. Grobelny, G. Celichowski, Appl. Surf. Sci. 254 (2008) 4273.[30] P.N.Manoudis, I. Karapanagiotis, A. Tsakalof, I. Zuburtikudis, C. Panayiotou, Langmuir

24 (2008) 11225.[31] A. Nakajima, K. Hashimoto, T. Watanabe, Monatsh. Chem. 132 (2001) 31.[32] W. Barthlott, C. Neinhuis, Planta 202 (1997) 1.[33] Z. Suo, F.T. Arce, R. Avci, K. Thieltges, B. Spangler, Langmuir 22 (2006) 3844.[34] A. Meroufel, C. Deslouis, S. Touzain, Electrochim. Acta 53 (2008) 2331.[35] X. Shi, P. Schillings, D. Boyd, Int. J. Prod. Res. 42 (1) (2004) 101.[36] D.E. Rumelhart, G.E. Hinton, R.J. Williams, in: D.E. Rumelhart, J.L. McClelland, PDP

Research Group (Eds.), Parallel Distributed Processing, MIT Press, Cambridge, MA,1986, p. 318.

[37] X. Shi, T.A. Nguyen, Z. Suo, Y. Liu, R. Avci, Surf. Coat. Technol. 204 (3) (2009) 237.[38] R. Vanlandingham, S.H. McKnight, G.R. Palmese, J.R. Elings, X. Huang, T.A. Bogett,

R. Eduljee, J.W. Gillespie, J. Adhes. 64 (1997) 31.[39] L. Zhu, Y. Jin, Appl. Surf. Sci. 253 (2007) 3432.[40] P.M. Barkhudarov, P.B. Shah, E.B. Watkins, D.A. Doshi, C.J. Brinker, J. Majewski,

Corros. Sci. 50 (3) (2008) 897.[41] T. Liu, L. Dong, T. Liu, Y. Yin, Electrochim. Acta 55 (2010) 5281.[42] B.T. Qian, Z.Q. Shen, Langmuir 21 (2005) 9007.