Sei sulla pagina 1di 10

Acta Materialia 53 (2005) 29092918 www.actamat-journals.

com

A study of mechanical alloying processes using reactive milling and discrete element modeling
Trent S. Ward, Wenliang Chen, Mirko Schoenitz, Rajesh N. Dave, Edward L. Dreizin
Department of Mechanical Engineering, New Jersey Institute of Technology, Newark, NJ 07102-1972, United States Received 18 November 2004; received in revised form 4 February 2005; accepted 4 March 2005 Available online 9 April 2005

Abstract The milling progress of mechanical alloying in SPEX shaker mills was investigated using dierent ball sizes and ball to powder mass ratios (charge ratios). Reactive materials were used, for which an exothermic reaction is mechanically triggered after a certain period of milling. The milling progress was determined experimentally from the temperature trace of the milling vial exhibiting a peak at the time of such a reaction. An expression for the milling dose was introduced to describe the eect of dierent milling parameters on the milling progress. In the rst approximation, the milling dose leading to the mechanically triggered reaction remained constant over a range of charge ratios and could, therefore, be used to gauge the milling progress. The milling progress was described theoretically using the discrete element method. The theoretical milling dose that correlates well with its experimental analog was found to depend on the head-on impact energy dissipation rate. The presented approach is suitable for scale-up and optimization of mechanical alloying of various materials. 2005 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Mechanical alloying; Milling dose; Milling progress; Discrete element method

1. Introduction Mechanical alloying is a rapidly developing technology capable of producing a wide range of dispersion strengthened, energetic, nanocrystalline, and other advanced alloys [1,2]. The materials processing is very simple and appears to be readily scalable: powders of new materials are produced by high-energy ball milling of starting ingredients. Many groups have reported significant progress in the preparation of advanced materials [3,4]. The critical role of the milling intensity in the formation of a specic structure has been addressed in [57]. It has been established that specic intensity thresholds exist for the formation of amorphous alloys, solid solutions, or intermetallic phases. Thus, a signicant degree of control over the structure of mechanical alloys
*

Corresponding author. Tel.: +1 973 596 5751; fax: +1 973 642 4282. E-mail address: dreizin@njit.edu (E.L. Dreizin).

produced in laboratory conditions has been achieved; however few have succeeded in production of useful mechanical alloys on a scale sucient for practical applications. The main challenge has been in the ability of evaluating and predicting the progress of mechanical alloying as a material processing technique. Even if the milling intensity is maintained within the range desired for synthesis of a specic material, the progress of mechanical alloying is aected by a large number of process parameters. To enable practical applications of mechanical alloying, it is necessary to predict how the milling progress is aected by such parameters as the amount of material being processed or type of the ball mill used. A model is needed to describe the progress of mechanical alloying quantitatively. Because of the large number of process parameters, it is reasonable to expect that a successful process description can be obtained using a numerical technique. A number of mathematical descriptions of mechanical

1359-6454/$30.00 2005 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.actamat.2005.03.006

2910

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918

alloying have been proposed in the literature [814]. However, the comparisons of the model predictions with the experimental data are dicult. The experimental results are expressed as changes in composition, particle size, morphology, structure, or crystallinity while models typically operate with such global parameters as number and energy of collisions between grinding media or local parameters such as coalescence and fracture probability, and size distribution. Therefore, even laborious experimental eorts on recovery of multiple intermediate milling products and the characterization of their properties needed to assess the mechanical alloying progress still do not enable straightforward comparisons with the computational results. An alternative approach to assess the milling progress has been initially suggested by Schaer and McCormick [15,16] and will be further developed in this paper. This approach makes use of a sharp temperature increase known to occur during milling of highly reactive components [17]. The temperature increase marks a mechanically activated reaction that occurs when a specic degree of structural renement is achieved. The structural renement can be dened as the reduction of the phase domains and the grain size in the microstructure of composite powder particles. For many materials, the structural renement is only possible when the milling intensity is above some threshold required to fracture and cold weld the elemental phases. This restriction on the milling intensity is similar to that necessary to produce specic types of mechanical alloy structures [6,7,18]. The following manuscript addresses the approach necessary to access the material processing progress for cases when the milling intensity is within the desired ranges, e.g., for achieving the structural renement required for mechanical activation of an exothermic reaction (or for the production of a mechanical alloy with a specic structure). It has been suggested that the mechanically activated reaction proceeds spontaneously if the adiabatic reaction temperature of the components exceeds 1800 K [17,19]. The processing time required to trigger the reaction can be detected by observing a milling vial temperature exhibiting a peak that can be used as a clear-cut indicator of the milling progress without any further processing of the product samples. Thus, mechanical alloying of highly reactive components, or reactive milling, will be exploited to develop an experimental data set characterizing the progress in the materials processing. Parallel to the experimental determination of milling progress, a numerical model will be developed and the computational results will be analyzed to establish a correlation with the experimental data. The main objective of this study is to develop a methodology enabling direct and ecient comparisons of the experimental and computational descriptions of mechanical alloying progress. In the future, such a methodology can be used to opti-

mize and scale up production of advanced mechanical alloys for a wide range of practical applications.

2. Experimental The high-energy ball mill used in the experimental study was a SPEX 8000 series shaker mill. The SPEX mill is a vibratory mill; its vial is agitated at high frequency in a complex cycle that involves motion in three orthogonal directions [20]. The SPEX mill provides a comparably high milling intensity, achieving higher strain rates and therefore more rapid grain size reduction than other types of ball mills [5]. The reciprocating velocity of the vial in the SPEX mill is directly proportional to the motor rotational speed. Under various loading conditions, the rotational speed of the actuator input shaft was measured with a stroboscope. The nominal rotational speed was 1054 rpm, yielding an oscillation frequency of 17.6 Hz. A hardened steel vial was used with hardened steel balls. No process control agent was used in this work and the materials were milled under argon. The measured diameters of the balls used were 2.36, 3.16, 4.76, and 9.52 mm (nominally 3/32, 1/8, 3/16, and 3/8 in., respectively). The ball-to-powder mass ratio (charge ratio, CR) was set to vary between 2.5, 5, and 10. Thermite mixtures known to have extremely high adiabatic reaction temperatures were used in these experiments. Specically, AlFe2O3 and AlMoO3 thermites with adiabatic reaction temperatures of 3135 and 3253 K, respectively, were used [21]. Starting materials were Al (98%, 1014 lm), Fe2O3 (99.5%, 325 mesh), and MoO3 (99.95%, 325 mesh) from Alfa Aesar. The total amount of material was 2 and 5 g in the case of AlFe2O3. The amount of the AlMoO3 mixture loaded in a single run did not exceed 2 g to avoid damaging the milling vial because of the high local temperatures during the reaction. A negative temperature coecient thermistor was mounted on the milling vial and connected to a personal computer-based data logger to monitor the temperature. The temperature acquisition rate varied depending on the milling time scale, typically the temperature was logged every 10th of a second to half a minute. The thermistor circuit was calibrated between two temperatures corresponding to 0 and 100 C. However, the exact temperature is not signicant; rather the time of initiation is more important and was always accurate within 1% for any sampling rate. Examples of the measured milling temperatures are shown in Fig. 1. The process consists of the activation period with a gradual temperature increase and stabilized temperature region. During this period, ne mixing and particle size reduction occurs. Ignition is mechanically triggered at a critical state by a ballball or ballvial

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918


80
0.9

2911

Milling Vial Temperature, C

CR =5
Coefficient of Restitution
60

Coated Balls

CR =10

CR =2.5

0.8

Uncoated Balls Huang et al. [22]

0.7

40

0.6

CR=10 CR=5

0.5

20

CR=2.5

0.4

10

20

30

40
0.3

Milling Time, min

Fig. 1. Typical temperature measurement prole for determining ignition time, shown as a function of the charge ratio. Proles taken from an AlFe2O3 trial with a ball diameter of 4.76 mm.

10

28

46

Ball Diameter, mm

collision. The reaction occurs with a large heat release where heat is transferred to the vial and balls, producing an abrupt increase in the vial temperature. After the reaction, the vial temperature slowly decreases as a function of the ambient conditions. In Fig. 1, each plot represents a dierent charge ratio, which changes as a function of the number of balls because the powder mass and ball diameter are constant. As the charge ratio increases between 2.5, 5, and 10, the initiation time decreases to 38.6, 24.1, and 9.3 min, respectively. Hence, a shorter milling time is required to obtain the same degree of renement at which the reaction is triggered. The processes of mechanical alloying are greatly inuenced by the impact characteristics of the grinding media such as friction and restitution coecient. The impact velocity, thickness and strength of the powder layer coating the balls, and the ball size contribute to the variation in restitution coecient [11,22]. The thin powder coating on the balls aects the collision by decreasing the elasticity of impact. To support the description of grinding media collisions in the numerical model described below, the coecient of restitution between the ball and vial surface was evaluated experimentally. The restitution coecient was determined for both uncoated and coated balls with the vial. Clean hardened steel balls and the same balls coated with powder (removed from an interrupted AlFe2O3 milling trial after 5 min) were used. In experiments, the balls were free falling onto a vial lid that was clamped to a heavy metal plate. The balls were dropped from a height of about 80 cm, which resulted in experimentally measured impact velocities ranging between 34 m/s. A high-speed video camera measuring 500 frames per second was used to determine the incident and rebound velocities in the impact zone. Sequences of four to ve frames immediately before and after the impact were used to obtain the respective average velocity values. The ratio of the rebound to the incident velocities produced the experimental coecient of restitution. The results of these

Fig. 2. Coecient of restitution results from impact experiments and literature. Lines represent numerical relationships used in the DEM simulation for the three charge ratios of 2.5, 5, and 10.

experiments are shown in Fig. 2 and yield an average restitution coecient of 0.7 for coated balls. The relative error among the trials varied between 1% and 8% except for the 3.16 mm coated balls where the error was 20%. This poor reproducibility is likely due to a less reproducible coating thickness, which reects the condition in the milling vial. The collisions involving the coated balls were consistently less elastic compared to the uncoated balls. Furthermore, the restitution coecient was less inuenced by the ball diameter for the coated trials. In similar investigations with larger balls and lower velocities, the restitution coecient was found to vary in the range of 0.380.8 depending on the powder thickness [22]. Huang et al. [22] systematically studied the effects of ball diameter, impact velocity, and powder thickness on the coecient of restitution for plate impacts. At a coating thickness of 0.52 mm and impact velocity of 2 m/s, the results from [22] are also shown in Fig. 2 and demonstrate a smaller average coecient of restitution of 0.385 for larger ball diameters. The lines in Fig. 2 show the trends used to describe the restitution coecient in the numerical simulations, as discussed in more detail below.

3. Numerical modeling Discrete element method (DEM) modeling was employed to study the milling progress of high-energy ball milling. In this numerical scheme, the motions of individual balls are traced and the interactions of the balls are monitored contact by contact. The term discrete element, in an approach originally proposed by Cundall [23], refers to the fact that the simulation models the particles as well as the device components as a system of individual or discrete elements. DEM has previously been implemented successfully to predict ball

2912

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918

paths due to the ball and boundary interactions of highenergy ball milling in horizontal, vibratory, and planetary mills [2426]. In this work, a partially latching spring model developed by Walton and Braun [27] for elasticplastic collisions was implemented in the DEM for all ball collisions. In this model, the loading resistance force is a linear spring, with the spring constant given by K1 K 1 0:01pEri ; 1 where E is Youngs modulus and ri is the ball radius. To ensure a suciently large normal stiness K1, the maximum ball deformation is limited to 1%. A stier linear spring with spring constant, K2, is used during the unloading (restoration) stage to account for a nite plastic deformation. K2 can be obtained from the coecient of restitution, e. It is assumed to be independent of the impact velocity for the elasticplastic case considered here, and is dened by r K1 : 2 e K2 The coecient of restitution is assumed to vary between 0.5 and 0.8 as a function of the thickness of the powder layer expected to coat the ball surfaces. This thickness is proportional to the ratio of total surface area of the balls to the powder mass. The coecient of restitution variation used in the model is illustrated in Fig. 2 for the charge ratios of 2.5, 5, and 10. The appropriate choice of the restitution coecient accounts for the presence of the milled powder, thus bypassing the problem of computationally expensive modeling of individual powder particles (see [28]). The tangential force is modeled by an incrementally slipping friction model, also proposed by Walton and Braun [27], where the eective tangential stiness is a function of the coecient of friction. Accordingly, this realistic soft sphere interaction model [29,30] is applied to the ballball and ballboundary interactions. A detailed description of the interaction-force model and the numerical implementation can be found in [27,29,30]. The numerical integration scheme employed in this model requires a suciently small time-step for contact force modeling, which is related to the normal stiness, K1, by the equation [23,31]: p pe mi =2K 1 ; 3 Dt n where mi is the mass of one ball and n is the desired number of time steps for one contact. Typically, the value of n is selected from 40 to 50 to ensure the accuracy of trajectories and energy losses [27]. A larger stiness results in a smaller time step and longer overall computation times. Inadequately small stiness results in a time step with unstable simulation results. It has been observed that under stable conditions, the value of the stiness does not

signicantly change the results, as long as signicant ball deformation does not occur during collisions. At the beginning of the simulation, balls are randomly positioned inside the milling vial and assigned small random velocities; however, the net momentum of the system is initially zero. For each time step, forces between balls are calculated for all contacting balls using the interaction force model. The new translational and rotational accelerations of the balls are calculated by Newtons equation of motion. The new velocity and position of the balls are obtained by explicit integration of Newtons equation via the time-centered, nite-dierence method [32]. After obtaining new positions and velocities for all balls, the program repeats the cycle of updating contact forces and ball positions. Checks are incorporated to nd new contacts and delete broken contacts for the neighbor list of each ball. During the simulation, average properties of the system are calculated and are written to a le at regular intervals to obtain quantities of interest. In order to theoretically study the milling progress, the energy transferred to the powder must be numerically predicted. This energy lost in all collisions during the time interval DtE due to the powder being deformed is described in terms of energy dissipation rate, Ed, by the following equation: Ed
Nc Nc X DEk X E 1 E 2 k ; Dt E DtE k 1 k 1

where E1, and E2 describe the energies of a binary impacting system before and after collision, respectively; k is the collision index, and Nc is the total number of collisions for the time interval DtE. A time interval of 0.2 s was selected to compute the energy dissipation rate. In a computational experiment, the energy dissipation rate was continuously calculated every 0.2 s until its temporal changes became insignicant. Thus, the simulation time is judged by the convergence of the energy dissipation rate and is typically between 4 and 8 s. Fig. 3 shows

3.5

Energy Dissipation Rate E d, W

Ball Diameter
2.36 mm 9.52 mm

3.0 2.5 2.0 1.5 1.0 0.5 0

Stable

Time, s

Fig. 3. Temporal changes in computed energy dissipation rate, Ed.

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918

2913

the energy dissipation rate versus time for ball diameters of 2.36 and 9.52 mm. For both diameters, Ed spikes at the beginning of the calculation and after approximately 0.51 s it begins to stabilize; it uctuates around its average value for the remainder of the simulation. The uctuation in Ed increases when the number of balls is reduced. To analyze the results, the Ed parameter assigned to each run is computed to be the average of all points in the stable region. In addition to quantifying the total rate of energy dissipation, the rates of energy dissipation due to dierent types of collisions were tracked. Specically, collisions were classied as a function of the impact angle. The complete set of input parameters listed in Table 1 corresponds to empirical measurables including: vial size, mill rotational speed, vibration frequency and amplitude, ball diameter, total mass of milling balls, Youngs modulus for steel, friction coecient, and restitution coecient. The DEM output results include the energy dissipation rate, impact frequency, impact velocity, and impact angle. The model was run for the same parametric variations used in the experimental study. Specically, the ball size and charge ratio (ball load) were systematically varied and the restitution coecient was adjusted accordingly as discussed above. Because the simulation results are described in terms of energy, the following assumption was used to compare the computational and experimental results. The milling time is proportional to the inverse of the energy dissipation rate the greater the energy dissipation rate, the shorter the
Table 1 Milling parameters used in the discrete element method Size of vial (mm) Angle of vial axis () Rotation speed (rpm) Vibration frequency (Hz) Vibration amplitudes, Ax, Ay (mm) Diameter of milling ball (mm) Total mass of milling balls (g) Youngs modulus of steel (GPa) Friction coecient Restitution coecient 38 57 15 1054 17.6 25, 6 2.36, 3.16, 4.76, 9.52 5, 10, 12.5, 20, 25, 50 200 0.4 0.50.8

milling time for the same degree of structural renement of the milled powder.

4. Results Experimentally observed times of spontaneous initiation during milling of AlMoO3 and AlFe2O3 are shown in Fig. 4. The values and errors shown are the results of 24 repetitions under identical conditions. The reproducibility is found to be on the order of 10%. This error is likely due to the fact that at-ended vials were used, hence powder trapped in the corners is milled ineectively, not participating in the reaction. This eect becomes more severe with increasing ball size. Milling times generally decrease with increasing charge ratio CR, as expected. For most cases, the larger ball diameter corresponds to shorter milling times, which could be understood qualitatively considering that larger balls result in somewhat increased milling intensities. At the same time, the trend is reversed for the largest, 9.52 mm, ball diameter, for the 2 g samples, when only one ball was used to produce the desired charge ratio. Clearly, using a single ball dramatically changes the pattern of the ball motion in the mill. Similarly, collective ball motions could be expected for the very large number of balls corresponding to the small diameter balls used for the 5 g sample. Because of the combined eects of milling intensity and dierent ball motion patterns, the interpretation of the observed trends for the milling time as a function of the ball diameter is not straightforward. The DEM model produced several descriptive characteristics of the ball milling process and the collision statistics. The collision frequency was determined as a function of the number of balls and is shown in Fig. 5. It is shown for the varied total ball mass and ball diameter, and a xed vibration frequency. The number of collisions ncoll is proportional to the number of balls nb: ncoll $ nb : 5 The trend described by Eq. (5) is generally maintained for dierent ball diameters. The total energy dissipation

Al + Fe2O3, 5 g

Al + Fe2O3, 2 g

Al + MoO3, 2 g single ball

Milling Time, min

100 70 40

10 7 4 2 3 4 5 6 7 8 9 3 4 5 6 7 8 9 3 CR 4 5
2.5

7
5

9 10
10

Ball Diameter, mm

Fig. 4. Experimental data on reactive milling of AlMoO3 and AlFe2O3 compositions presented as the milling time plotted versus ball diameter.

2914
Ball Diameter 2.36 mm

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918


0.8 0.7

107

Collision Frequency, 1/s

4.76 mm 9.52 mm

1/Ed~t, 1/W

10

3.16 mm

0.6 0.5 0.4 0.3

CR

2.5

10

105

10

0.2 0.1

103

10

Ball Diameter, mm
102

100

101

102

103

Fig. 7. Eect of charge ratio on the inverse of the total energy dissipation rate for a 2 g sample mass.

Number of Balls
Fig. 5. Collision frequency as function of number of balls, given for all ball diameters.

rate Ed was computed as a function of the ball mass and is shown in Fig. 6. As noted above, this energy dissipation rate is the average of the values for each 0.2 s iteration after the simulation results are stable. The trend in Fig. 6 illustrates that Ed is independent of ball diameter and is proportional to ball mass mb: E d $ mb : 6 The trend expressed by Eq. (6) can be qualitatively understood for the milling process, in which the general pattern of the ball motion does not change signicantly as a function of the ball size. In addition to the energy dissipation rate computed for each ball mass, the Ed was examined as a function of charge ratio. The inverse energy dissipation rate 1/Ed is considered as the numerical analog to the experimental milling time t. The DEM results for a 2 g sample mass corresponding to the experiments presented in

Ball Diameter 2.36 mm

Energy Dissipation Rate E d, W

10 8 6

3.16 mm 4.76 mm 9.52 mm

Fig. 4 are shown in Fig. 7. It is apparent that t is justiably proportional to 1/Ed, since the empirical and numerical curves demonstrate lower milling times as the charge ratio increases. At the same time, the specic shapes of the calculated curves showing change of inverse of the total energy dissipation rate as a function of ball diameter cannot be unambiguously correlated with the experimental curves of milling time as a function of the ball diameter, as shown in Fig. 4. Therefore, further analysis is needed, e.g., considering whether the collisions with dierent impact angles have dierent inuence on the achieved milling progress. By tracking the impact angle during the simulation, the collision energies have been categorized as either head-on: impact angles less than 30, or glancing: impact angles greater than 30. The energy dissipation rate was then accordingly computed for both head-on and glancing impacts. The sum of the head-on and glancing energy dissipation rates is the total energy dissipation rate. Fig. 8 illustrates the inverse Ed versus the ball diameter for the total Ed, head-on Ed, and glancing Ed for both 2 and 5 g sample masses. It appears that the calculated curves for the inverse Ed for the head-on collisions approach the experimental curves of milling time as a function of the ball diameter much better than similar curves for the 1/Ed for the glancing collisions.
1.2
Head-on Ed

5g

2g

1.0

Glancing Ed Total Ed

1/Ed ~t, 1/W


6 8 10 28 46

0.8 0.6 0.4 0.2

0.0 3 4 5 6 7 8 9 3 4 5 6 7 8 9

Ball Mass, g

Ball Diameter, mm

Fig. 6. Energy dissipation rate as function of ball mass, given for all ball diameters.

Fig. 8. Inverse of the energy dissipation rate for total, head-on, and glancing collisions with CR = 5.

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918

2915

The head-on collisions account for approximately 46% and 34% of the total collisions for the 2 and 5 g sample, respectively. This 12% dierence indicates that a greater number of balls in a xed volume result in a system with fewer head-on collisions.

5. Discussion Results presented in Fig. 4, and Figs. 7 and 8 represent experimental milling times and their computational analog. The trends observed in experiment and numerical simulations are qualitatively similar. However, direct comparisons are dicult. The signicance of changes in the milling times as a function of ball size for a given charge ratio needs to be assessed. Also, it is unclear which collision types and their respective rates of energy dissipation are critical in achieving structural renement in the mechanically alloyed powder. The following discussion is aimed to address the above questions. Recently, it has been suggested that the progress of mechanical alloying or reactive milling can be described using the specic milling dose, dened as the work performed on the powder [33,34]. The structural renement of the milled powder increases with the amount of work W performed per mass of powder. In the rst approximation, the work is proportional to the product of time t, the number of collisions ncoll, and energy per collision Ecoll: W $ ncollEcollt. This work is also proportional to the mass of powder, therefore, the milling dose, Dm, is introduced similar to [34] as Dm W ncoll Ecoll t $ ; mp mp 7

at a specic state of the milled material, that is, a specic degree of grain renement. Therefore, the eect of milling intensity on the ignition triggering is also neglected. Based on the above assumptions, the critical milling dose D m can be introduced that characterizes the milling progress when the achieved degree of grain renement is adequate for triggering the ignition mechanically. Eq. (7) can be analyzed considering the trends observed in the DEM results relating the number of collisions to the number of balls, Eq. (5), and the energy per collision to ball mass, Eq. (6). Thus, the critical milling dose can be expressed as D m $ nb m b t C R t; mp 8

where the denition for the charge ratio CR = nb mb/mp is used. For a certain specic degree of renement, it is expected that C R tinit const; 9 where tinit is the time when the reaction is initiated. This reasoning suggests that the milling time required to trigger initiation depends on the diameter of the milling balls exclusively via the charge ratio. A similar relation was suggested by Delogu et al. [35], however, the analysis of mechanical alloying processes presented in [35] led to a conclusion that the product of the charge ratio squared and milling time, C 2 R t init , should be constant. The product of the measured milling times leading to initiation and the charge ratios, CRtinit, is plotted as a function of the ball diameter in Fig. 9. The values of CRtinit for a series of experiments with dierent CR superimpose and do not change signicantly. Thus, to rst approximation, the present observations appear to support the tentative trend as expected from Eq. (9). Signicant deviations from constant behavior predicted by Eq. (9), or even linearity exist, however, especially for smaller ball diameters where milling times are greater than expected. The observed relative deviations from linearity are dierent for AlFe2O3 powder with dierent sample mass and therefore a dierent total number of balls. These deviations are however identical for
Al-Fe2O3, 2g Al-MoO3, 2g

where mp is the powder mass. It was assumed that the value of Dm determines the state of the milled material achieved as a result of specic progress of mechanical alloying. Eq. (7) does not consider the eect of the milling intensity on the state of milled material or degree of structural renement achieved. This requires, as discussed above, that the milling intensity remains within the range necessary to achieve the desired renement or structure. It is also assumed that ignition is triggered
550 325

Al-Fe2O3, 5g

CR t, min

100 78 55 32 2 3 4 5 6 7 8 9 3 4 5 6 7 8 9 3 4 CR 5 6
2.5

8
5

9 10
10

Ball Diameter, mm

Fig. 9. Experimental data on reactive milling of AlMoO3 and AlFe2O3 compositions presented as the product of charge ratio and milling time plotted versus ball diameter.

2916

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918


2g 5g
CR=2.5 CR=5 CR=10

AlFe2O3 and AlMoO3 with the same mass and therefore the same total number of balls. This indicates that variations in collision statistics are observed depending on the total number of balls. It is interesting to note that especially for small balls, a resonance condition appears to exist as the number of balls is varied. Fig. 10 shows the milling dose for Al Fe2O3 powder at a constant ball diameter, but varying number of balls. It is observed that at 10 g, all balls follow a collective motion with only small relative velocities, resulting in drastically weaker impacts and therefore longer milling times. The comparison of CRtinit between the 2 and 5 g sample mass for the AlFe2O3 proves that the CR in Eq. (9) accurately scales the sample mass, however, the observed dependence on the ball diameter is not intuitive. The values of CRtinit are dierent for AlMoO3 and AlFe2O3. It is reasonable to expect that dierent materials or dierent stoichiometry require mixing of the components on dierent length scales for initiation to occur. The proportionality among material systems could be further quantied in future work as a function of the activation energy of the exothermic reaction. The DEM results characterize the energy dissipated in the milling process, while the experimental results characterize the time or energy input per unit mass of powder. A theoretical milling dose function can be introduced as D m C R mp ; Ed 10

Milling Dose CRmp/Ed, g/W

10 8 7 6

Head-on Total

10

Ball Diameter, mm

Fig. 11. Theoretical milling dose calculated for head-on and total collision energy dissipation rate. Simulations were performed for the 2 and 5 g sample mass and for the three charge ratios of 2.5, 5, and 10.

where mp is the mass of the powder load and Ed is the average rate of energy dissipated. Dierent types of collisions can be more or less signicant regarding the energy dissipation rate as demonstrated in Fig. 8. The theoretical milling dose for head-on and total Ed as a function of the ball diameter is shown in Fig. 11. The milling dose is shown for the number of balls that correspond to the three charge ratios and the two powder loads of 2 and 5 g, as in the experiments illustrated in Fig. 4. The calculations based on the total Ed reveal a

relatively constant milling dose for all ball diameters. The milling dose plot with the head-on energy dissipation rate, Eh, shows a dierent trend that correlates better with the experimental milling dose as shown in Fig. 9. Although not shown, the trend for the glancing Ed is the dierence of the total Ed and head-on Eh, thus a maximum for Eh is a minimum for the glancing Ed, resulting in the constant total Ed trend. Based on the better correlation of the experimental milling dose and its computed analog derived from 1/Eh, the head-on collisions are suggested to be more signicant in dening the milling progress. This nding agrees with reports stating that the sliding or glancing impacts do not contribute signicantly to the deformation, coalescence, and fragmentation [11]. Therefore, the glancing Ed should be neglected in the formulation of the theoretical milling dose. The theoretical milling dose can then be reexpressed in terms of the head-on energy dissipation rate E h: D m C R mp Eh 11

600

2g 5g

400

200

0 0 10 20 30 40 50

Total Ball Mass, g


Fig. 10. Milling dose recorded for AlFe2O3 thermite as a function of total ball mass for 2.36 mm balls.

Fig. 9 and the top portion of Fig. 11 can now be directly compared. The theoretical milling dose has a divergence similar to the experimental milling dose with small diameter balls, but demonstrates less ecient milling for CR = 10 rather than CR = 2.5 as the experimental results. For CR = 5 and 2 g milling sample for both systems, the trends are in good agreement with the DEM prediction. The validity of Eqs. (9) and (11) is also well represented in the DEM results with a near constant milling dose for 4.76 mm balls among all charge ratios. This intersection point represents the most ecient milling condition for the examined charge ratios. In a scaled up operation, the convergence, similar to that observed in this work for 4.76 mm diameter balls, may occur at a

CRt, min

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918

2917

dierent ball diameter, but numerical modeling is now expected to identify the convergent ball diameter and generate the additional parameters required for successful production. The theoretical milling dose, although expressed in terms of energy, behaves similarly to the experimental milling dose and demonstrates the models ability to capture the physical interactions during milling. 6. Conclusions This research established that the sharp temperature increase occurring in reactive milling of powders capable of highly exothermic reactions is a useful indicator of the milling progress. The milling progress for a range of parameters can be expressed by a relationship called the milling dose, a function that is proportional to the product of the milling time and the charge ratio. Experimentally, the milling dose was found to be constant for a specic material system over a range of charge ratios. The concept of the milling dose validated the experimental approach as a practical technique to determine the milling progress. The mechanical alloying process in a SPEX shaker mill was successfully simulated using a DEM numerical scheme with a soft sphere interaction model. The DEM model produces numerical values for the energy dissipation rate that are assumed to be inversely proportional to milling time. The head-on collision energy dissipation rate was selected for use in the theoretical milling dose because it demonstrated a better correlation to the experimental milling dose. As the model qualitatively predicts the trends observed in the experiments, it is suggested that the model accurately describes the progress of mechanical alloying. Although this model does not consider local phenomena and its implementations are restricted for the cases when the milling intensity is within the range required for synthesis of an alloy with specic structure, it conrms that an energy transfer approach is adequate to model the progress of mechanical alloying. Acknowledgments This work was supported in parts by ONR, Grant N00014-00-1-0446, NSWC Crane Division, Award N00164-02-C-4702, TACOM-ARDEC, Award DAAE30-03-D-1015, and New Jersey Commission on Science and Technology, Award No. 01-2042-007-24.

References
[1] Benjamin JS. Mechanical alloying a perspective. Metal Powder Rep 1990;45:1227.

[2] Suryanarayana C. Mechanical alloying and milling. Prog Mater Sci 2001;46:1184. [3] Schoenitz M, Ward T, Dreizin EL. Preparation of energetic metastable nano-composite materials by arrested reactive milling. Mater Res Soc Symp-Proc 2003;800:8590. [4] Zhang DL. Processing of advanced materials using high-energy mechanical milling. Prog Mater Sci 2004;49(34):53760. [5] Eckert J, Borner I. Nanostructure formation and properties of ball-milled NiAl intermetallic compound. Mater Sci Eng A 1997;239240(12):61924. [6] Gaet E, Abdellaoui M, Malhouroux-Gaet N. Formation of nanostructural materials induced by mechanical processings (overview). Mater Trans, JIM 1995;36(2):198209. [7] Delogu F, Schini L, Cocco G. The invariant laws of the amorphization processes by mechanical alloying I. Experimental ndings. Philos Mag A 2001;81(8):191737. [8] Aikin BJM, Courtney TH. The kinetics of composite particle formation during mechanical alloying. Metall Trans A 1993;24A(3):64757. [9] Aikin BJM, Courtney TH. Modeling of particle size evolution during mechanical milling. Metall Trans A 1993;24A(11): 246571. [10] Aikin BJM, Courtney TH, Maurice DR. Reaction rates during mechanical alloying. Mater Sci Eng A 1991;147(2):22937. [11] Maurice D, Courtney TH. Modeling of mechanical alloying: Part I, deformation, coalescence, and fragmentation mechanisms. Metall Mater Trans A 1994;25A:14758. [12] Maurice D, Courtney TH. Modeling of mechanical alloying: Part II. development of computational modeling programs. Metall Mater Trans A 1995;26A:24315. [13] Harris JR, Wattis JAD, Wood JV. A comparison of dierent models for mechanical alloying. Acta Mater 2001;49(19): 39914003. [14] Courtney TH, Maurice D. Process modeling of the mechanics of mechanical alloying. Scripta Mater 1996;34(1):511. [15] Schaer GB, McCormick PG. Anomalous combustion eects during mechanical alloying. Metall Trans A 1991;22A: 301924. [16] Schaer GB, McCormick PG. Displacement reactions during mechanical alloying. Metall Trans A 1990;21A:278994. [17] Takacs L. Prog Mater Sci 2001;47:355414. [18] Lu L, Lai MO, Zhang S. Modeling of the mechanical-alloying process. J Mater Process Technol 1995;52(24):53946. [19] Munir ZA, Aselmi-Tamburini U. Mater Sci Rep 1989;3(7 8):277365. [20] Maurice D, Courtney TH. The physics of mechanical alloying: a rst report. Metall Trans A 1990;21A:289303. [21] Fischer SH, Grubelich MC. A survey of combustible metals, thermites, and intermetallics for pyrotechnic applications. In: 32nd AIAA/ASME/SAE/ASEE Joint Propulsion Conference, July 13; 1996. [22] Huang H, Dallimore MP, Pan J, McCormik PG. An investigation of the eect of powder on the impact characteristics between ball and a plate using free falling experiments. Mater Sci Eng A 1998;241:3847. [23] Cundall PA, Strack ODL. A discrete numerical model for granular assemblies. Geotechnique 1979;29:4765. [24] Hashimoto H, Watanabe R. Model simulation of energy consumption during vibratory ball milling of metal powder. Mater Trans, JIM 1990;31(3):21924. [25] Watanabe R, Hashimoto H, Lee GG. Computer simulation of milling ball motion in mechanical alloying (overview). Mater Trans, JIM 1995;36(2):1029. [26] Dallimore MP, McCormick PG. Dynamics of planetary ball milling: a comparison of computer simulated processing parameters with CuO/Ni displacement reaction milling kinetics. Mater Trans, JIM 1996;37(5):10918.

2918

T.S. Ward et al. / Acta Materialia 53 (2005) 29092918 [31] Chen W, Dave RN, Pfeer R, Walton O. Numerical simulation of mechanofusion system. Powder Technol 2004;146(12):12136. [32] Allen MP, Tildesley DJ. Computer simulation of liquids. Oxford: Clarendon Press; 1987. [33] Delogu F, Deidda C, Mulas G, Schini L, Cocco G. A quantitative approach to mechanochemical processes. J Mater Sci 2004;39(1617):51214. [34] Cocco G, Delogu F, Schini L. Toward a quantitative understanding of the mechanical alloying process. J Mater Synth Proc 2000;8(34):16780. [35] Delogu F, Orrhu R, Cao G. Chem Eng Sci 2003;58:81521.

[27] Walton OR, Braun RL. Viscosity and temperature calculations for assemblies of inelastic frictional disks. J Rheol 1986;30(5): 94980. [28] Kano J, Hiroshi M, Saito F. Correlation of grinding rate of gibbsite with impact energy of balls. Particle Technol Fluid 2000;46(8):16947. [29] Walton OR. Numerical simulation of inelastic, frictional particle particle interactions. In: Roco MC, editor. Particle two-phase ow. Boston (MA): Butterworth-Heinemann; 1993. [30] Walton OR. Numerical simulation of inclined chute ows of monodisperse, inelastic, frictional spheres. Mech Mater 1993;16: 23946.

Potrebbero piacerti anche