Splat morphology of plasma sprayed aluminum oxide reinforced with carbon nanotubes: A comparison between experiments and simulation

Transcription

1 Accepted Manuscript Splat morphology of plasma sprayed aluminum oxide reinforced with carbon nanotubes: A comparison between experiments and simulation Anup Kumar Keshri, Arvind Agarwal PII: S (11) DOI: doi: /j.surfcoat Reference: SCT To appear in: Surface & Coatings Technology Received date: 21 December 2010 Revised date: 30 April 2011 Accepted date: 11 July 2011 Please cite this article as: Anup Kumar Keshri, Arvind Agarwal, Splat morphology of plasma sprayed aluminum oxide reinforced with carbon nanotubes: A comparison between experiments and simulation, Surface & Coatings Technology (2011), doi: /j.surfcoat This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

2 Splat Morphology of Plasma Sprayed Aluminum Oxide Reinforced with Carbon Nanotubes: A Comparison between Experiments and Simulation Anup Kumar Keshri 1, 2 and Arvind Agarwal 1,* 1 Plasma Forming Laboratory Mechanical and Materials Engineering, Florida International University, Miami, FL 33174, USA 2 Manufacturing Division, School of Mechanical and Building Sciences Vellore Institute of Technology University, Vellore , Tamil Nadu, India Abstract This study elucidates the effect of carbon nanotube (CNT) addition on the splat formation in plasma sprayed aluminum oxide (Al 2 O 3 ) composite coating using experimental and computational methods. CNT content was varied as 0, 4 and 8 wt. % in Al 2 O 3 matrix. With an increasing CNT content, splat morphology became more circular and diskshaped. The average diameter of disk-shaped splats increased from 28.6±1.4 µm for Al 2 O 3 to 43.2±1.3 µm for Al 2 O 3-8 wt. % CNT. The population density of splats with fingers, fragments, and voids was the lowest for the highest (8 wt. %) CNT content. The addition of CNTs resulted in two simultaneously competing phenomena viz. increased thermal capacity and increased viscosity of the melt. Increased thermal capacity delayed the localized solidification resulting in higher splat diameter while agglomeration of CNTs at the periphery of the splat results higher viscosity of the melt which suppresses the splat fragmentation that leads to increased population density of disc shaped splats. Splat morphology of three compositions was also simulated using SIMDROP software, which showed a good agreement with the experimentally collected splats. Keywords: Aluminum oxide, Carbon nanotube, Plasma spraying, Splat * Corresponding Author: Arvind Agarwal, Ph: , Fax: agarwala@fiu.edu 1

3 1. Introduction Splat is the smallest unit of the microstructure of plasma sprayed coatings. The properties of the coatings are largely dependent on the splat morphology and their stacking [1-4]. Incomplete melting and improper stacking of splats can result in globular voids, poor adhesion at the inter-splat boundary which can have adverse effect on the mechanical, thermal and electrical properties of the coatings [5-7]. Hence splat formation and its morphology play a significant role in tailoring the coating properties. Several studies have been reported on the morphological aspect of splats [1, 3, 4, 8-10] which suggest that splat morphology largely dependent on (i) feedstock material properties [8], (ii) thermal and kinetic state of the in-flight particle [8, 10] and (iii) substrate chemical state, roughness and temperature [3, 4, 11]. Elsebaei et al. [8] performed a study on the morphology of individual splats for different set of plasma operating parameters (arc current: 700, 870 A, stand-off distance: 80, 100 mm) for the regular yittria stabilized zirconia (YSZ) (particle size: µm) and the spherical agglomerate of YSZ (agglomerate size: µm) synthesized from the nano-ysz powder particle. Spherical agglomerate of nano YSZ was used with the intent to melt the periphery of the agglomerate and to retain the nano-features in the core. Such coating resulted in bimodal microstructure. Lima et al. [12] studied thermal spray coatings synthesized from the nanostructured ceramic agglomerated powder and concluded that it was necessary to avoid the full melting of the agglomerates to preserve nanostructure in the coating. Elsebaei et al [8] concluded that circularity (degree of roundness) and flattening degree (ratio of the diameter of splat to starting droplet) of YSZ 2

4 splats synthesized from the spherical agglomerate were larger than the regular-ysz splats. This was attributed to the smaller particle size, higher particle temperature and velocity of agglomerated-ysz particles [8]. Splats synthesized using spherical agglomerates of YSZ were more disk shaped (i.e. better circularity) as compared to regular-ysz splats at a higher stand-off distance [8]. This was due to the higher particle velocity of agglomerated-ysz compared to regular-ysz particle [8]. Bianchi et al. [13] deposited single splats by spraying micron-size zirconia on cold 304L stainless steel substrate (~100 C) and observed a highly fragmented morphology. Perfect disc shaped splats were obtained for the substrate heated to 300 C [13]. Fukumoto et al. [14, 15] studied the relationship between the splat morphology (from the micron-sized feedstock particles) and the substrate temperature and observed distinct changes in the splat morphology. Threshold transition temperature for the substrate was first identified by Fukumoto et al. [15], beyond which splat morphology changes from fragmented to disk shaped. Sampath et al. [3, 4] carried out a study on the effect of substrate temperature on the splat formation for partially stabilized micron-sized zirconia particles. The threshold transition temperature was found in the range of C [3]. In the case of cold substrates, initiation or localized solidification is responsible for spreading instability which leads to flattening splashing [3, 4]. Preheated substrate above the transition temperature provides better contact and uniform heat conduction which minimizes the localized solidification [3, 4]. The substrate heating also allows condensates and adsorbates desorbtion. Li et al. [16] studied the effect of substrate preheating temperature and surface organic covering on splat formation. Splats (aluminum, nickel, copper, Al 2 O 3 and molybdenum) were deposited on polished stainless 3

5 steel substrate surface covered with different organic substances (xylene, glycol and glycerol). It was found that when the preheating temperature exceeded 50 C over the boiling point of organic substance, the regular disk type splats were formed [16]. An optimum substrate preheating is required to strike a balance between better splat formation and minimal residual stress. These studies indicate that powder feedstock, processing conditions and substrate conditions have significant effect on the splat formation. Our research group has worked extensively on the synthesis of CNT reinforced aluminum oxide coatings by plasma spray technique [17-27]. Comprehensive process maps have been successfully developed to synthesize lowest porosity plasma sprayed Al 2 O 3 composite coatings with 0, 4 and 8 wt. % CNT reinforcement [26]. Balani et al. obtained ~200% improvement in the elastic modulus [19], 57% improvement in the fracture toughness [17] and 49 times enhancement in dry sliding wear resistance [21] by adding 8 wt. % CNTs in Al 2 O 3 coatings. Improved elastic modulus, fracture toughness and wear resistance was mainly attributed to excellent dispersion of CNTs in the Al 2 O 3 matrix that promotes toughening mechanisms such as CNT bridging, crack deflection at CNT/Al 2 O 3 interface [17-21]. A majority of our past work on Al 2 O 3 CNT coatings was focused on studying the materials property as function of process, microstructure and CNT content. However, the role of CNT in the splat formation was never addressed. Bakshi et al. [27] did a preliminary study on the role of CNT in metallic Al-Si splat formation and concluded that splat shape is governed by the viscosity and thermal conductivity of the droplet which are dependent on the CNT content. Higher CNT 4

6 content (10 wt. %) leads to disc shaped Al-Si splat as compared to lower CNT content (5 wt. %), which was attributed to increase in the viscosity of the melt due to increased CNT content [27]. Motivated by this scenario, the objective of this study is to understand the role of CNT in the Al 2 O 3 splat formation. The effect of varying CNT content on the splat morphologies has been investigated. Splat formation can be optimized by experimentation but it requires extensive and time consuming experiments due to large number of processing variables involved in plasma spraying. Splat morphology simulation for the given processing variables can save considerable amount of time. In this study, splat morphology simulation has also been performed using SIMDROP (Simulent Drop 3.0, Simulent Inc, Toronto, Canada) and comparison has been made with experimentally deposited splats. 2. Experimental 2.1 Powder Feedstock Sub-micron sized Al 2 O 3 (~150 nm, average diameter) powder and multiwall carbon nanotubes (95% +purity, nm outer diameter, μm in length) were used as starting materials. Since sub-micron sized fine powder and CNTs cannot be fed in the plasma flow using conventional carrier gas due to their high interparticle friction and resulting inconsistent flow, spray drying was implemented to manufacture micron-sized agglomerates. Spray drying also allows homogeneous dispersion of CNTs in Al 2 O 3 matrix. Sub-micron Al 2 O 3 powder was spray dried (referred as A-SD) to obtain spherical 5

7 agglomerates of 30±10 μm in diameter. The diameter of the spherical agglomerates was measured from the 5-6 different SEM images. From each image, measurements of agglomerate diameter were taken. Powder size distribution of A-SD, A4C-SD and A8C- SD are shown in Figure 1a-c respectively. Spray drying of sub-micron Al 2 O 3 with 4 wt. % CNTs (referred as A4C-SD) and 8 wt. % CNTs (referred as A8C-SD) resulted in spherical agglomerates of 26±7 μm and 24±5 μm respectively. A-SD powder served as the control sample to investigate the effect of CNT addition. 2.2 Synthesis of Single Splat A-SD, A4C-SD, A8C-SD powders were plasma sprayed using SG 100 gun (Praxair Surface Technology, Danbury, CT, USA) on polished (R a = 0.03 m, R Z =0.098 m) AISI 1020 steel substrate (22 mm 19 mm 3.2 mm) to collect splats. Diameters of the splats were measured using Image J software ( A total of ~100 splats were taken into consideration from 5-6 different SEM images. Splats were deposited at optimized plasma process parameters which showed lowest porosity in the coating [26]. Details of the optimization of plasma process parameters can be found elsewhere [26]. Substrate preheat temperature was maintained at 453 K which was same as in the optimization study for the lowest porosity coating [26]. Table I summarizes the plasma spray operating parameters for splat experiments. Carrier gas flow rate was adjusted for three different powder feedstock to maintain a constant powder feed rate of 3 g/min. Figure 2a shows the set up for plasma spraying of single splats. Temperature and velocity of the in-flight powder particle were measured using AccuraSpray in-flight diagnostic sensor (Tecnar Automation Ltée, QC, Canada). In-flight particle flow pattern 6

8 was diagnosed at 75 mm of spray distance which was same at which splats of Al 2 O 3 and Al 2 O 3 -CNT were deposited. The sensor was located at the middle of the particle flow pattern. Accuraspray diagnostic system provides ensemble average data which represents the particle characteristics in a measurement volume of approximate 75 mm 3. The minimum temperature that can be measured with the sensor is 900 C with 0.5% precision while the minimum velocity that can be measured is 5 m/s with 0.5% precision. Error in the velocity and temperature measurement is <1.5 m/s, and <15 C respectively at crosscorrelation factor > 0.9. Experimentally obtained ensemble surface temperature and velocity of in-flight particles were used for the splat simulation of A-SD, A4C-SD and A8C-SD. A shield plate with a series of 2 mm diameter holes was used to collect well dispersed splats as shown in Figure 2b. The holes in the shield plate were aligned with the particle flow pattern. The steel substrate was positioned behind the shield plate at a total spray distance of 75 mm. The substrate was preheated to 453 K using a heating gun (STEINEL, HG 2510 ESD, MN, USA) as shown in Figure 2c. The substrate temperature was continuously measured using K-type thermocouple (KMQSS-020U, Omega Engineering Inc., wire diameter: 1 mm) inserted through the substrate thickness. Figure 3a is the picture of the substrate showing hole (1 mm diameter) at the center of the substrate and through thickness. One end of the thermocouple was positioned into this hole while the other end was connected to the data logger for the substrate temperature measurement. The reading rate of the data logger was 12 measurements/minute. Figure 3b illustrates the schematic of substrate showing dimension of all faces and hole. Plasma 7

9 gun speed was 25 mm/sec, which was same as used for the optimization study for the lowest porosity coating [26]. 2.3 Microstructural Characterization FEI PHENOM (in back scattered mode) and JEOL JSM-633OF field emission (in secondary electron image mode) scanning electron microscopes were used at an operating voltage of 15 kv to investigate the powder morphology and splat morphology of A-SD, A4C-SD and A8C-SD. 2.4 Splat Simulation Splat morphology simulation has been performed using SIMDROP (Simulent Drop 3.0, Simulent Inc, Toronto, Canada) software. SIMDROP software, developed by Pasandideh-Fard et al. employs three-dimensional finite-difference algorithm which solves the Navier-Stokes equation including heat transfer and phase change [28-31]. Volume of fluid (VoF) tracking algorithm is used in this model to track the droplet-free surface [28-31]. Also, thermal contact resistance (usually 10-7 m 2 K/W for thermal spray process) at the droplet-substrate interface is included in this model [28-31]. Splat simulation was performed for A-SD, A4C-SD, and A8C-SD powders. Experimental data from the plasma spraying was used as the input for the simulation. One of the major challenges encountered in the simulation was the non-availability of materials properties (Kinematic viscosity, thermal conductivity, specific heat, surface tension of the liquid phase, thermal conductivity and specific heat of solid phase, heat of fusion etc.) especially for CNT reinforced Al 2 O 3, because high temperature thermophysical 8

10 properties of CNTs are non-existent in the literature. Thermophysical properties of CNT were approximated as of graphite [32]. Thermophysical properties for Al 2 O 3 were obtained from the literature [33, 34]. However, most of the thermophysical properties in the literature are listed at temperatures different than in-flight temperature of A-SD, A4C- SD and A8C-SD particles exiting from the plasma. SIMDROP software is capable of making the best fit of the materials property as a function of temperature from the available data [35]. Thus, it can generate the property at the desired temperature by interpolating or extrapolating the fitted curve [35]. For example, in case of one of the materials property i.e. solid phase specific heat, format of the input file is as following: N = n T = T 1, T 2, T 3, T 4..T n C = C 1, C 2, C 3, C 4..C n where, N is the total number of paired data sets, T is the array of temperatures (K), and C is the array of solid phase specific heat at corresponding temperature (in W/m-K). SIMDROP can generate best fitting curve as C=f(T). Similarly, other thermophysical properties were inserted in the above format for the splat simulation. Rule of mixtures (ROM) was used to estimate the apparent thermophysical properties of the complex Al 2 O 3 -CNT composite materials from intrinsic properties of both Al 2 O 3 and CNT and inserted as an input parameter for the splat simulation. Hence, splat simulation was treated as a single phase flow problem. Table II shows the calculated ROM values of the thermophyical properties for A-SD, A4C-SD, and A8C-SD. The accuracy of the computed ROM values has been addressed by comparing the simulated results with the experimental results in section 3.3. Table III lists droplet size, droplet surface temperature, droplet velocity, substrate temperature and substrate roughness, which were 9

11 considered as input parameters for A-SD, A4C-SD and A8C-SD splat simulation. The average powder size for each composition was assumed to be the droplet impinging diameter. The powder size distribution for all compositions ranged between μm in diameter (A-SD: 30±10 μm, A4C-SD: 26±7 μm, A8C-SD: 24±5 μm). Surface temperature and velocity of the particles were obtained experimentally using an in-flight particle diagnostic sensor, as explained in previous section 2.2. As a consequence of the splat simulation, two output files were obtained: (i) substrate temperature vs. time (i.e. solidification curve) and (ii) the volume fraction of the solid vs. time. 3. Results and Discussion 3.1 Powder Morphology Figures 4 (a), (b), and (c) show high magnification SEM micrographs of A-SD, A4C-SD and A8C-SD agglomerates, respectively. The inset shows lower magnification images of spherical agglomerates of each composition. The spray dried agglomerates have improved flowability into the powder hoses and injector, due to their spherical shape and reduced interparticle friction. From Figure 4(b) and (c), it can be seen that the CNTs were dispersed uniformly on the surface of agglomerate. Uniform dispersion of CNTs is critical in splats morphology which is discussed later. 3.2 Effect of CNT Content on Splat Formation Figure 5a-c shows SEM images of A-SD, A4C-SD and A8C-SD splats. Splat morphology changed from splashed and fragmented to almost disc-shaped with an increase in CNT content. In addition, relatively higher splat diameter and lower finger 10

12 length was observed with the increasing CNT content. Figure 6a-c shows the splat size distribution for A-SD, A4C-SD and A8C-SD splats. The average splat diameters are plotted in Figure 7a. Diameter of the fingered splat was calculated by measuring the radius of the splat from the center to the end of the finger. Only those fingered splats were considered for the measurement which showed lowest variation in the finger length. Since the frequency of the fingered splats with lowest deviation in finger length was much higher, it is safe to consider only these splats for the diameter measurement. Figure 7a shows the variation in the splat diameter as a function of CNT content. Splat diameter increases with an increasing CNT content. Similarly, Figure 7b shows that with the increasing CNT content, the average length of the fingers radiating from the periphery of the splat reduced. Splats obtained from these experiments can be categorized into four different types based on their geometry and are shown in Figure 8a. These splat categories are shown schematically in Figure 8b: (1) disc shape, (2) splats with finger, (3) fragmented splats, and (4) splats with voids. Perfect disk shape splat is formed due to longer time of the melt. The fragmented or splashed splat might be the result of localized solidification of melt. The solidified layer could obstruct the outward spreading liquid which leads to the fingers radiating out from the periphery of the splat. Splat with voids might be formed due to the entrapped gas. Experimental population density of these splat geometries was measured using image analysis from several images and has been plotted as a function of CNT content. Figure 9 show the population density for four splat geometries in A-SD, A4C-SD and A8C-SD samples. The percentage of disc shape splats increased with an increasing CNT 11

13 content. Population density of splats with finger, fragmented splats and splats with voids is the lowest for the highest (8 wt. %) CNT content. To understand the underlying mechanism, splats were observed in SEM at a higher magnification. Figures 10a-b show the single splat of A4C-SD and A8C-SD, respectively with CNTs distributed in the splat matrix. In addition, agglomeration of CNTs at the periphery of A4C-SD (Figure 11a) and A8C-SD (Figure 11b) splats is also observed. Since, splat diameter of A8C-SD was relatively larger, complete splat is shown as an inset in Figure 11b. Such a varying degree of CNT dispersion in the splat is responsible for the increased splat diameter and increased percentage of disc shaped splat. Distributed CNTs in the melt (Figure 10) leads to enhanced thermal capacity resulting in the lower viscosity. Specific heat capacity of Al 2 O 3 is 1358 Jkg -1 K -1 while the specific heat capacity for CNT is 2145 Jkg -1 K -1 at 2327 K [32]. Hence, specific heat capacity calculated using ROM (as shown in Table-II) for A4C-SD is 1404 Jkg -1 K -1 while its 1444 Jkg -1 K -1 for A8C-SD respectively at 2327 K. Specific heat capacity for A4C-SD and A8C-SD are 3.38 % and 6.33 % respectively higher than A-SD at 2327 K. The higher thermal capacity of the melt would increase the time required for the heat loss to occur for identical thermal resistance at the substrate/splat interface. Hence, localized solidification of the melt will be delayed, which results in enhanced spreading and larger splat diameter in CNT reinforced splats (figure 7). Agglomeration of CNTs (Figure 11) at the periphery of the melt will produce a counter effect resulting in an increased viscosity. Increased 12

14 viscosity of the melt will suppress fragmentation and hence, higher percentage of disc shaped splat was observed in case of CNT reinforced splats. Both of the above phenomena are schematically elucidated in Figure 12, which shows the droplet of Al 2 O 3 (Figure 12a) and Al 2 O 3 -CNT (Figure 12b) at t=0 sec i.e. just before the impact on the steel substrate. At t=t 1, higher spreading of Al 2 O 3 -CNT will take place due to higher thermal capacity of the melt as a result of CNT addition. Higher thermal capacity of the melt leads to lower viscosity which contributes towards higher spreading of CNT containing splats. At t > t 1, localized solidification occurs in Al 2 O 3 splats, that restricts their size and causes finger formation. In case of Al 2 O 3 -CNT splats, CNTs flow easily towards the periphery of the splat due to their lower specific mass (2.1 g/cc) as compared to Al 2 O 3 (3.9 g/cc) and start agglomerating at the edges (as seen in Figure 11). Agglomeration of CNTs at the edges will contribute towards increased viscosity and suppresses the splat fragmentation resulting in higher population of the disc shape splats. Thus, it is clear that CNTs play an important role in modifying the splat shape and size which ultimately affect the final properties of the coating. The experimental optimization of the splat geometry by varying in-flight temperature and velocity, CNT content and substrate conditions requires lots of intensive effort and time. Computer simulation of the splat geometry can provide valuable information about the processing conditions. Simulation of Al 2 O 3 -CNT single splat is discussed in the following section. 13

15 3.3. Simulated Splat Morphology Figures 13a, c, and e depict the simulated views of A-SD, A4C-SD and A8C-SD splats. All distances in the simulated view are in millimeter. The simulation was performed at cells per radius (CPR) value of 15. The CPR is an indicator of the resolution and can be calculated for each calculation domain from the following equation [35]. CPR = Radius of the impinging droplet No of grids Calculation domain A higher CPR number is an indicator of a higher resolution. It is recommended that CPR to be at least 10 and the optimum number of CPR is 15 [35]. Increasing the CPR number greater than 15 requires lots of extra computational time and resources [35]. Figures 13b, d, f are SEM images of the experimentally obtained splats of A-SD, A4C-SD and A8C- SD at the same parameters as simulation. Relatively higher splashing with broken fingers was observed in the simulated view of A-SD (Figure 13a) which is also seen in the experimental A-SD splat (Figure 13b). Lowest splashing of the droplet was observed in both simulated and experimental results of A8C-SD splat (Figures 13c and f). Hence, splats geometries obtained from the experiment and simulation showed good qualitative agreement. It was observed experimentally that splat diameter increases with an increasing CNT content. In order to compare the experimental results with the simulated results, spreading ratio (D/D 0 ) for experimental and simulated splats were plotted. D is the final diameter of splat and D 0 is the initial diameter of the droplet considered here as the average diameter of the size distribution. Figure 14 shows the spreading ratio as a 14

16 function of CNT content for experimental and simulated splats. Both, experimental and simulation results show the similar increasing trend in spreading ratio (D/D 0 ) with an increasing CNT content. However, experimentally obtained spreading ratio is lower than computed which is attributed to the powder feedstock characteristics. Spray dried powder, which contains ~30-40% of voids [20, 24, 26] was used for synthesizing the splats. Due to the presence of those voids in powder, molten droplet size is expected to be smaller resulting in smaller splat diameter and lower spreading ratio. During simulation, molten droplet size was assumed to be of as initial powder size which leads to larger diameter of splats as compared to experimentally obtained splats. An accurate method to obtain droplet size could be to collect the plasma sprayed powder in a liquid and then analyze the particle size distribution. Mean diameter obtained by this size distribution can be put as a more accurate input parameter (initial droplet diameter) into the simulation. The simulation also provides the volume percentage of solid in A-SD, A4C-SD, and A8C-SD splats at different time intervals, as shown in Figure µs of the after impact, the volume percentage of solid in A-SD splat is 26% while it is 14% and 8% in A4C-SD and A8C-SD, respectively. This indicates a delayed solidification process in CNT reinforced splat which is attributed to enhanced thermal capacity of the melt as a result of higher specific heat of graphite [32] µs of the after impact, the volume percentage of solid in A-SD splat is 66%, while it is 58% and 53% in A4C-SD and A8C- SD splat, respectively which becomes constant. Between 1.5 and 5 µs after impact, small increment in volume fraction of solid was observed for all splats (ΔV=+ 2% for A-SD splat, ΔV=+ 1.2% for A4C-SD splat and ΔV=+ 0.91% for A8C-SD splat). These results 15

17 indicate that complete solidification may be occurring over a greater time interval, which could be order of magnitude greater. This is in accordance with the solidification of the real molten droplets in plasma spraying. Typically, molten ceramic droplet solidify at a cooling rate of ~10 6 K/s during plasma spraying [36-38], which indicates that droplet solidification time is ~ 0.5 milliseconds for the in-flight particle temperature of 2750 K. Since, cooling does not follow a linear relationship (according to Newton s law of cooling) in between 1-5 µs, the solidification time is much lower (in microseconds) in the present study. Solidification behavior of the splat can also be understood from the change in the substrate temperature obtained from simulation. Figure 16 shows the computed substrate temperature variation vs. time for A-SD, A4C-SD and A8C-SD splats. Three different zones can be observed in Figure 16. Region I shows a significant increase in the substrate temperature at 0.1 µs for A-SD, A4C-SD and A8C-SD splats. This is attributed to the heat transfer from the molten droplet to the substrate immediately after its initial impact. Higher increase (ΔT= C) in the substrate temperature was seen for CNT reinforced splats which is attributed to higher thermal capacity of CNT containing molten Al 2 O 3 melt resulting in higher heat transfer to the substrate. In region II, substrate temperature increases for A-SD splats whereas substrate temperature remains constant for A4C-SD and A8C-SD splats. Increasing substrate temperature in case of A-SD splat is due to higher rate of solidification (as shown in Figure 15) resulting in higher amount of heat energy release. Released latent heat of 16

18 solidification will be absorbed by the substrate and substrate temperature will increase. CNT reinforced melt has higher thermal capacity and remains molten for a long time. The substrate temperature in CNT containing melts has reached maximum in Region II. The further change in the temperature of CNT containing melt will occur only if it loses heat due to solidification. Since solidification rate is lower in CNT containing melt (Figure 15), hence no appreciable change in the substrate temperature was observed in region II for A4C-SD and A8C-SD splats. In region III, substrate temperature decreased for all splats (A-SD, A4C-SD and A8C-SD). However, higher rate of change (dt/dt) of temperature was observed for A4C- SD and A8C-SD splats compared to A-SD. This is attributed to lower solidification rate in CNT reinforced melt (volume of solid: 58% and 53% in A4C-SD and A8C-SD splat respectively at 0.75 µs) compared to A-SD (volume of solid: 66% in A-SD splat at 0.75 µs), as seen in Fig. 15. Lower solidification rate will lead to lower latent heat release and hence less heat is available for the substrate to absorb. While, higher solidification rate in A-SD splat leads to higher latent heat release, which will be absorbed by the substrate. Due to the additional higher heat absorbed by the substrate (i.e. substrate will loose heat slowly with time); lower rate of change of temperature (dt/dt) was observed in A-SD splat. In the case of CNT reinforced melt, since, there is less heat available for the substrate to absorb (i.e. substrate will loose heat slowly with time), higher rate of change of temperature was observed in case of CNT reinforced melt. 17

19 Though, computed substrate temperature decreases in the µs range (in region III), whereas, volume percentage of solid remains almost constant as shown in Figure 15. This can be understood from Figure 17 which shows a liquid droplet impacting on the substrate and spreading into a thin splat at two different interval of time. Once the droplet impacts on the substrate, molten liquid which is in contact with the substrate, starts solidifying. A solid layer of thickness s exists between the top and bottom surface of the splat as shown in Figure 14. As the time progresses from t 1 to t 2, the thickness of the solid layer, which has already solidified and in contact with the substrate, increases to s+s 1. With the growth of solidified layer, a temperature gradient exists between the top and bottom surface of the splat. Temperature at the bottom surface of the splat keeps on decreasing with the increase in the fraction of solid, and hence reduction in the substrate temperature is observed in region III. However, the percentage volume of solid is constant in region III in Figure 15 which is attributed to the lost material due to flattening splashing. These lost solidified particles do not contribute in the computations of volume percentage of solid. Salimijazi et al. [39] studied the solidification behavior and splat morphology of vacuum plasma sprayed Ti alloy by computational modeling using SIMDROP software and by the experimental results and mentioned that approximately 30% of the droplet material is lost due to flattening splashing. The splashing mechanism of the droplet at the time of impact and during its flattening has been well explained by Cedelle et al. [40]. It is concluded that CNTs play an important role in tailoring the morphology of splat, which ultimately affect the coating s final properties. CNT reinforcement resulted 18

20 in relatively lower splashing and increased population density of disc shape, which will contribute towards obtaining the denser coating due to effective packing. Further, splat simulation is an effective tool for simulating the plasma sprayed single splat morphology. This can contribute significantly towards optimization of splat morphology and can reduce the large number of experiments which involves lots of time and money. Conclusions Plasma sprayed CNT reinforced (0, 4 and 8 wt. %) Al 2 O 3 single splats were obtained on the polished steel substrate at the preheat temperature of 453 K. Results strongly indicates that CNTs play a critical role in the splat morphology. Splat diameter increased from 28.6±1.4 µm to 43.2±1.3 µm with the reinforcement of 8 wt. % of CNT. Increased splat diameter was due to enhanced temperature of the melt as a result of higher specific heat of Al 2 O 3-8 wt. % CNT (1444 Jkg -1 K -1 ) compared to Al 2 O 3 (1358 Jkg - 1 K-1). Agglomeration of CNTs at the periphery of the melt produced a counter effect resulting in an increased viscosity. Increased viscosity of the melt suppresses fragmentation and hence, higher percentage of disc shaped splat was observed in case of CNT reinforced splats. Simulation of single splat showed good match in splat morphology with the experimentally obtained splat. Volume percentage of the solid and substrate temperature during splat formation elucidates the solidification behavior of CNT reinforced splats. 19

21 Acknowledgements Authors acknowledge the financial support received from Office of Naval Research (N ). AKK also acknowledge the support from the Dissertation Year Fellowship awarded by University Graduate School at FIU. The support from Advanced Materials Engineering Research Institute (AMERI), FIU for facilitating the characterization facilities is greatly appreciated. References [1] P. Fauchais, A. Vardelle, B. Dussoubs, J. Therm. Spray Technol. 10 (2001) [2] X. Jiang, J. Matejicek, S. Sampath, Mater. Sci. Eng. A 272 (1999) [3] S. Sampath, X. Y. Jiang, J. Matejicek, A. C. Leger, A. Vardelle, Mater. Sci. Eng. A 272 (1999) [4] S. Sampath, X. Jiang, Mater. Sci. Eng. A (2001) [5] E. E. Balic, M. Hadad, P. P. Bandyopadhyay, J. Michler, Acta Mater. 57 (2009) [6] T. Chraska, A. H. King, Surf. Coat. Technol. 157 (2002) [7] R. Dhiman, A. G. McDonald, S. Chandra, Surf. Coat. Technol. 201 (2007) [8] A. Elsebaei, J. Heberlein, M. Elshaer, A. Farouk, J. Therm. Spray Technol. 19 (2009) [9] S. Fantassi, M. Vardeue, A. Vardelle, P. Fauchais, J. Therm. Spray Technol. 2 (1993) [10] K. Shinoda, T. Koseki, T. Yoshida, J. Appl. Phys. 100 (2006) [11] S. Amada, K. Imagawa, S. Aoki, Surf. Coat. Technol. 154 (2002) [12] R. S. Lima, B. R. Marple, J. Therm. Spray Technol. 16 (2007)

22 [13] L. Bianchi, A. C. Leger, M. Vardelle, A. Vardelle, P. Fauchais, Thin Sol. Films 305 (1997) [14] M. Fukumoto, Y. Huang, M. Ohwatari, C. Coddet, Thermal Spray, Meeting the Challenge of the 21st century, ASM International, Metals Park, OH 401 (1998) [15] M. Fukumoto, S. Katoh, I. Okane, in: A. Ohmori (Ed.), Proceeding of the Integrated Thermal Spray Conference, Kobe Japan, (1995) [16] C.J. Li, J.L. Li, W.B. Wang. In: C. Coddet, Editor, Thermal Spray, Meeting the Challenges of the 21st Century,, ASM International, Metals Park, OH (1998), p [17] K. Balani, A. Agarwal, Surf. Coat. Technol. 202 (2008) [18] K. Balani, A. Agarwal, Nanotechnology 19 (2008) [19] K. Balani, A. Agarwal, J. Appl. Phys. 104 (2008) [20] K. Balani, S. R. Bakshi, Y. Chen, T. Laha, A. Agarwal, J. Nanosci. Nanotechnol. 7 (2007) 1-10 [21] K. Balani, S. P. Harimkar, A. Keshri, Y. Chen, N. B. Dahotre, A. Agarwal, Acta Mater. 56 (2008) [22] K. Balani, T. Zhang, A. Karakoti, W. Z. Li, S. Seal, A. Agarwal, Acta Mater. 56 (2008) [23] Y. Chen, K. Balani, A. Agarwal, Appl. Phys. Lett. 92 (2008) [24] A. K. Keshri, J. Huang, V. Singh, W. Choi, S. Seal, A. Agarwal, Carbon 48 (2010) [25] A. K. Keshri, J. Huang, W. Choi, A. Agarwal, Surf. Coat. Technol. 204 (2010) [26] A. K. Keshri, R. Patel, A. Agarwal, Surf. Coat. Technol. 205 (2010) [27] S. R. Bakshi, V. Singh, S. Seal, A. Agarwal, Surf. Coat. Technol. 203 (2009) [28] M. Pasandideh-Fard, S. Chandra, J. Mostaghimi, 1st International Thermal Spray Conference, Montreal, Quebec pp [29] M. Pasandideh-Fard, S. Chandra, J. Mostaghimi, Int. J. of Heat Mass Transfer, 45 (2002)

23 [30] M. Pasandideh-Fard, V. Pershin, S. Chandra, J. Mostaghimi, J. Therm. Spray Technol., 11 (2002) [31] J. Mostaghimi, M. Pasandideh-Fard, S. Chandra, Plasma Chemistry and Plasma Processing, 22 (2002) [32] FACTSAGE Thermodynamic Equilibrium Software, version 5.0, GmbH, Aachen, Germany, 2001 [33] E. Dorre, H. Hubner, Alumina-processing, properties and applications. Berlin: Springer-Verlag; 1984: [34] Y. S. Touloukian, E. H. Buyco, Specific heat-nonmetallic solids. Thermophysical properties of matter. New York: IFI/Plenum; 1970: 5 [35] Simulent Drop 2.2, Droplet impact and solidification software, Totonto, Canada ( [36] P. Fauchais, M. Fukumoto, M. Vardelle, J. Therm. Spray Technol. 13 (2004) [37] H. B. Parizi, L. Rosenzweig, J. Mostaghimi, S. Chandra, T. Coyle, H. Salimi, L. Pershin, A. McDonald, C. Moreau, J. Therm. Spray Technol. 16 (2007) [38] J. Suffner, H. Hahn, S. Dosta, I. G. Cano, J. M. Guilemany, Surf. Coat. Technolo. 204 (2009) [39] H. R. Salimijazi, M. Raessi, J. Mostaghimi, T. W. Coyle, Surf. Coat. Technolo. 201 (2007) [40] J. Cedelle, M. Vardelle, B. Pateyron, P. Fauchais, Thermal spray 2004: Advances in technology and application: Proceedings of the international thermal spray conference (2004)

24 List of Tables Current (A) Table I: Plasma spray operating parameters for splat formation Voltage (V) Plasma process parameters for synthesizing single splat Primary Gas, Argon (slm) Secondary Gas, Helium (slm) Stand-off (mm) from the substrate Powder feed rate (g/min) Substrate preheat temperature (K)

25 Table II: Thermo-physical properties of A-SD, A4C-SD and A8C-SD S. No. Property Specific mass (kg/m 3 ) Liquid phase kinematic viscosity (m 2 /s) Liquid phase thermal conductivity (W/m-K) Liquid Phase Specific Heat (J/kg-K) Liquid Phase Surface Tension (kg/s 2 ) Solid Phase Thermal Conductivity (W/m-K) Al 2 O 3 (A-SD) Al 2 O 3-4 wt. %CNT (A4C-SD) Solid Phase Specific Heat (J/kg-K) Al 2 O 3-8 wt. %CNT (A8C-SD) x10-5 at 2473 K 7.86 at 2327 K 1358 at 2327 K 0.69 at 2327 K at 298 K 5.90 at 1312 K 772 at 298 K 1273 at 1312 K 1358 at 2327 K 1.098x10-5 at 2473 K at 2327 K 1404 at 2327 K 0.69 at 2327 K at 298 K at 1312 K at 298 K 1320 at 1312 K 1404 at 2327 K x10-5 at 2473 K 377 at 2327 K 1444 at 2327 K 0.69 at 2327 K at 298 K at 1312 K at 298 K at 1312 K 1444 at 2327 K Melting Point (K) Heat of Fusion (J/kg) x x x

26 Table III: List of molten droplet and substrate parameters used for simulation of A-SD, A4C-SD and A8C-SD splats A-SD A4C-SD A8C-SD Initial particle diameter (µm) Particle surface temperature (K) Particle velocity (m/s) Initial substrate temperature (K) Substrate surface roughness( µm)

27 List of Figure Captions: Figure 1: Particle size distribution of (a) Al 2 O 3 (A-SD) spherical agglomerates (b) Al 2 O 3-4 wt. % CNT (A4C-SD) spherical agglomerates (c) Al 2 O 3-8 wt. % CNT (A8C- SD) spherical agglomerates. Figure 2: (a) Plasma spray set-up for collecting single splats (b) Shield plate in front of the plasma gun having a series of holes of 2 mm diameter (c) Heat gun used to preheat the steel substrate (22 mm 19 mm 3.2 mm) which is positioned on the back of the shield plate. Figure 3: (a) Digital picture of the steel substrate showing through thickness hole (Ø=1 mm) for inserting thermocouple (b) schematic of steel substrate showing dimensions of all faces and hole. Figure 4: High magnification SEM images of spray dried (a) Al 2 O 3 (A-SD) powder particle (b) Al 2 O 3-4 wt.% CNT (A4C-SD) particle showing homogeneous dispersion of CNTs on the surface (c) Al 2 O 3-8 wt.% CNT (A8C-SD) particle showing homogeneous dispersion of CNTs on surface. Inset shows low magnification image of spray dried agglomerate of each composition. Figure 5: SEM images of plasma sprayed single splat on polished steel substrate for (a) A-SD (b) A4C-SD (c) A8C-SD. Substrate was preheated to 453 K. Figure 6: Splat size distribution of (a) A-SD (b) A4C-SD (c) A8C-SD. Total of ~100 splats were considered from 5-6 different SEM images to plot the splat size distribution. Figure 7: (a) Variation in average splat diameter with function of CNT content. (b) Variation in finger length with function of CNT content. Figure 8: (a) Experimentally collected splats showing different morphology (b) Schematic showing different morphology of splats Figure 9: Population density of different type of splats as function of CNT content. Figure 10: SEM image showing distributed CNTs in (a) A4C-SD (b) A8C-SD splats. Figure 11: SEM image showing distributed CNTs in the matrix and agglomerated CNTsat the periphery of (a) A4C-SD (b) A8C-SD splats. Figure 12: Schematic of splat formation in the case of Al 2 O 2 and Al 2 O 3 -CNT splat. 26

28 Figure 13: (a) Simulated top view of A-SD splat (b) SEM image of experimentally obtained A-SD splat (c) Simulated top view of A4C-SD splat (d) SEM image of experimentally obtained A4C-SD splat (e) Simulated top view of A8C-SD splat (f) SEM image of experimentally obtained A8C-SD splat. Figure 14: Spreading ratio with varying CNT content for experimental and simulated splats. Figure 15: Volume percentage of solid as a function of time for A-SD, A4C-SD and A8C-SD splats. Figure 16: Substrate temperature as a function of time for A-SD, A4C-SD and A8C-SD splat. Figure 17: Schematic of the liquid droplet landing on the substrate and spreading into a thin splat at two different interval of time. 27

29 Figures (a) No. of Particles 20 Al 2 O 3 (A-SD) Mean : m 15 SD : 7.72 m Particles Size ( m) (c) No. of Particles Al 14 2 O 3-8wt% CNT (A8C-SD) Mean : m 12 SD : 7.3 m Particles Size ( m) Figure 1: Particle size distribution of (a) Al 2 O 3 (A-SD) spherical agglomerates (b) Al 2 O 3-4 wt. % CNT (A4C-SD) spherical agglomerates (c) Al 2 O 3-8 wt. % CNT (A8C-SD) (b) No. of Particles Al 2 O 3-4wt. % CNT (A4C-SD) Mean : m SD : 6.92 m Particle Size ( m) spherical agglomerates. A total of ~100 particles were considered from 5-6 different SEMimages.. 28

30 (a) (b) (c) Steel substrate Shield plate 25 mm 25 mm Shield plate Heat gun 25 mm Figure 2: (a) Plasma spray set-up for collecting single splats (b) Shield plate in front of the plasma gun having a series of holes of 2 mm diameter (c) Heat gun used to preheat the steel substrate (22 mm 19 mm 3.2 mm) which is positioned on the back of the shield plate. 29

31 1.0 mm 19 mm 3.2 mm 22 mm (a) (b) Figure 3: (a) Steel substrates showing through thickness hole (Ø=1 mm) for inserting thermocouple and (b) schematic of the steel substrate showing dimensions of all faces and hole. 30

32 (a) (b) Figure 4: High magnification SEM images of spray dried (a) Al 2 O 3 (A-SD) powder particle (b) Al 2 O 3-4 wt.% CNT (A4C-SD) particle showing homogeneous dispersion of (c) CNTs on the surface (c) Al 2 O 3-8 wt.% CNT (A8C-SD) particle showing homogeneous dispersion of CNTs on surface. Inset shows low magnification image of spray dried agglomerate of each composition. 31

33 Figure 5: SEM images of plasma sprayed single splat on polished steel substrate for (a) A-SD (b) A4C-SD (c) A8C-SD. Substrate was preheated to 453 K. 32

34 (a) No. of Splats A-SD Splats 14 Mean Diameter : 28.6 m 12 SD : 1.43 m Splats Diameter ( m) (c) No. of Splats (b) No. of Splats A4C-SD Splats Mean Diameter : 34.7 m SD : 1.48 m Splats Diameter ( m) A8C-SD Splats 16 Mean Diameter : 43.3 m 14 SD : 1.31 m Splats Diameter ( m) Figure 6: Splat size distribution of (a) A-SD (b) A4C-SD (c) A8C-SD. Total of ~100 splats were considered from 5-6 different SEM images to plot the splat size distribution. 33

35 Splat diameter ( m) 42 (a) CNT content (wt.%) CNT content (wt%) Figure 7: (a) Variation in average splat diameter (based upon ~100 splats) with function of CNT content. (b) Variation in the finger length with function of CNT content. Error bars corresponds to standard deviations associated to average value. Finger length ( m) (b) 34

36 (a) (b) Disc splat Fingered splat Fragmented splat Figure 8: (a) Experimentally collected splats showing different morphology (b) Schematic showing different morphology of splats Void (1) (2) (3) (4) Splats with voids Splat with voids 35

37 Population (%) Perfect disc splats with finger Fragmented splats having voids CNT content (wt.%) Figure 9: Population density (based upon total ~100 splats) of different type of splats as function of CNT content 36

38 (a) A4C-SD (b) A8C-SD CNT CNT (a) Figure 10: SEM image showing distributed CNTs in (a) A4C-SD (b) A8C-SD splats. 37

39 (a) A4C-SD Agglomeration of CNTs Distributed CNTs Figure 11: SEM image showing distributed CNTs in the matrix and agglomerated CNTs at the periphery of (a) A4C-SD (b) A8C-SD splats. A8C-SD Distributed CNTs Agglomeration of CNTs (b) 38

40 Figure 12: Schematic of splat formation in the case of (a) Al 2 O 2 and (b) Al 2 O 3 -CNT splat. (a) (b) 39

41 Splashing Broken finger (a) Splashing (b) Finger Splashing (c) (e) Finger Splashing Broken Finger Splashing (d) Finger (f) Figure 13: (a) Simulated top view of A-SD splat (b) SEM image of experimentally obtained A-SD splat (c) Simulated top view of A4C-SD splat (d) SEM image of experimentally obtained A4C-SD splat (e) Simulated top view of A8C-SD splat (f) SEM image of experimentally obtained A8C-SD splat. 40

42 Spreading ratio (D/D 0 ) Simulation results Experimental results CNT Content (wt. %) Figure 14: Spreading ratio (ratio of final splat diameter to initial droplet diameter) with varying CNT content for experimental and simulated splats. 41

43 Volume of Solid (%) T= 0.75 µs 10 HD-A-SD-180C A4C-SD A8C-SD HD-A4C-SD-180C HD-A8C-SD-180C Time ( s) Figure 15: Volume percentage of solid as a function of time for A-SD, A4C-SD and A8C-SD splats. 42

44 Figure 16: Computed substrate temperature as a function of time for A-SD, A4C-SD and A8C-SD splat. 43

45 Figure 17: Schematic of the liquid droplet landing on the substrate and spreading into a thin splat at two different interval of time. 44

46 Research Highlights The effect of carbon nanotube (CNT) addition on the splat formation in plasma sprayed aluminum oxide (Al 2 O 3 ) composite coating using experimental and computational methods is studied in this article. The addition of CNTs makes splat morphology more disk-shaped with an increasing diameter and lower splashing. The addition of CNTs resulted in two simultaneously competing phenomena viz. increased heat content and increased viscosity of the melt, which were responsible for higher splat diameter and increased population density of disc shaped splats respectively. Splat morphology was also simulated and showed a good agreement with the experimentally obtained splats. 45