Materials Science and Engineering A 375–377 (2004) 120–126

Development and applications of ultrafine aluminium powders

Martin Kearns

The Aluminium Powder Company Limited, Forge Lane, Minworth, Sutton Coldfield, B76 1AH, UK

Abstract

Over the last 20 years or so, a variety of new technologies has been developed to produce sub-micron powders. Among the productsattracting interest is nanoaluminium which is being evaluated in specialist propulsion and exothermic end-uses. This paper examines theadvances made in ‘nanopowder’ production in the context of the existing aluminium powder industry where finest commercial grades havea median size of∼6�m (one or two orders of magnitude coarser than nanopowders) and which today supplies the markets being targetedby nanopowders with coarser, but effective products. Are there genuine market opportunities for nanoaluminium and if so, how will they beproduced? One the one hand there are the novel nanopowder production methods which are high yielding but generally slow and costly, whileon the other, there is the very fine fraction from conventional atomising routes which generate a very low yield of sub-micron powder but whichnevertheless can translate into a meaningful rate as part of the bulk production. Can conventional routes ever hope to make sufficient volumesof nanopowders cost effectively and which will be the favoured routes in future? Moreover, what of the ‘ultrafine’ size range (∼0.5–5�m)which is of more immediate potential interest to today’s powder users. This paper seeks to identify the near term opportunities for applicationof low volume/high value ultrafine and nano powders.© 2003 Elsevier B.V. All rights reserved.

Keywords: Aluminium; Atomising; Superfines; Nanopowders

1. Introduction

Annual sales of aluminium powders and granules(∼<1 mm) worldwide is estimated at∼200 k tonnes perannum (tpa) primarily comprising sales to the metallurgical,chemical and paint and pigment industries[1]. Specialistend uses include rocketry, explosives, thermal spray, powdermetallurgy, etc. The vast majority of powders are producedby conventional air or inert gas atomisation though signifi-cant quantities are made by granulation of foils. Several ofthe bulk applications where aluminium is used as a pow-erful reductant or as a precursor in the chemical industryattract low premiums. There is therefore an economic driveto maximise production of superfine fractions (those de-fined as having median particle sizes in the range 5–15�m)which are precursors for the manufacture of increasingly so-phisticated pigments for decorative finishes on automotivesand domestic appliances.

For other end uses, such as thermal spray or powdermetallurgy, removal of superfines can be advantageous inimproving powder flow characteristics and reducing the

E-mail address: mkearns@alpoco.co.uk (M. Kearns).

hazards associated with airborne superfines and associateddust collection systems[2]. For the powder producer, thereis a drive to develop a balance of customers requiring com-plementary size fractions in order to sell the entire atomisedsize distribution. In order to increase the value of the port-folio, gas atomising technology has focused on achievinghigher yields of fine powders to satisfy the demand for finepigments. The natural progression is towards nanopowders,but with the exception of a few specialist niches, the fo-cus is on stepwise size reductions to service establishedAl powder markets. Atomising technologies are thereforebeing geared towards maximising the yield of superfines.While gas atomisation processes are generally regarded asinefficient, with as little as 1% of applied energy beingutilised in size reduction, significant amounts of sub-micronmaterial are generated during atomising. It is debatable howmuch further the technology can be developed to producesuperfines, but it will almost certainly be unable to bridgethe gap to nanopowder production methods (Refs.[3–11]).Ultrafines (∼0.5–5�m) would, however, be a more realistictarget.

Fig. 1 shows a schematic relationship between ‘productsize’ and approximate price range for aluminium. In the‘coarse’ limit, 660 kg sows or T-bars are available on the

0921-5093/$ – see front matter © 2003 Elsevier B.V. All rights reserved.doi:10.1016/j.msea.2003.10.160

M. Kearns / Materials Science and Engineering A 375–377 (2004) 120–126 121

1000

fuels

pigments

nano

Reductants

LME

ingot

Specialist

superfines

Ap

pro

x. p

rice, £

/kg

100

10

1.0

0.1

10-9

10-6

10-3

100

Particle Size, m

Fig. 1. Approximate prices for Al as a function of the product size: fromingot to nanopowders.

LME at ∼£0.9–1.1 kg−1 and this contrasts with a quotedprice of up to £700 kg−1 for nanopowders[12]. Fig. 2 is anapproximate indication of the volumes of aluminium pro-duced in different sizes. Without attaching too much signif-icance to the relationship, there is a striking trend betweenthe world’s aluminium product capacities and the physicalsize of the product—from smelter to nanopowders: this sug-gests an ultimate market size of 10–100 tpa.

2. Production of aluminium powders

A variety of methods is available for production of alu-minium powders from the melt but the most significantvolumes of fine powders are prepared by air and gas atom-isation. For gas atomisation an empirical relationship hasbeen determined relating the metal/gas ratio to the medianparticle size. For a given molten metal flux, then increasingthe incident gas flux will reduce the median particle sizeresulting in an increase in the yield of fines. The width ofparticle size distribution is also dependent on the design ofthe atomising head and detailed process variables and thisalso affects the yield of the fine fraction.

108

rodcoarse

powders

superfines

nanopowders

World Smelting Capacity

106

Mark

et

Vo

lum

e,

To

nn

es

104

102

100

10-9

10-6

10-3

100

Particle Size, m

Fig. 2. Approximate market size for Al as a function of product size:from ingot to nanopowder.

101.0

Gas/Metal Ratio, m3/kg 0.1

Band of powder size

results

1000

Med

ian

Par

ticl

e Si

ze, m

icro

ns

100

10

Fig. 3. Gas/metal ratio vs. Al particle size for close-coupled atomizationprocesses.

The efficiency of gas atomising processes is low (typi-cally quoted as∼1%) and whilst several process innovationshave enabled reasonable yields of superfines to be achieved(type and temperature of atomising gas, nozzle and mani-fold design), there are difficulties in achieving stable runningconditions if parameters are set to achieve very fine mediansizes.

Fig. 3 shows that the median particle size achievable bygas atomisation[13] tends towards an asymptotic limit withincreasing gas/metal ratio, suggesting that direct atomisa-tion of nanopowders would require unsustainable gas/metalratios. Nevertheless, as the median size decreases, the pro-portion of sub-micron product increases and while yieldswill inevitably be low, this is compensated for by the factthat the atomising process runs continuously at relativelyhigh throughputs (tonnes per day) compared with nanopow-der methods (typically kg per day). Therefore, meaningfulnet production rates may be achieved for nanopowders ifeffective separation can be achieved.

3. Ultrafine aluminium powders from gas atomisation

Fig. 4 shows SEM images of a fine fraction of gas atom-ised powder which has been separated from a coarser, gasatomised distribution. The SEM images confirm that indi-vidual particles are effectively spherical and that there is alarge size range from∼100 to 2000 nm. Laser particle sizeanalysis (Malvern Mastersizer) inFig. 5 shows a mediansize of∼1.9�m and that there is∼20% by mass which issub-micron.

The fact that atomisation has been performed in inert gasand the particle size is relatively coarse results in a relativelypure powder product with a relatively low oxide content.Analysis shows that there is 0.55% oxygen and 0.02% C.The low carbon level contrasts with levels found in somepowders produced by plasma technology[5]. Depending onthe final application, this is potentially advantageous sinceit is known that adsorbed carbonaceous species can stifle

122 M. Kearns / Materials Science and Engineering A 375–377 (2004) 120–126

Fig. 4. SEM images of 2�m median size gas atomised aluminium powder.

surface activity of Al powders[14]. The specific surfacearea (SSA) of this sample was determined by BET (surfaceadsorption) analysis to be∼2 m2/g. This value is consis-tent with that predicted from an assumption of the oxidethickness and the measured oxygen content of the powder.Values measured elsewhere on fine powders are in the range∼2–4 nm (Refs.[3–5]). Tetronics quote a typical analysis of15% oxygen (33% Al2O3) in their ‘aluminium’ nanopowder.

Fig. 6shows the relationship between median particle sizeand SSA and oxide content for selected gas atomised prod-ucts. For comparison, published data for nanopowders pre-pared elsewhere are included in the figure. It appears thatthe gradient of the relationship between the SSA and me-dian size is<0.5 (0.5 is the value expected of monodispersespherical powders) while the gradient for percent oxygenversus median size is >0.5. The deviation from the expectedgradient may indicate different size distributions and differ-ent oxide thicknesses on different particle sizes. There isvery good agreement with the values determined by Higaet al. [6] for chemically produced powders and by Tetron-ics [5] for plasma-produced nanopowders. Argonide’s Alexpowder appears to have a slightly thinner oxide in relationto its size, but overall there is a remarkably good correspon-

Particle Size Distribution

0102030405060708090

100

0.1 1 10Size, microns

%

Fig. 5. Malvern Mastersizer size analysis: Alpoco 2�m secondary col-lector powder.

dence between the amount of active aluminium (oxide level)and the particle size. Technanogy[12] claim to be able tocontrol oxide thickness in their product and so the relation-ship inFig. 6would not necessarily apply to their products.

Within measurement errors,Fig. 6appears to confirm thatbelow 10�m, particles may be treated as spherical and havea consistent oxide thickness. If the relationship is examinedat coarser sizes, there is a deviation towards relatively highoxide contents and surface areas which is indicative of in-creasingly irregular particle shape.

It is known that fine aluminium powders (<45�m)can form explosive mixtures with air. Indeed it is thehighly exothermic reaction of aluminium with oxygen (�H(298 K) = −1678 kJ/mol of Al2O3) that confers the de-sirable reducing and propulsive properties of the powder.The hazards associated with aluminium powders have beendescribed recently[2] and it is not surprising that thesebecome more acute as the powder size reduces, particularlybelow 45�m. Increased reactivity of nanoaluminium hasbeen commented on by Tepper[3] and Tetronics[5] andinitial explosibility tests have been conducted on the gasatomised 2�m powder to examine its behaviour versus

0.01

0.1

1

10

100

0.1 1 10

Median Particle Size, microns

SS

A (

m2 /g

), %

Oxy

gen

SSAOxygen

SSA & %Oxygen vs Particle Size

Fig. 6. Relationships between median particle size and specific surfacearea (SSA), oxygen content. Includes data for atomised powders andquoted values from refs.[4,2].

M. Kearns / Materials Science and Engineering A 375–377 (2004) 120–126 123

Table 1Shows a selection of explosibility data for different aluminium powders

Powder size Sample origin MiE (mJ) MEC (g/m3) Test house

−75, +45 Screened air atomised 300–400 750 Chilworth−38�m Screened air atomised 50–60 60 Chilworth6�m median Gas classification 13 n/a Chilworth2�m median Secondary filter 8.5 70 Wolfson BS 5958

0

100

200

300

400

500

0 10 20 30 40 50 60

Median Particle Size, microns

Min

. Ig

nit

ion

En

erg

y, m

J

Fig. 7. Relationship between minimum ignition energy and aluminiumpowder size.

other commercial powders. Explosibility testing is used rou-tinely for the safe design of production plant and this uses a20 l sphere apparatus for measuring explosion pressure tran-sients following ignition of a dust cloud within. Among pa-rameters characterising explosibility, the minimum ignitionenergy (MiE) is a measure of the reactivity of the powderand the minimum explosible concentration (MEC) is thethreshold concentration that will support ignition (Table 1).

Tests were done by Wolfson Electrostatics on a sampleof fine powder collected from Alpoco’s secondary filtrationsystem[15]. The minimum ignition energy of 8.5 mJ for2�m aluminium powder is somewhat lower than typicalvalues for superfines (Fig. 7). Note that measurements oneven finer samples may be affected by agglomeration of fineswhich will not allow true dispersion into the most potentdust cloud.

The increased reactivity of superfine powders is furtherconfirmed by DSC analysis in air which shows the earlyonset of oxidation (509◦C (2�m) and 562◦C (6�m)) andthe much enlarged exothermic peak areas compared withcoarser powder (seeFig. 8).

Table 2shows a selection of manufacturing methods for nanopowders

Method Producer Quoted size (nm) Production rate SSA (m2/g) Reference

Electro-explosion of wire Argonide 50–500 100 g/h 10–15 [3,4]DC Plasma torch Tetronics 50–150 2 kg/h short run 25–30 [5]Inert gas condensation physical vapor synthesis Nanophase Technologies 10–50 tpa 20–60Chemical: alane adducts US Navy 65–500 Low [6]Sodium flame encapsulation (SFE) AP Materials Industrial scale [7]Gas condensation Technanogy 20–200 kg/h per reactor >50 [18]Inert gas atomisation Alpoco 100–5000 0.5 kg/h 2

-6

-4

-2

0

2

4

6

500 550 600 650 700 750

Temperature (C)

Heat

flo

w (

mW

/mg

)

2 m powder

6 m powder

+38 m powder

Exothermic

Endothermic

Fig. 8. DSC traces showing oxidation followed by melting of aluminiumpowders.

4. Production of nanopowders

Several techniques are being developed today for the man-ufacture of nano aluminium. With the exception of gas atom-isation, exploding wire and plasma wire fragmentation, mostmethods seek to grow powders from an induced gaseousphase (a ‘bottom up’ approach rather than a fragmenta-tion method).Table 2lists some of the leading practition-ers: Alpoco’s ultrafine gas atomised powder is included forcomparison.

In general the methods break down into chemical/pyro-lysis methods and high energy plasma or current-assistedmethods. Typical production rates for nanoaluminium arequoted at between∼200 g/h and 2 kg/h. The fastest produc-tion route appears to be flame pyrolysis which is claimed tobe capable of delivering 1000–5000 tpa from a single reactorfor oxides and carbon black. AP Materials claim a uniquesodium flame and encapsulation technology for pyrolysisand subsequent encapsulation of particles before agglomer-ation occurs. They claim advantages in scale of production

124 M. Kearns / Materials Science and Engineering A 375–377 (2004) 120–126

and low cost versus IGC. Singhal and Skandan[11] quotes∼US$ 50 kg−1 production cost for a reactor with burner di-ameter of 12.5 cm scaled up to 100 tpa for oxides. Particlesize can be controlled by controlling pressure in the reactor.Reduced pressure in flame pyrolysis leads to rapid quench-ing and fine particles, but at normal pressure there is an op-portunity for coarsening by coalescence.

Nanophase Technologies employs plasma evaporation ofa metal substrate followed by cooling with carrier gas andcollection (physical vapour synthesis (PVS) or inert gascondensation (IGC)). The conditions in the carrier stagecan determine size and level of agglomeration. Aluminawith median size 10–50 nm can be produced with SSA of15–90 m2/g. Nanoalumina is offered at∼US$ 200 kg−1. TheIGC method involves evaporation of the precursor followedby homogeneous nucleation/condensation of powder in alow partial pressure of inert gas. Importance is placed oncontrol of the colloidal behaviour of nanoaluminium andcomplementary coating technology enables subsequent dis-persion.

Technanogy also operates an IGC method (reactor out-put∼lbs/h) and claim techniques for control of particle sizewithin ±10 nm and control of oxide thickness between 1and 5 nm[12]. They expect to deliver 20 t nanoaluminiumproducts in 2002. The SSA is >50 m2/g and the main appli-cation is in propulsion where the goal is to increase effectivepayload of space vehicles. IGC methods are also operatedby Nanoproducts Inc., who describe preparation of a pre-cursor solution of the required stoichiometry which is thensubjected to plasma evaporation. Higa et al.[6] have a solu-tion method for decomposition of solutions using transitionmetal catalysts. Choice of catalyst and concentration of reac-tants determines particle size. In general, chemical methodsappear to be advantageous in giving tight size distributionsand low impurity levels but production rates are relativelylow.

Argonide operate a unique exploding wire method formaking aluminium powders by passing high currentsthrough fine wires. This is a semi-continuous process andthe plasma formed subsequently condenses to form fineparticles (∼100 nm) in an argon gas environment. This re-quires expensive equipment and production rates are low.Tetronics’ DC Plasma route also has a wire feed, but theapplication of a plasma torch results in significantly finerparticle sizes (seeTable 2) and a noticeable level of carbonpick up (∼2.4 w/o) and other significant impurity levels.The very high surface area of nanopowders generally meansthat they are prone to adsorb impurities from the manu-facturing environment. Other highly energetic sources suchas spark discharge are also being applied to production ofnanopowders (e.g. CyTerra Corp., Advanced Materials andProcesses, November 2001).

The extremely high reactivity of fresh nanopowdersmeans that each technique has a method for collecting theproduct in an inert atmosphere and must ensure a controlledpassivation (e.g. controlled oxidation, salt or organic coat-

ing) before onward processing can be done safely (this isdescribed inSection 6).

5. Demand for ultrafine and nanopowders?

Demand for ultrafine and nanopowders can be expectedfrom several quarters. Paint and pigment producers havedriven the demand for finer powders for making productswith high covering power and novel aesthetics. The pigmentproducers are interested in finer and more consistent prod-ucts, because these will improve still further the coveringpower of the pigment. Allied to this, there are some con-sumers who produce conductive pastes and inks for elec-tronic devices. As miniaturisation continues, so the needto offer thinner and more uniform/consistent coatings willgrow.

Most of the projected applications for nanoaluminium aimto exploit the vast quantities of energy stored in nanoalu-minium. Fuels for space and naval vehicles and propellantsfor the military are perhaps the areas of greatest interest forapplication of nanoaluminium. Aluminium powder is usedtoday in solid rocket boosters, e.g. for the Space Shuttle andAriane 5 and there is an ongoing drive to reduce launchcosts and increase payload. Tetronics refer to an increase inburn rate of between 2 and 10× when using nano versusregular aluminium powder fuel[5] and this is in line withArgonide’s claim that their ‘Alex’ nanoaluminium doublesthe burning rate and increases manoeuvrability and thrustcompared with standard 20�m sized spherical aluminiumpowder[3,4].

Tepper describes a means for capturing their productsdirectly in kerosene to produce a powerful fuel which isfully consumed before it exits the engine giving maximumburn efficiency. At the same time, it is clear that relia-bility and consistency are critical for successful missionsand qualification trials will rightly be highly demanding.There is interest in the potential of the turbulent reactionof finely divided aluminium in contact with water to propelsuper-cavitating naval vehicles and ordnance[16]. Spe-cialist military pyrotechnics is another potential market:super-thermitic reactions are described by Lowe[12] whichcan be of particular interest for pyrotechnics, primers,detonators, etc.

Powder metallurgy is an underdeveloped market foraluminium: a consequence of the difficulties in sinteringpowders with such a tenacious oxide layer. While nanosizepowders will never be a commercial prospect for bulk PM,there is evidence that as initiators they could stimulate de-velopments in Al and ferrous systems. Tepper[3] reportsthat pressed pellets with other powders such as B, Ni canbe made to react at low temperatures and reference hasbeen made to in situ formation of NiAl coatings. The use ofconventional superfine (6�m) aluminium powder as a sin-tering aid for stainless steels has already been demonstratedby Degnan et al.[17]. In this case local thermitic reduction

M. Kearns / Materials Science and Engineering A 375–377 (2004) 120–126 125

1000

-3 -1 1 3 7 5 9

Water

Spun

Granule

Gas

Atomised

Die

Cast

Gravity

Casting

Rapid

Solidification

Splat/Ribbon

Exponent (x) of Cooling Rate, 10x Ks-1

Den

drit

e A

rm S

paci

ng, m

icro

ns

100

10

1.0

0.1

0.01

Fig. 9. Relationship between cooling rate and microstructural refinement.

(2Al + Cr2O3 (surface oxide)= 2Cr + Al2O3) generatesheat to initiate early reaction.

Groza[18] reports that early sintering of nanopowders canoccur at∼0.2 Tm for, e.g. Fe, W compared with∼0.5 Tm forconventional sintering. Tetronics report the onset of an oxi-dation exotherm at∼538◦C and Argonide at 548◦C (mea-sured by DSC[5], DTA [19]). The attraction of nanopowdersin this sense is not only in accelerated, lower cost sintering,but in allowing processing on thermally sensitive substrates,e.g. electronic interconnects described above. Another ex-ample is in the joining of dissimilar materials using a reac-tive interlayers, e.g. joining stainless steel to aluminium.

In addition to sintering aids, then ultrafine alloy powderswill also have interesting properties and strength levels areexpected to be much enhanced as grain size decreases. Inparticular, dispersion hardening alloys should attain highstrength and toughness. Such powders could be of interestfor surface application via cold or thermal spraying tech-niques[20]. Fig. 9 shows a schematic of the cooling rateversus particle size. Rapid solidification in finer powdersconfers desirable microstructural refinement that is ex-ploited in some powder metallurgical products to achieve,e.g. optimum dispersion strengthening or superplastic form-ing properties in consolidated parts[21].

Metallurgical reduction processes: Ultrafine aluminiummay be exploitable in some metallurgical processes, e.g. toenhance firing rates in the exothermic production of Cr, butthere are some practical difficulties which may preclude itsuse. For example, control of the reaction, safe handling offine mixes and extraction/collection of air-borne dust.

6. Practical aspects

The very feature that makes nanopowders so attractivefor some applications, namely its high reactivity, can alsolead to problems in handling and the preservation of reactiveproperties. As described above, as powder size decreases,explosibility increases and safe collection and encapsulation

methods are critical[2]. It may be desirable for the prod-uct to be kept as active as possible, but unpassivated Al ispyrophoric and can not be safely air-freighted. Added careis required in the containment, packaging and transport ofthese products.

Various producers refer to controlled oxidation, varyingthickness of the oxide film[12], capture in organic media,coating with salt or hydrophobic species, etc. Specific meth-ods will be geared towards certain end uses and it will beimportant to characterise the metastability and reproducibil-ity of reactivity of such species if their ultimate performanceis determined by the state of the film. Preservation of thesurface condition during subsequent processing will needto be ensured for applications as critical as rocketry wherea long shelf life is required and performance is absolutelyparamount.

7. Conclusions

A variety of methods are emerging for the production ofnanosized aluminium powders which are seen as having po-tential for application as propellants, high power metallurgi-cal reductants/sintering aids and high efficiency coatings inparticular. Each of the production processes discussed haslimitations: all have low throughput and some demand ex-pensive equipment and result in high oxide and impuritylevels. Between the established superfines (5–15�m) mar-ket and the future nanosize market, an ultrafines categorymay be defined between 0.5 and 5�m which could be moreaccessible and of more immediate commercial interest.

Results shown here demonstrate that even at 2�m me-dian size, there is evidence of exceptional reactivity. Whilethis size is at the limit of what can be produced today bygas atomising, this may yet be a more feasible short-termprospect for supply of meaningful quantities of powder. Lit-tle work has been done so far to characterise this fraction,but the initial signs are that further work is merited.

Acknowledgements

Thanks are due to Prof. D.G. McCartney of the School 4Min the University of Nottingham, for arranging SEM micro-graphs, to Dr. Rob Claridge of QinetiQ for arranging someof the explosibility tests reported here and to Dr. Isaac Changof Birmingham University, for DSC analyses and valuablediscussions. Last, but not least, thanks to Emma Wilby andPaul Winder of Alpoco, Anglesey, for other powder analyses.

References

[1] Aluminum Association APMI Statistics, 2001.[2] J.W. Wright, in: Proceedings of the Conference of PM in Automotive

Applications, Aluminium Powders: Hazardous or not? APMI, Troy,MI, November 2000.

126 M. Kearns / Materials Science and Engineering A 375–377 (2004) 120–126

[3] F. Tepper, Int. J. Powder Metall. 35 (7) (1999) 39–44.[4] F. Tepper, Powder Metall. 43 (4) (2000) 320–322.[5] Tetronics Company Literature, Mark Hull, Powder Metall. 45 (1)

(2002) 8–9.[6] K.T. Higa, et al., US Patent No. 5,885,321 (1999).[7] http://www.apmaterials.com.[8] J. Murakami, et al., US Patent No. 6,146,505 (2000).[9] M.C. Roco, T.C. Lowe, M. Krebs, Adv. Mater. Processes 42–43

(2001).[10] H. Eifert, D. Kupp, B. Gunther, Powder Metall. 43 (4) (2000) 310–

313.[11] A. Singhal, G. Skandan, Int. J. Powder Metall. 35 (7) (1999) 45–

51.[12] T. Lowe, Adv. Mater. Processes 63–65 (2002).

[13] J.J. Dunkley, in: Proceedings of the 1999 International Conferenceon Advances in Powder Metallurgy and Particulate Materials, vol.1, Vancouver Publishers, MPIF, Princeton, NJ, June 1999, pp. 3–12.

[14] T. Sercombe, G.B. Schaffer, Acta Mater. 49 (2001) 2671–2678.[15] Personal Communication, Rob Claridge, QinetiQ, 2002.[16] Scientific American, Advanced Propulsion Systems, May 2001, pp.

69–71.[17] C. Degnan, et al., Mater. Sci. Technol. 17 (2001) 1624–1634.[18] R. Groza, Int. J. Powder Metall. 35 (7) (1999) 59–66.[19] M.M. Mench, K.K. Kuo, C.L. Yeh, Y.C. Lu, Combustion Sci. Tech-

nol. 135 (1998) 269–292.[20] S.J. Harris, et al., Mater Sci. Forum 331–337 (2000) 519–526.[21] R. Grimes, et al., in: Proceedings of the Institute of Materials

Congress 2000, Cirencester, UK, April 2000.