Ana]ysis of Reaction Products and Conversion Time in the Pyrolysis of Cellulose

and Wood Particles

R. S. MILLER and 1. 13 ELLAN Jet Propulsim laboratory California Institllte of lkchIologM Pasadena,
CA 9]109-8099

(submitted March 15, 1996)

AIWHUCT-A detailed mathematical model is presented for the temporal and spatial accurate modeling of did-fluid
r~~ctions in porous particles for which vohrmctric reaction rate data is known a priori and both the porosity and the
Permeability of the particle are large enough to allow for continuous gas phase flow. The methodology is applied to the
pyrolysis of spherically symmetric biomass particles by considering previously published kinetics sehcmcs for both cellulose
and wood. A parametric study is performed in order to illustrate the effects of reactor temperature, heating rate, porosity,
initial particle sim and initial temperature on char yields and conversion times. It is observed that while high tcmperatunx
and fast heating rates minirniz~ the production of char in both reactions, practical limits exis[ due to endothcrnlic reactions,
heat capacity and thcmml diffusion. Three pyrolysis regimes arc identified: 1) initial heating, 2) primary reaction at the
effective pyrolysis tcnqxxature and 3) fiMl heating. The relative durations of each regime are independent of the reactor
temperature and arc approxir;~tcly 200A, 60°A and 20% of tlw total conversion time, respectively. The results show that
modck which neglect the thermal and species boundary layers exterior to the particle will generally over predict both the
pyrolysis rates and exq)erimcntally obtainable tar yields. An evaluation of the simulation results through comparisons with
cxpcrirncntal data indicates tlu~t the wood pyrolysis kinct ics is not accurate; particularly at high nnctor temperatures.

Key words: biomass, cellulose, modeling, porous particle, pyrolysis, wood

NOMENCI.ATURE

A Frequency constant.
c. Cellulose.
c Specific heat.
c1 Characteristic pore length scale.
1) Molecular species diffusivity.
e Specific internal energy.
1’: Activation energy.
K Reaction rate.
M Molecular weight,

1
 Numerical time lcveJ.
lotal number of species.
Pressure.
Radial coordinate.
Radial position.
Universal gas constmt.
Reaction source/sink term.
Time.
Temperature.
Gas phase velocity.
Wood .
Reaction molar fraction.
Gas phase mass fraction.
Solid plmse mass average excluding char.

I Icat of reaction.
E Porosity.
IJ Divergence of the velocity.
A qhcrmal conductivity.
Molecular viscosity.
Partial density.
True density.
Stefcan-Bolt7Lnmnn consLlnt.
w Emissivity.

o Initial value.
c (onversion.
eff Effective.

.9 Gas phz$c.
i Spccics i.

~ Species j.

2
 Particle.
Radiation.
Reactor.
Solid phase.
Total (all spcciesandphascs).
Thermal.
Constant volume.
Phase a.
Excluding char.

1 INTRODUCTION

Solid porous particle combustion (coal, solid waste, solid propclkmts, etc.) is a subject of wide spread interest
for both fundamental research and industrial applications, Modeling of this phenomenon is inherently difllcult
duc to complexities associated with the multi-phase .aspcct. Particle porosity, two- ,and three-phase interactions,
and ill-defined boundaries due to solid phase reaction products (e.g. char) need to be considered in a complete
rnodcl. 31c combustion of a solid fuel particle can be divided into six primary physical processes:

a) Solid fuel reactions at both the particle surface and the interior.
b) Secondary solidlliquid phase reactions at both the particle surface and the interior.
c) Gas phase reactions both intcmal and cxtemal to the particle.
c) Gas and liquid phase diffusion (“pore diffusion”) and/or convection within the particle.
f) MrEs transfer with the surroundings.
g) IIeat transfer with the surroundings.

“1 ‘he combination of the above processes which determines the combustion characteristics is dependent on the
particular reaction of interest. IIIC ratio of reaction to diffusion time scales is in general too small for kinetically
controlled models to be effective (see e.g. Di Blasi, 1996b). “Pore diffusion” can control the rate of reaction by
limiting the surrounding oxidizer delivery to the particle’s interior antior by cooling the particle’s surface due to
emerging interior gases, Solid phase fuels may produce additional solid and/or liquid phase products such as char
which can act to thcnnally insulate the particle. In addition, reactions of gas phase species outside of the particle
can influence both heat and mass transfer (e.g. cxothermic reactions).
Significcmt early research in this area h,as been directed at incorporating the above mentioned processes into
theoretical models of char combustion and gasification. Early models were limited to analyzing the external
particle surface reactions coupled with gas phase transport (Cararn & Arnundson, 1977; Mon & Amundson,

3
 1978; Sundaresan & Amundson, 1980). Extensions included the addition of inner-particle diffusion, reael:ion and
pore growth effects (Sotirchos & Amundson, 1984a; Sotirchos & Amundson, 1984b). Further efforts relied on
stochastic descriptions of pore distributions and overlap. This latter approach typically employs assumptions of
either spherical voids (1.acwcnbcrg et. al., 1987), infinite cylindrical voids (Gavalas, 1980; Gavakas, 1981), or
arbitrary void sizes and distributions (Bhatia & Perlmuttcr, 1980; 13hatia & Perlmutter, 198 1). All of the above
mentioned models rely on u priori knowledge of pore surface regression rates as a function of carbon oxidation
and invoke a quasi-steady state assumption for the particle surroundings.
While the above mentioned models have shown some success for modeling char combustion, these approaches
arc not directly applicable to other types of solid particle reactions. Reaction rates may occur volumetrically (not
only at exposed surfaces) and are more easily measured in this manner in the laboratory. Density chan,ges due
to thermal swelling and/or intermediate solid or liquid phase species may cause temporary pore shrinkage. In
addition, the presence of gas phase species within the particle results in a nearly uniform diffusion time scale
across the particle boundary and a quasi-steady assumption for the particle exterior cannot bc justif]ed.
One ty-pe of solid reaction which exhibits several of the above char~ctcristics is tic pyrolysis of biomass. As
biomass is heated in the absence of an oxidizer it produces char, tar and volatile gases, 11 is now wideJy accepted
that as the heating rate is increased, relative proportions of tar and volati Ie gases can bc incrcasscd while producing
Jitlle, if any, char (Reed et. cd., 1980; Dicbold, 1980). Fast pyrolysis in the ablation regime has been investigated
for wood rods in contact with a hot, spinning disc by Lede et. al. (1985) who found that fmt pyrolysis is possible
when both high heating rates and cf?icicnt removal of the reaction products arc present. Possible applications of
these processes involve the rapid heating of ground biomass particles (e.g. waste saw dust) in either fluidizcd
beds (Lirn el. d., 1995) or vortex reactors (Diebold and Power, 1988) in order to maxirnizc tar and volatile gas
yields, Ihe collected gases and tars can bc further processed for use in adhesives, resins or for hydrogen fuel
production whcrc,as large c}mr yields arc desirable for charcoal production.
“J’hc large diversity of biomass fccdstock has motivated analyses of the somewhat simplified case of CCIIU1OSC

pyrolysis. ]n general, biomass is composed of approximately 50’% cellulose by mass (I)i Bl~si, 1993b) and many
of the kinetic and hydrohhcrmo- dynamic processes involved in cellulose pyrolysis may be common to the more
general case of biomrm. The majority of previous modc]s for both CCIIU1OSC and wood pyrolysis arc based on first
order kinetics schcmcs. These models range in comp]cxity from one-step global to multi-s[ep kinetics involving
both primary and secondary reactions (see Di Blasi (1993b); Antal and Varhcgyi (1995) for rcccnt reviews).
Only the most recent of these models have attempted to incorporate hydrodynamic and thcrmodyn.mnic effects.
Kothmi and Antd (1985) investigated the effects of hcatup time and dcvolatilization time on the flash pyrolysis
of cellulose. lhcy found that time delays and endothermic reactions place practical limits on attainable palticle

4
temperatures. Simmons and Gentry (1986) studied the kinctieall y controlled regime of the cellulose pyroiysis of
sub-millimeter sized particles. Using a mathematical model with prescribed particle surface conditions, they were
able to make predictions ofthc range ofkinctic control as a function of particle size and heating temperature. Poor
comparisons with experiments were attributed to neglect of the external thermal boundary layer in the mode]. Di
131asi (1994) extended the kinetics scheme of Bradbury et al. (1979) to include secondary reactions of volatiles
to simulate the pyrolysis of cellulose slabs. l%e model accounts for both heat and nm.ss transfer within the slab
through an equation for intcmal energy and I)arcy’s law for the gas phase velocity, and has been cxtcndcd to
multi-step kinetics for wood pyrolysis (Di 131asi, 1992; Di Blasi, 1993a). ~l~c model is applicable only within the
slab and the particle surface conditions arc assumed functions of the rcaetor tcmpxaturc. However, effects “on
pyrolysis due to the thermal boundary layer and chemical reactions outside of the particle can become significant
and have recently been connected to the wide variation of kinetic parameters measured in experiments (Narayan
and Antal, 1996).
“Ilc objcctivc of this paper is to present a model of solid particle reactions which is sufllcicntly robust to
incorporate all of the above listed physical proecsses both internal and external to the particle, and which is
applicable to volumetric reaction rate data, In particular, no quasi-steady assumption is made for the particle
surroundings, and the particle surface cxmditions arc allowed to evolve in a “natural” manner determined by the
far field temperature and pressure. The model is then applied to cxamp]c Ca.SCS of spherically symmctnc cellulose
~and wood particle pyrolysis in an initially quiescent environment of high temperature stc’am. The effects of
reactor temperature, heating rate, initial particle size, initial porosity and initial particle temperature on both char
formation and conversion times are investigated. A ~n~parison of the simulated results with experimental data
is nmde and is discussed in evaluating the accuracy of tic assumed kinetics schcmcs. 3fic paper is organized as
follows: Section 2 presents the general model equations together with the kinetics schemes and properties for both
CCIIUIOSC and wood reactions. Numcrieal solutions and a parametric study are presented in Section 3 along with
discussions of char yields, conversion times, spatial tar distributions ~d Comparisons with exprimcnts. section
4 is devoted to conclusions and further discussions.

2 TI II;ORETICAL FORMULATION

Consider a single porous solid particle having both a porosity (ratio of pore volume to total volume) and a
pcrlncabi] it y sutlicient] y large to al]ow for continuous gas flow. Assume that the particle is allo~ved to react
and that only volumetric reactions arc known. In this CaSC, a dct’ailed analysis of the internal pore structure is
superfluous bccausc only the bulk “cffcctivc” proper-tics of the porotls rnatcria] arc nccdcd. We allow fbr the
general case in which both liquid and gaseous reaction products may lx formed; however, both convection and

5
diffusion of the liquid arc neglected. ‘His last assumption is justified provided that either the viscosity of the
liquid is subst.antiaily larger than that of the gases or the reaction time scale of tic liquid is much smaller than its
convection and diffusion time scales.

2.1 Govemitg equations

21c governing equations for the solid particle dynamics are pmscntcd in spherically symmetric form; the extension

to muhi-dimensions being straight forward, A combination of two perspectives is employed to dcscnbc the particle
dynamics. In the first approach, the various spccics and phases within the particle are viewed as a “mixture. ” In
this c,asc, it is the partial densities which arc relevant. We denote partial densities as Po,i and true densities as ~Q,i,
where the subscripts denote both the phase [~ == s (solid or liquid), g (gas)] and the species (i == 1, . ..JV. where
N is the total number of species). Note that the above assumptions itnply that solid and liquid phax species are
treated identically. The mass conservation equation for the solid phase spccics is:

(1)

where the appropriate source/sink terms due to reactions arc contained in 3 (subscripts as dcfmed above) and
both convection and diffusion are neglected. In similar fishion, the gas ph.asc ecmtinuity equation is:

(2)

where the radial coordinate is r, tic radial velocity component is u, the gas phase partial density is p~ =- >Jg p~,i
,and the corresponding source term is ~g=- ~.g ~g,i (the summation Zadcrlotcs a SLUII OVer all s~cics of phase
a). SO far, the porosity of the particle (s) only appears indirect] y through the relation between the partial and true
densities:
(3)

(4)

III general, the true densities of the solid phase species arc described through additional equations of state;
however, for the purpose of the present work they are assumed to have prescribed constant values, Fly definition,
the porosity is based only on the gas phase volume whereas the stationary medium includes both the solid aild
the liquid phases.
2?IC above equations arc coupled through both source terms and additional equations for species, momentum
and energy conservation, along with an appropriate equation of state for the gas ph~zsc. I’hc spccics conservation
equations arc formulated in terms of the gas phase mass fraction ~ (Z9 Yi =- 1):

5)
I}c species diffusion is assumed to be Fickian l~ti an ef%ctive nlolecular diffusi~ty ~$~f due to porosity
effects. Following classical porous media theory and empirical measurements, the effective diffusivity is assumed
to bc proportional to the porosity and is modeled as D$/, = e~~(i) where l~ii) is the molecular diffusivity of
species z (Rear, 1972). In general, the porosity is represented by a rank two symmetric tensor to properly describe
anisotropic pore distributions; however, only locally isotropic porosities and diffusivities arc considered in the
current work as more precise information is not available.
lhc conservation equation for the gas phase momentum is modeled through a “channel” description (second
pcrspcctivc), i.e. as a mixture of gases flowing through individual “channels” (pores) within the particle:

where the velocity divergence is:

11 == -l-S- 7’2zf , (7)
r’ 6% ( )
the pressure is p and the effective molecular viscosity is peff. ~le fully compressible form of the momentum
equation is employed, as a source term in the continuity equation results in non-zero divergence of the velocity
field. The effective viscosity is modeled in a similar manner as the diffusivity; however, the values of the gas
spccics’ molecular viscosities (p(i)) are locally nm.ss averaged to account for mixture effects:

(8)

Although the gas diffuses according to the effective viscosity relation, additional drag forces arc experienced by a
flow convecting through a porous medium. l’hcse effects arc due to the geometry of the voids and to viscous shear
stresses along solid-gas boundaries. In order to avoid a complicated analysis of the pore geometry, these effects
arc lumped together and modeled by “damping” the convective terms in the momentum equation proportional to
the porosity. lhis damping is somewhat arbitrary, yet appears to bc rc,asonable under the previous restriction of
relatively large porosity ardor permeability. lle above momentum equation offers several advantages over the
traditional use of Darcy ’s Law which skltcs that the velocity is proportional to the product of the pressure gr,adient
and the permeability. Equation (6) is derived theoretically and with relatively few, and known, assumptions. It is
valid for both the interior and the exterior of the particle and takes full account of transient effects. Darcy’s Law
is completely empirical and has not been correlated for the case where gas phase sources occur within porous
media, nor for flows in which transient effects are important (Bear, 1972).
~hc tempcmturc is obtained through a “mixture” modeled conservation equation for the internal energy, Local
thermal cquilibnum is assumed for all species and ph,ases (1’ =- l;, all i). All species are assumed calorically
perfect such that the partial internal cnerg y (ei) is proportional to the specific heat at ccmskwlt volume; ei =

7
pi<~~)l’. With this noLltion, the g,as phase partial internal energy is:

and the total internal energy is:

()
eg == pg ~’ C$) 2“, (9)

(lo)
“=-C94 (~ps’c(’))7’
where C$*) is the constant volume specific heat of gas phase species i, and C(i) is the specific heat of either solid
or liquid phase spccics i. With this notation, the tokli internal energy is governed by:

(11)

where Fourier conduction with effective thcrrnal conductivity Aef ~ is assumed, only gas phase energies arc allowed
to convect and viscous dissipation is ncglcctcd. The last term on the right hand side (rAs) represents a sutnrnation
over all reactions which accounts for both heat rclc.asc and/or phase changes with heats of reaction A hi. Modeling
the effective conductivity is more complex than the previous transport properties due to the fact that heat transfer
occurs simultaneously through all ph,ases and species. We choose to model the muiti-pimsc heat transfer in terms
of a parGIlleI conduction model inspired by previous empirically tested models of two-species conduction in porous
media (Bear, 1972):
>:’ ~i~(~)
Aeff == (1 –&) + ~~,~ + ~ ~ Y;_’4ii) , (12)
{ x’ Pi 1
where ~(i) is the thermal conductivity of species i. Gas phase spccics are assumed to be transparent to radiation
while the solid and liquid phase radiation ,are modeled as in Chart et. al. (1988) with an effective conductivity:
~r,~ = a7’3d/w, where a == 5.67 x 10-1’ kJ/nz2sK4 is the Stef,art-iiloltzmann ccmst.ant, w is the ernissivity

and d represents a characteristic length scale for the pore size, The above model quantifies a “flow” of heat
simult.ancously (in pwalIe/, implying volume averaging) through both the solid and gas phases, where individual
spccics conductivitics arc mass averaged among species of tic same ph,me. While other conduction models have
been proposed (e.g. series and combinations of series-paraiIeZ), all arc at best intuitive and compare siniilarly
with cxpcrimcnts (Bear, 1972).
7?]c above set of governing equations is completed by an equation of state. lie total pressure driving the
~ascs is assumed to bc related to the temperature and dcmity through tic perfect gas law:

(13)

where Mi is the molecular weight of species i and 77 is WC universal gas constant. ‘Ihc prescncc of the porosity in
the equation of state indicates that the pressure is related to the ‘true” gas density jg. Equations (1 )-(13) describe
the essential physical processes of reactions in porous solid particles with nlulti-spccics/phase interactions. Ihcy

8
arc valid for both the interior and the exterior (c + 1) of the particle and provide a fully transient description of
the particle behavior with relatively fcw assumptions.

2.2 Biomass pyrolysis

I?lc equations derived in %ction 2.1 are applied to the pyrolysis of both cellulose and wood partichx by imple-
menting the kinetics schemes compiled by Di Illasi (1994) and Di Blasi (1993a), respectively. Ilc two reaction
schemes both employ a simplified decomposition of the primary solids to form three lumped product groups; solid
char, klr vapors and low molecular weight gas (Fig. 1). Ihc cellulose is additionally considered to pass through
can intermediate solid form labclled active cellulose. All of the reactions arc first order, irrcvcrsiblc and follow
Arrhcnius temperature dependencies of thc form Ki == Ai eXp[--}I;1/~~2’] where Ki is the reaction mtc, Ai is the
rate frequency constant, J;i is the activation energy and the subscript refers to the reacting spccics. The reaction
rates arc applied to the particle model to fom~ source terms of the form:

for the reaction of solid phase spccics i (~i) with corresponding production of spccics j (~j) and ~j is tic molar
fraction (note that there is no summation over repeated indices throughout the paper). Similar forms for the gas
phase reactions arc in terms of the partial densities. Vrducs of the activation cncrgics, rate constants and heats
of reaction are provided in Table 1. Reaction K2 is assumed to prcduce both char and gas in the respective
ratios of 0.35 and 0.65 and all remaining Xj arc equal to unity. All solid phase reactions, with the exception of
the primary celhrlosc decomposition arc endothermic while secondary gas reactions are exothemlic. Ilc reaction
frcqucncics for the wood pyrolysis are consistent with the original values as ca.lculatcd by Thumer and Mann
(198 1). 71c frequency constants for the primary wood reactions used in Di Blasi (1993a) appear to bc larger than
the originals by a factor of 3600; we were informed that this was a modification made by the author (Di Blasi,
1996a) to duplicate cxpcrimcntrd char densities. Implications of this modification will be discussed below,
Ihc particles considered in this work are initially at equilibrium in an environment composed of inert supcr-
hcatcd steam. All material properties arc taken from the compiled data of Di Blasi (1 994) and Di Blasi (1993a)
for the reacting species, while t.hc 1120 properties are from appropriate steam tables. Ihc wood kinctii;s and
reactions were originally compiled from several sources and do not correspond to any specific wood type. ‘Ihblcs
2 and 3 contain the solid and gas phase properties, respectively. ‘I_hc steam data is taken at a rcfcrcnce pressure
of 1) =-- 100k1’a and temperature of Y‘ .= SOOK, and unavailable gas properties arc assumed to bc the sarnc as the
steam. All of the true solid densities are defined in terms of the initial porosity of the material. ~hcrc arc two
advantages to this approach. First, expcrimcnta] measurements of the true density are very difllcult, whereas the
apparent, or partial, density is relatively simple to nlcasurc. Second, witi the density definition emp]oycd here,

9
the initial porosity may be varied in order to study its influenee without altering the total mass of the particle.
31w true density of char is assumed to bc the same as the corresponding primary reactant. The emissivity is taken
to bc w == 1 and the radiation Icngth scale is d== 4.0 x 10- 5m taken from Chan er. al. (1988).

2.3 Iilrbulence considerations

Ilc equations presented above do not include a turbulence model. Although the particles arc assumed to be

initially located in a quicsccnt environment, significant gas phase velocities can result in turbulent diffusion
around the particle. This could result in three primary alterations to the particle surroundings: 1 ) Incrcascd
tcmpcmture gradients near the boundary can increase the total heat flux to the particle. 2) Enhanced mixing of gas
phase reactants can significantly increase the rate of chemical conversion near the particle for diffusion flames.
3) ~’urbtllcnt diffusion of the gases can change the spatial distributions of exothermic reactions and therefore
indirectly affect the particle heating rate. Neglect of a turbulence model for the present biomass pyrolysis is
considered to bc justified in regard to 2) and 3) bccausc only first order reactions are considered (non-difllrsion
limited) and only mildly cxothermic heats of combustion arc involved. In regard to issue 1), the present study
treats the thermal radius, and therefore the heating rate, as a free parameter. ~hesc considerations combined with
a posterior analysis of the simulation results indicate that the neglect of a turbulence model is justified for the
present biomass particle pyrolysis.

2.4 lnitinl conditions, boundary conditions and the thernzd rodirts

A particle of initial radius }i,o and uniform temperature ~~,o is exposed at time t == O to a quiescent environment
composed of super-heated steam. Numerical solutions are obtained for the spherical 1 y symmetric domain within
the interval [0, }tR] where the outer domain radius, Ii!H (referred to as the reactor radius), corresponds to the
reactor conditions. Symmetry conditions arc employed at the inner domain boundary; i.e. u = O and d/& = O
for all remaining variables. We model the exterior boundary in a nmrmcr similar to the “sphere of influence” (S01)
approach of Bcllan and Cuffcl (1983) which was originally proposed to account for thermodynamic interactions
among liquid droplets in sprays (clusters). ~hc SOI represents a characteristic length scale for these interactions
and corresponds to the radius of a fictitious sphere located at the droplet center. ~fiis radius, }?It, is equal to one
half the mean droplet ccntcr separation. The proper boundary conditions at r == J?It are obtained by matching the
tcmpcraturc and pressure (and therefore the density) to the local reactor conditions while assuming that t3/i% = O
for all remaining variables. RR is a fl]nction of both the local number density and the three dimcnsiomrl pricking
factor. Wc extend the original SOI concept to introduce a “thermal radius” Ill < ~tR which is dcfrncd such that
the tcn)peraturc is held constant and equal to the IOCZJ reactor conditions, i.e. 7’ =- 7}{, for all T > lt~. In this
manner, the eflectivc heat flux reccivcd by the pafiiclc can bc altered indcpcndcntly of the reactor temperature

10
while resolving gas phase reactions exterior to the particle. lhe results presented in the following section arc
obtained for a single particle and a fixed normalized rcaetor radius ~?R == 101~,0 with constant rcaetor pressure
p = ]OOkPa. In addition to other quantities, both the reactor temperature and the themml radius are varied in
order to independently study the effects of the heating ternpcraturc and the heating rate, respectively.

2.5 Sohtion procedure

The modeled equations are solved numerically utilizing a procedure based on finite difference approximations for
both spatial and temporal derivatives. All cxmvective and first order derivatives are approximated via upstream
differences in order to maintain proper ellipticity of the equation set, while diffusion seeond derivatives are seeond
order ccntrzd. Acoustic waves are filtered through use of a semi-implicit iterative pressure solver. The basic method
is as follows: The solid phase density, gas mass fraction and internal energy equations are integrated one time
step (from time lCVCI 71 to n + 1 ) using an explicit forward time difference. These values give an approximation
for the gas phase density and all necessary properties at time level n + 1. A predictor value of the velocity is then
calculated from the gas phase continuity equation. Next, the pressure is obLlincd by solving a Poisson equation
obtained by taking the divergence of the momentum equation. The pressure is then used to eorrcct the gas density
and the process is repeated until ecmvcr-gcncc is achieved. All simulations are performed with a compressed and
staggered 64 grid point spatial resolution of the radial coordinate with compression increasing towards the origin
such that 21 grid points are within the initial particle boundary. Sixty eight simulations were performed on a
Cray-YMP supereomputer requiring a total central processor time of approximately 50 hours.

3 RESULTS

Before proceeding with a parametric study it is informative to illustrate the particle evolution through a “baseline
case” simulation for both cellulose and wood and to compare their behavior. The baseline case conditions are:
7~,o =-- 500K, 2~ == 900K, R7, == 51~,o, l~,o == 5nwz and e. = 0.7. qhc simulation is terminated at a final
time t, (the conversion time) at which the remaining solid mass, minus the char, has reached 0.l% of the initial
particle mass. ‘he ecmversion times for the ecllulose and wood simulations arc LC == 253.7s and t, == :347.8s,
respectively.
Figure 2 depicts the temporal evolution of the partial char density for both cellulose and wood. in both
cases, no char exists initially. ~hc small amount of char forming outside of the radial position lL,O is due to
the nccessar-y smoothing of the initial particle gradients for numerical resolution across the outer boundary of
the particle. As the particles are exposed to the high temperature steam environmen~ heat diffusion into the
particle results in a pyrolysis wave having a thickness ~ lnmt which travels radially inward producing solid char
residue. Approximately six times more char by mass is formed by the wood reaction than by the cellulose. Ihc

11
final porosity values are nearly uniform within each of the particles; however, the average values differ and are
E(t = tc) s 0.98 and E(L = t.) w 0.91 for cellulose and wood, respectively. The smaller final char densities near
the origin are due to biomass remaining at the termination of the simulations.
As the pyrolysis wave travels through the particle, both La vapors and gas arc produced from the reacting
solids and, due to pressure gradients, are ejected from the particle. The production of both tar and gas drives the
inert steam out of the pores and away from the particle. Figure 3 indicates that the maximum tar mass fractions are
located within the particle. In this region, the tar fraction maintains a nearly uniform value duc to near uniform
tcmpcraturcs and reaction rates within the particle (discussed below). In addition, the slightly larger tar mass
fmctions observed for the cellulose particle indicate that the effective production rate of tar is larger for cellulose
th,an for wood. As tar is ejected, it encounters the hot environment resulting in increased conversion rates to gas.
In fact, for both cellulose and wood, nearly all of the tar is converted to gas within the range 7.< 51~,cI. This
r,angc therefore represents the eflkctive range for which exterior gas phase reactions may influence the pyrolysis
evolution due to exothcrmicity. The rapid tar decomposition near the particle has consequences for experimentally
obtairmblc pyrolysis product meastirements and is discussed in more detail below.
~hc temporal evolution of the temperature profiles can provide insight into the importance of the outer particle
regions in aflecting the pyrolysis process. Figure 4 illustrates this effect for both cellulose and wood. The
high temperature environment maintains relatively large temperature gradients at the particle surface, whereas
endothermic reactions smooth the internal thermal gradients. These competing processes result in a relatively thin
reaction zone located at the particle surface. The final core temperature for the wood panticlc is larger than that
corresponding to the cellulose particle despite the larger nm.w (with the same endothermic heat of reaction) of
the wood particle. This is a result of the smaller reaction rate of the wood particle as can be deduced from the
above listed conversion times. I?-Ic slower wood reactions absorb heat endothcm~ically at a lesser rate than the
cellulose; thus allowing for thermal diffusion to heat the particle to larger temperatures, Examination of Fig.4
reveals that the actual particle surface temperature never rcachcs the reactor temperature 7 ‘n. In fact, during the
majority of the conversion, the particle surface temperatures are w 200K less than l}~. Analysis of the surface
tcmpcraturc indicates that its rate of increase with time cannot accurately be modeled as a linear function . ~hesc
rcsu]ts suggest that pyrolysis models which neglect the outer particle thermal boundary layer may substantially
over predict the reaction rates.
Pressure gradients resulting from the conversion of solid to vapor produce maximum gas phase velocities
at the location of the pyrolysis front (Fig,5). Convection carries endothermically cooled interior tar and gas
out of the particle and thus aids in maintaining a relatively low particle surface temperature. The magnitudes
and relative protllcs of thc interior Vclocitics are in general agreement with the cellulose pyrolysis simulations

12
based on Darcy’s Law by Di Blasi (1994). Outside of the particle there arc no gas phase mass sources and
-
the velocities decay roughly as r 2 . The cellulose actually yields a larger nuxximum velocity than the wood
although it produces substantially less gas phase species by nmss. ll~is is because the velocity is related to the
reaction rate which is larger for cellulose. Analysis of these results along with others not shown here indicates
that an outer computational radius of 10l~,o is suflicicnt to capture the pertinent physics of the particle pyrolysis
without significant artificial boundary effects. In addition, due to the complete conversion of tar to gas within the
computational domain, further discussions are primarily limited to char yields with the caveat that the remaining
products are converted entirely to the gas species within the particle’s near field surroundings. However, in order
to provide insight into the role of the exterior gas phase reactions, a discussion of spatial tar distributions as a
function of the reactor temperature is provided in Section 3.1.3.

3.1 E~Jects of the reactor temperature

in this section, the focus is on the values of the conversion times and on the final char yields. Knowledge
of the final char yields is necessary in order to either maximize or minimize char formation depending on the
particular application; e.g. charcoal or volatile gas production, rcspcctivcly. On the other hand, conversion times
dic~atc the types of reactors and residence times which arc capable of producing the desired yields. 13c most
obvious parameter influencing both t, and the pyrolysis products is the reactor temperature. Thus simulations
arc conducted for both particle types for 600K < l’~ ~ 1500K with all other parameters equal to the baseline
case values (RT u 51~,o, l~,o u 57nm, E. =- O,T and Tb,o == 500 K). The cxmvcrsion times and char yields are
presented in Fig.6 for both cellulose (c.) and wood (w.). For both particles, the most substantial effects of the
reactor temperature occur for relatively low ambient temperatures. A “transition” value of the reactor temperature,
I}i M 7001{, appears to separate two regimes of pyrolysis behavior. The value 700K is not meant to quantifi
a precise transition temperature but is only listed as an apparent range for which the observed changes occur,
Below this value, the conversion times increase rapidly with decreasing reactor tcmpcraturcs (from minutes to
hours). ‘Ihe char yield for the cellulose particle also shows a relatively sharply increasing trend in this region.
~’hc wood particle shows a nearly constant char yield of x 0.32 for all reactor temperatures; however, the char
yields decrease slightly for 7R < 700}{. On the other hand, for ~~ > 700}{ only relatively mild variations of
both the char yield and the conversion time arc observed for both particles.

3.1.1 Kinetic limitations

‘I ‘he pyrolysis trends observed in Fig.6 arc best explained in terms of two competing processes; 1) reaction
kinetics and 2) ‘thermal inertia.” Kinetics dictate the limitations imposed on pyrolysis due to reaction rates,
whereas thermal inertia accounts for limitations imposed on both the particle tcrnpcraturc and its rate of increase

13
due to diffusion, heat capacity and endothermic solid reactions. If the ratio of the thermal diffusion time scale
to the reaction time scale << 1, then particle pyrolysis occurs in the kinetieall y ecmtrolled regime. The particle
temperature is instantaneously raised to the rcaetor tcrnpcrature and any variations duc to exothcm~ickndothcrrnic
reactions are smoothed by diffusion infinitely fast. Under these conditions, it is possible to define both a kinetic
char limit and a kinetic conversion time limit for the rcaet.ions, both being functions of TR only. Tabk 4 lists
the definitions of these limits for both cellulose and wood pyrolysis where the reaetion rates (KJ are evaluated
at ~’ = TR. The conversion time is evaluated as that corresponding to a residual mass of 10-3 which makes the
definition independent of the initial particle size or mass. The time to convcti raw celhdosc to the active form
has been neglected as being much smaller than the remaining reaction times, and scecmdary char production has
been included for wood pyrolysis.
Figure 7 presents the kinetic limits on char yields and conversion times for both cellulose and wood pyrolysis
as a function of !f’~. 3’hc limiting values of the ecmversion times are observed to be strong functions of the
reaction tcmpcraturc. Complete pyrolysis at the kinetic limit can require anywhere from weeks to micro-scemds
within the range of reactor temperatures in current use. For all praetieal ranges for which char production is to
bc minimized (large I’J, the wood reaction is always slower than the cellulose reaction. It is well known that
commercial processes which aim at rna,,irnizing char yields generally employ long rcsidenec times with low final
heating temperatures. For example, char yields as high as 50’%0 have been reported from the pyrolysis of bagasse
at 530K with 65 hour heating times (see e.g. Antal and Mok, 1990). The kinetic char yield limit for cellulose
pyrolysis is in good qualitative agreement with these observations since the yield decreases monotonically with
reactor temperature. However, the wood reaction shows a peak char yield at ?‘~ x 6501{ with a relativcl y sharp
drop in yield below this temperature. Ilis apparent contradiction with cxpcrirnental observation may indicate a
flaw in the wood pyrolysis kinetics scheme.
In order to quantify the relative ratio of the kinetically controlled and the diffusion controlled pyrolysis regimes,
it is convenient to define two kinetic “cfllcicncy” factors. Both a yield efficiency and a conversion efllciency are
defined as the mtio of the kinetic limit to the actual simulation value of the final char yield and the conversion
time, respective] y. Values of these ratios which approach unity indicate the kinetically controlled regime, whereas
ratios approaching zero indicate strongly diffusion controlled pyrolysis, Both ratios must be in agreement as a
yield cfflcicncy of unity alone is not sufficient to conclude kinetically controlled pyrolysis. Both cff!ciency factors
arc illustrated in Fig,.8 M a function of l~t. It is apparent that diffusion effects are substantial throughout the
entire range of temperatures cxmsidered, Only as 7‘~ is decreased below 6001< and conversion times bceomc on
the order of hours and larger can biomass pyrolysis truly bc considered to bc kinetically controlled. In the case
of wood pyrolysis, kinetically controlled predictions of product yields may give reasonable results, even though

14
the conversion times indicate that diffusion effects are strong. This is an artifact of the present kinetics scheme
for wood pyrolysis which yields only mild variations in char yield with reaction temperature. The quantitative
results presented thus far are valid only for the particulru particles under consideration and are expected to be
functions of particle size and heating rate in addition to other properties (conductivitics, heats of reaction, etc. ).

3.1.2 Efective py?dysis temperahlre

A comparison of the results of Figs.6 and 7 suggests that the simulated pyrolysis of both cellulose and wood
is occurring within a relatively narrow effective reaetion temperature range slightly larger than 600K. This
hypotbcsis is confirmed by examining the mass averaged particle temperature:

(15)

~vherc pj indicates the partial solid phase density excluding the char. Figure 9 depicts the mass averaged particle
temperature evolution as a ftmction of nondimensional time for several values of the reactor tempemturc in the
range [7001(, 1200K]. The curves arc almost identical for both cellulose and wood when time is normalized with
tC. Examination of the mean particle temperature indicates three primary phases of evolution for both cellulose and
wood pyrolysis: 1) An initial rapid increase in temperature as heat diffuses into the particle from the surroundings.
2) A primary reaction regime at a nearly constant “effective pyrolysis temperature” during which endothem~ic
reactions strongly resist any further particle heating. 3) A secondary reaction regime within which only relatively
small particle mass remains and the total heat loss due to reactions is significantly smaller than heat diffusion into
the particle. The majority of the pyrolysis occurs in the prima~ regime within which the particle temperatures
for both cellulose and wood are limited to values in the range of approximately (600K, 6501Y). Only relatively
small variations of char yield and conversion time can bc expected for reactor tcmpcraturcs above this range,
311c behaviors associated with regimes 1) and 2) arc in agreement with the numcnczd results of flash CC1IU1OSC

pyrolysis by Kothari and Antal (1985) obtained using a much simpler particle model. Ihcir results indicate a
larger maximum particle temperature than Fig.9 indicates; approximately 7751f for particle diameters < 0.5rnnI.
I lowcvcr, these particles may have a mass too small for the endothermic heat absorption to overcome the inward
thermal diflllsion. The present results indicate that the effective pyrolysis temperature is dctcrmincd primarily
by the magnitude of the primary reaction endothermicity, the particle nm.ss and the reaction rate: The pyrolysis
temperature is therefore not directly related to kinetic and diffusion reaction limits.
Further evidence for the three pyrolysis regimes is found in Fig. 10 which depicts the particle heating rate as a
function of both nondimensional time and ~}t. For the purpose of this study, the heating rate is defined as the time
derivative of < 2; >. This definition is appropriate for the current discussion because the endothermic reaetions
maintain a fairly uniform temperature profile within the solid and the spherical mass averaging of the temperature

15
gives a larger “weighting” to the actual pyrolysis temperatures near the particle surface (the nonlinear temporal
dcpcndcnce of the surface temperature was discussed above). In addition, this definition eliminates any ambiguity
in prescribing the location of the particle surface and also neglects contributions from residual chm regions left
behind the pyrolysis wave which no longer influence the temperature via endothermicity effects. Figure 10 clearly
reveals the three pyrolysis regimes as distinguished by regions o~ 1) relatively large positive, but decreasing,
heating, 2) virtually zero heating rate when inward thermal diffusion balances endothermic absorption, and 3)
positive and increasing heating rates when the particle mass has bcc.omc too small for endothermic reactions to
prevent the particle tcmpcraturc growth. Both Figs.9 and 10 sug.gcst that the relative durations of the pyrolysis
regimes (with respect to tJ appear to bc independent of the reactor temper-ature and arc approximately 20%, 60°/0
and 20°/0, respectively. Ihe heating rates in the nom~alizcd time frame are almost identical except for early times
when large reactor temperatures result in increased heat fluxes into the particle. In general, the heating 12dteS are
always less ttmn 10K/s (a factor of ten larger than reactor heating rates used in traditional TGA experiments;
e.g. 13ilbao e(. al., 1992) and their functional dcpendcnee on time is highly nonlinear.
glc presrnce of the three regimes is not always readily apparent. For example, Bilbao et. CII. (1992) studied
the effect of the reactor heating rate on the thermal decomposition of spherical pine particles in the size range
4 .Ocmz s l~,o s 11 .2cnz. hey presented cxpcrimcntally obtained results for the temporal dependence of the
particle temperature for various radial positions inside of the particle. TIc initial particle masses are relatively
large and thus endothem~ic heat absorption should be substantial; however, only the particle core temperatures
exhibit the primary pyrolysis regime (as observed by intervals of near constant temper-ature). The reason for this
apparent discrepancy is twofold: First, the experimental mca.surcments for outer radial positions do not consider
that the pyrolysis front passes through these positions, leaving behind only char with no further endothermic
reactions to balance t}vmrnal diffusion. Second, the experiments ccmsidercd particles which began in a “cold”
reactor which was slowly heated to the final reactor temperature (from 303K to 923K at rates < 12K/ rnin).
Examination of their data shows that nc,arly 50?40 of the pyrolysis is cornplcte before the reactor tetnperaturc
rcachcs TOOK, even for the largest particle size and heating rote. 311c primary reaction regime will only bc easily
distinguishable when the reactor temperature is significantly larger than the effective pyrolysis temperature. This
is indeed the case for the results of Figs.9 and 10 in which the reactor temperature is constant and ~’R 2 700K

3,1.3 Ihr distribtttions

lmporkmt issues arc the accurate mcasurerncnt of tar yields and whether tar yields increase at a constant rate with
increasing reactor tcmpcraturc.
The general form of the tar profiles was demonstrated in Fig.3 and in the corresponding discussions. Another
consequence of the rapid depletion of tar in the pafiiclc’s vicinity is its influence in the expcnmcntal analysis of

16
pyrolysis vapor yields. 31w typical experimental method of vapor collection is to rapidly quench the gas phase
species in order to hault the reactions and to therefore achieve representative sarnpies of the various species yields.
For example, Scott et al. (1988) measure both cellulose and wood pyrolysis products using a “cryovottactor”
which quenches the gas phase products via injection of a turbulent stream of cryogenic nitrogen immediately after
the pyrolysis. Accurate measurements of the true product yields are only possible when the time lag between
pyrolysis and quenching is much sma.llcr than the tar conversion time scale. The results presented previously in
Fig.3 suggest that the sample measurements may vary substantially from the actual exit yields due to rapid gas
phase reactions occurring in the immediate vicinity of the particle.
An examination of the tar distribution within the particle surroundings provides a qualitative analysis of the
limitations on gas phase pyrolysis measurements as caused by tar decomposition near the particle. Figure 11
presents both the maximum normalized radius at which the tar mass fraction has decayed to 0.05 (the “5Y0 tar
radius”) and the maximum tar mass fraction for both CC1IU1OSC and wood as a function of the reactor temperature
(“maximum” indicates for all times). Radial values > 10l~,o indicate that the 5°/0 radius is Iocatcd exterior to
the computational domain. All of the actual tar profiles arc of the same general form as those previously shown
in Fig.3. These two parametetx provide a measure of how closely to the particle the tar is distributed and how
much tar is produced, respectively. ‘Ile 5°/0 tar radius and the maximum tar fraction show opposite dependence
on the reactor temperature. Figure 11 indicates that as the reactor temperature is increased, the tar evolves an
increasingly narrower distribution adjacent to the particle surface. on the other hand, the maximum tar (at the
surface and interior to the particle) increases as the reactor temperature is increased. These observations have
important consequences to both experiments and conmlercial reactors,
Figure 11 suggests that expcrimcnta] mcasurcmcnts may under predict the actual tar yields which exit the
particle as further tar decomposition occurs both rapidly and near to the particle. This argument can be used
to explain the results of Di Blasi (1996b) who used a numerical pyrolysis model, based on the same cellulose
kinetics scheme as employed in this paper, to predict the tar, gas and char yields measured by Scott et, al. (1988)
in the “cryovortactor.” 7’he Di Blasi model does not include reactions in the outer particle domain and the total
tar yields correspond only to the total tar exiting the particle surface. Ihe numerical model was applied for a
nearly kinetically controlled particle size in an attempt to duplicate the experimental conditions. Comparisons
showed good agreement for the total char yield predictions; however, with significant over prediction of the total
tar yields, Further tar decomposition may have occurred outside of the particles before the cryogenic nitrogen
stream could quench the reactions. In this case, it would be cxpcctcd that a pyrolysis model which neglects the
particle surroundings would display the above trends. 3hat is, an over prediction of experimental tar yic[ds, even
if the kinetics schcmc is accurate and the total char yield is well predicted.
 The results of Fig. 11 also indicate that there exists an optimal temperature for applications which aim at
maximizing tar yields (e.g. biomass pyrolysis for commercial hydrogen production). Although increasing reactor
temperatures produce increasing total tar yields, the relatively narrow distributions of the tar near to the particle
surface limit the effcctivc tar yields which can be hru-vested. Caution must bc cxcrciscd in raising the reactor
tcmpcraturc because the quenching of the pyrolysis products may not be fast enough to capture the incrmsed tar
productions, In this case the added expense of heating the reactor to large tempcmtures may not only be costly,
but may actually yield lCSS trrr than a lower, and less cxpcnsivc, optimal rcaetor temperature. TIc present model
is capable of predicting such optimal reactor temperatures. Experimental measurements of optimal tar prcxiucing
reactor temperatures as a fimction of both the reactor type and the biomass fccdstock are needed to validate the
predictions.

3..? Effects of additional parameters

In order to make useful predictions of biomass pyrolysis, it is necessary to extend the results of the previous
section to include the effects of R1, (lcating rate), 1~,0, ~0 and T1,,o. As the variation of each parameter is
studied, the remaining parameters are kept constant and take the baseline case values; 7~ = 900K, RT’ = 5%,
l;, = 5nznz, so = 0.7 and 7~,o == 500K. lle kinetic limits on yield and conversion time are not dependent on
these parametcm. As such, the efllciency factors are inversely proportional to the proceeding simulation results.
Therefore, quantitative values for these factors are only discussed intem~ittently and when pertinent; however, the
eflicicncy magnitudes can be calculated from the data provided in Table 4. In addition, the parameters studied in
this section show significantly less influence on the tm distribution than did the reactor temperature. As such, tar
distributions are no longer discussed; however, we note that with I}t = 900K all of the cases ccmsidcred in this
section are characterized by total tar decomposition within the simulated domain.

3,2.1 I:flects of the thermal radius

Heating rates arc known to strongly influence biomass pyrolysis yields and evolutions (SCC e.g. Antal and Mok,
1990). Previous numerical investigations which address this issue have been restricted to models which either
mode] the particle surface conditions or which usc steady state assumptions for the particle surroundings (see
e.g. Di Biasi, 1993a). Variations in the heating rate within experimental and cmnmcrcial reactors arc difficult
to model quantitatively. In addition to l’R, the heating rate is dctennincd by many factors. For example, free
stream vclocitics, turbulence intensity, contact with solid surfaecs and clustering of the biomasss particles within
the reactor may all stl-ongly affect the heating rate each particle cxpericnccs. Therefore, for the purpose of tic
present work, effects due to the heating rate are studied only in a qualitative manner by varying the thermal radius,
RT,.

18
 3%c final time char yields and conversion times for the biomass particles are presented in Fig. 12 for thermal
radii in the range I. I 5 s EIT/~,0 < 10. me lwwst effect of the heating rate is observed in the variations
of the conversion times for both cellulose and wood. In agreement with previously observed trends (Antal and
Mok, 1990), an increase in the heating rate dccrcascs both the char yield and the conversion time. When tic high
temperature source (at r = RT) is moved closer to the particle surface, t. changes as a result of two competing
processes: the increase in 2!> due to the increasing heating rate, and the decrease in temperature due to endothermic
reactions. 3%c reduced conversion times imply that it is the first process which dominates in the small R7,/1~
limit. Ilowcvcr, the char yields only show mild relative decreases for the smallest RT values, indicating that the
effective pyrolysis temperature is only raised by a relatively small amount for the present parameters. E:vcn for
the most rapid conversions, the efficiency ratios indicate that diffusion effects cannot be neglected in the accurate
simulation of the particle pyrolysis considered here.
An additional interesting result suggested by Fig. 12 concerns the free stream boundary placement. ~le variation
of }/1, indirectly gives a measure of the distance from the particle at which the computational boundary must be
placed in order to correctly simulate particle pyrolysis witl)in an infinite domain. Numened simulations of liquid
droplet combustion generally require external droplet resolutions greater than 25 times the initial droplet radius in
order to eliminate effects due to the boundary placement (see Harstad and Behn, 1991). Figure 12 indicates that
a much smaller domain can be considered in the ease of solid particle pyrolysis. In fact, only very small variations
in both char yield and conversion time are observed for values of ~/T > 5}~,0. 31e differences between pyrolysis
and liquid droplet simulations are due to the nature of the processes. Liquid fuel combustion generally involves
initial evaporation followed by seeond order rcaetion kinetics with rates which are mixing (diffusion) controlled.
Boundary placement near the droplet will significantly increase the mixture fraction of free stream spccics at the
droplet surface. In addition, exothcrmicity of liquid fuels is generally much larger than the pyrolysis gas phase
reactions, resulting in substantial thermal expansion effects relatively far from the droplet. The trends observed
for both cellulose and wood particles indicate that the value RR == 101~,0 is sufficiently large to make boundary
placement effects negligible for the final yields and conversion times.

3.2.2 fi.jtects of the particle size

It is commercially desirable to pyrolyze large biomass particles as the grinding process necessary to rechrce the
particle size is overly expensive (Di Bkasi, 1993). Kinetic limits arc not necessarily applicable for modeling
biomass pyrolysis in reactors. Simmons and Gentry (1986) investigate kinetic limits for pyrolysis control as a
function of both l~,o and the heating temperature, Using a n~athcmaticd model with prescribed particle surface
conditions they estimated that a 200pnz biomass particle is heat transfer limited for temperatures larger than
775}{. Ilowcvcr, comparisons with experiments indicated limiting values well below the model predictions. This

19
discrepancy was attributed to the neglect of the thermal boundary layer outside of the particle in the model. The
current work addresses the issue of particle size on pyrolysis yields and kinetic cfllcicncics for relatively large
initial particle sizes and accounts for all inner and outer particle effects.
Figure 13 illustrates the effect of the initial particle size on the final char yields and conversion times. The
initial particle diameter is varied in the range [0.05cnt, 2.0c77t]. Both the char yield and & arc observed to
bc increasing functions of 1~,0. However, only relatively small changes in char yield arc found for particles
larger than l~,o x 0.4crn. Below this value there is insufficient particle mass for the endothermic reactions to
overcome heat diffusion from the sunoundings. Therefore, the particle is pyrolyzed at higher effcctivc heating
temperatures. For particles larger than this limit, endothermicity maintains an approximately const<ant effcctivc
heating tcrnpcrature in agreement with the results of Fig.9. For the given reactor conditions, the optimal particle
size for char nlaximiyation is approximately 0.4crn. Larger particles will produce only relatively small increases
in char yield, at the expense of substantially longer reactor residence times (minutes-~how-s). The data also
reveals that only wood pyrolysis approaches the kinetic control limit within the range of parameters investigated.
71c conversion efficiencies for the smallest initial particles sizes are 0.004 and 0.42 for cellulose and wood,
respectively.

3.2.3 lj’ects of the initial porosity

The initial porosity of the biomass particle can also affect the pyrolysis behavior. Porosity is directly related to the

particular biomass specimen under eonsidcmtion; however, it is not an intrinsic property of the substance. Values
of the porosity in hard woods have been reported in the range 0.71 < E. <0.85 by Magnatcrm et. al. (1992)
using both porosimetry and electron scanning microscopy and may bc smaller for compactified biomass. Results
arc presented in Fig. ] 4 for char yields and conversion times as a function of .sO in the range [0.4, 0.9]. This
range of values is sufficient to capture the pertinent characteristics of a variety of viable pyrolysis materials, yet is
sufficiently large to remain within the bounds of the assumptions used in the model derivation. The total nm.ss of
each particle remains const~ant due to the prescribed values for the initial partial densities of cellulose and wocd
(Table 2). Figure 14 reveals several interesting influences of EO. The char yields for both cellulose and wood
rcrnain nearly constant over the entire range of considered porositics due in part to the fixed initial particle mass.
On the other hand, t, shows an increasing trend with initial porosity for both cellulose and wood. Ihcreforc, the
effective heating rates of the particles are being dccrcascd with increasing ~o. This can be explained in terms
of the thermal conduction model employed in tic internal energy equation. Ihe cffcctivc conductivity Aeff is
modeled in terms of a parallel conduction model corresponding to volume averaging over the respective solid
and gas phase ~nductivities. qhblcs 2 and 3 show that the mnductivity for the solids is in gene~ an order of
magnitude larger th,an that of the gases. ~hcrcfore, the parallel rnodcl results in relatively smaller cffectivc heat

?0
diffusion into the particle for increasing CO. A mass averaged conduction model would be expected to show m-y
little effect of co on L, Further experimental results are needed in order to ascertain which type of conduction
model is more realistic. Expcnmcntdly dcnved plots of the char yield and t~ versus e may indicate the correct
modc]ing approach for the effcctivc conductivity.

3.2.4 Ayicts of the iniiiol particle temperature

Onc method of testing the reliability of results in experiments is to prc-treat the biomass before analysis. Varhcgyi
and Antal (1989) employed thermal prc-treatment in addition to investigating the effects of catalysts (lVuCl,
J’cS04, Z?i2’12) on Aviccl cellulose pyrolysis. Although the presence of the caki ysts ajtercd the reaction kinetics,
prc-treatment of the pure spccimcn at 5351< for 1 hour did not produce significant changes in the observed
evolutions. Other expcrirncnts have addressed the issue of ambient pressure and its effect on the pyrolysis. For
cx.arnplc, Richard and Antal ( 1992) were. able to increase char yields from 60/0 to 410/0 by raising the pressure
from 0.1 A41’a to 1.O&fPa for the pyrolysis of cellulose in a packed bcd reactor. 7%c present CC11U1OSC and wood
kinetics schemes do not allow for the investigation of catalyst thermal prc-treatment duration or pressure effects.
I lowcvcr, the value of 7~,o can be expected to affect tC, and possibly the char yield. Figure 15 illustrates these
effects by varying l~,o in the range 300K s 7&. < 600K. It is evident that 7~,o dots not significmt]y affect
the effective pyrolysis temperature as both CC11 U1OSC and wood char yieJds remain ncady constant. Conversion
times arc decreased with increasing Tp,o, as expected. However, the time required to raise the p,artic!c tcrnpcrature
from 300K to 600K, as estimated from the difference in conversion times, is approximately 100s. This value
is significantly less than the total conversion time and gives further evidcncc that it is the endothermicity of the
solid phase reactions which primarily governs conversion times and, therefore, on effective heating rates.

3.3 Comparison with experiments

7 ‘he simulations presented in this work have not been conducted with the intent of representing either the conditions
or prmunctcrs of specific experiments. }Iowcver, comparisons with p,ast laboratory results arc uscfhl in interpreting
the validity of the particle model, as well as the cellulose and the wood kinetics schemes. The minimum
requirement from a model or kinetic scheme is that it should predict the total char (and therefore the vapor)
yields and also the required time for complete conversion. 10 the author’s knowledge, there arc no experiments
with measured yields ar~d/or conversion times for “large” isolated ccllulosc or wood particle pyrolysis. This is

because it is advantageous to consider very fine particle. sizes (kinetically controlled pyrolysis) when mm.suring
reaction rate parameters. Ncvcrthclcss, by comparing simulations with available data for a variety of conditions
,and materials, it is possible to qualitatively evaluate both the particle model and kinetics schemes. The two
kinetics schemes addressed in this work place limits on the total possible char yields and conversion times (SCC
Fig.7). Limiting the discussion to TIZ in the range [550K, 1100K] (ecm-esponding to total pyrolysis time sedes
less than N 106s), then the possible available char yield must be less than R 0.2 for the ceilulose scheme, whereas
the wood scheme limits char yields to a range of approximately (0.27, 0.32). Any experimental evidence for char
yields outside of these ranges would indicate an inconsistency in the modeled pyrolysis kinetics.
In Fig. 16 comparisons are made for the total char yield as a function of 7’R for the baseiine case eeiluiose and
wood p,articlcs with the results of seven experiments. Two different sets of measurements conducted with pure
CC11 U1OSC arc included. Shaikadeh et. al. (1979) performed experiments using ecllulose powder in a vacuum tube
f~lmace at relatively low temperatures. The very fine powder sizes in addition to the low heating temperatures
suggests that the pyrolysis is nearly completely kinetically eontroilcd. In agreement with this hypothesis, the
experimental data compare very well with the kinetic limits of the present mode]. Cellulose pyrolysis in both
fluid bed and entrained flow rcaetors was studied for a larger temperature range and particle size (l~)o R 50pm)
by Scott et. al. (1988). As may be expected, the lager particle size shows slightly more char yield than dots
the vacuum pyrolyzed powder. The data reveal that the slightly larger particles employed by Scott ef. al. still
pyrolyze near the kinetic limit. The much larger particle size (l~,o == 57nm) considered in our simulations is
dominated by diffusion limitations and, as such, shows larger char yields than either experiment. However, the
smallest cellulose particle considered in the current work has q,. == 0.25ntm and was pyrolyzed at 1’H Z= 900K
(see Fig. 13). In this case, the final char yield is approximately 0.02; comparable to the data of Scott ct. al. (1988).
Both sets ofexperirnentrd cellulose char yields are consistent with the limits imposed by the model kinetics scheme
(< 0.2). The cellulose kinetics scheme is thus considered to be in good agreement with expcnments, and the
numerical particle model shows the correct qualitative behavior for diffusion dominated pyrolysis.
Experimental data from wood pyrolysis show that the wood kinetics scheme is inadequate. Figure 16 includes
four sets of experimentally obtained data for wood pyrolysis. Thumer and Mann (1981) measured char yields
from the pyrolysis of Oak sawdust (l~,o w I m771), over a relatively low temperature range, using a furnace
rc.actor at atmospheric pressure. Their results suggest a nearly constant char yield of approxirnatc]y 0.3 over
the temperature range [6021{, 665 K]. Good agrccmcnt with both the wood particle model and the kinetic limit
is observed (Fig. 16). This is not surprising as the primary reaction constants used in the current model were
extracted from the llur-ncr and Mann measurements. The small over prediction as compared to their ciata is a
result of the secondary char producing reaction (~s) which was not originally included in their kinetics scheme.
l%c remaining wood pyrolysis data are for Ibplar wood and olive husk rnea.nrrtd in a semi-batch moving bcd
reactor by M&$chio (1992), and for Eastern Red Maple pyrolyzed in both a fluid bed and a transport reactor by
Scott et. al. (1 988). All of these expcrirncnts used relatively small par-ticlc sizes in order to rcrnain near the
kinetic control limits, Relatively good agrccmcnt is found bctwccn the experiments and the model for only the

22
lowest temperatures ecmsidered (near kinetic control). However, all of the experiments display a much sharper
reduction in char yield with increasing reactor temperatures than either the particle mcxicl or the kinetic scheme
is cap~blc of producing. Analysis of Fig.7 indicates that the wood kinetics scheme is not capable of reproducing
the lower experimental data for any reactor temperature. These results reveal a major flaw in the present wood
pyrolysis scheme in predicting char yields.
Although not included in the data of Fig. 16, the present model results ean be used to make a qualitative
assessment of the modified wood kinetics scheme used in both Di Blasi (1993a) (paper 1) and Di Fllasi (1992)
(paper 11). The modified wood scheme is essentially the same as the present wood kinetics scheme execpt that
the original primary reaction frequency conskwts presented in lhrmcr and Mann (1981) were modified by the
author (apparently increased by a factor of 3600) (Di 131asi, 1996a). Both papers I and II provide tabular data
for char yields at a time when 10!4o of the original particle mass remains, and with 2’~ ~ 1100K. In both cases,
nearly constant char yields are reported with very little deviations over a relatively wide range of pammcters. The
respective mean char yields are approximately 0.41 and 0.49 for papers 1 and H. lIc discrepancy between char
yields obtained with apparently the same reaction sclicme lias not been explained. I/i addition, the final yields
are even higher when the remaining particle mass is consumed. Since a comparison with Fig. 16 shows that these
char yields are exuedingly high, it seems that the modified scheme of Di Blasi is of questionable merit.
Quantitative comparisons with experimentally measured ecmversion times are diflicult to make. I%is is par-
ticularly true since such reported data arc usually in terms of particle residenee times in the reactor. Even when
cornplctc pyrolysis conversion is aehievcd, the final times are not generally tabulated. However, some data exist
which show that the present model is in approxitnate agreement with experiments. Figure 13 illustrates that the
smallest cellulose particle (~,0 = 0.25rr1n~) ecmverts completely in 1.6s at a reactor temperature of 71t == 900K.
qlis value compares favorably to the 0.7s residence time for an Avieell cellulose particle of initial radius of
l~,>o = 0.3nlnz in a 973K reaetor measured by Scott et. al. (1988). Oak sawdust pyrolysis at temperatures
627K ~ 1 it ~ 665K occurs on total reaction time scales bet wecn 23 and 31 minutes according to Tlmmer
and Mann (198 1). Again, these values arc similar to the present simulation results as depicted in Fig.6 for
slightly larger particles (= 2000s). Ihesc eornpansons suggest that the present particle model is able to reproduee
expcrirncntal conversion times for both cellulose and wood particles with a reasonable dcgrcc of accuracy.

4 CONCI,USIONS
A mathematical model is presented which is capable of modciing both the temporal and the spatial evolutions
of porous solid particle reactions for which volumetric reaction rate kinetics are known a priori. ~~c model is
derived for conditions in which both the porosity and the permeability arc assumed large enough to allow for

23
continuous gas phase flow within the particle, and is valid in both the interior and the exterior particle regions.
Although the equations arc relatively intensive cmmputationally, they provide an accurate representation of particle
reactions which may be used to evaluate various kinetics schemes for detailed comparisons with cxpcrimcnts and
previous models.
qlc n~athematieal model is used to simulate the pyrolysis of spherical biomass particles in initially quiescent
super-heated steam environments by considering previously published kinetics schemes for both cellulose and
generic wood reactions. As many as five eompcting reactions arc employed, including complctc property varia-
tions, thermal and mass transport, and both endothermic and exothermic heats of reaction. Numerical solutions
to the modeled equations are used to illustrate the effects of the reactor temperature, heating rate, porosity, initial
particle size and initial particle temperature on both char yields and conversion times. Solutions for “baseline
case” particles reveal that the modeled equations are capable of reproducing the qualitative evolutions of the
pyrolysis process as observed in past cxpcrimcnts and models.
In general, the variation of any parameter which produces an increase in the cffcctivc pyrolysis temperature will
rcducc both the total char yields and the conversion times for both cellulose and wood particles. Such parameter
changes include; 1) increasing the reactor temperature, 2) increasing the heating rate, or 3) decreasing tic initial
particle size. However, practical limits on n~a~inmm effective pyrolysis temperatures are imposed by the relative
endotherrnicity of the interior pyrolysis reactions (as related to the particle mass), and also to a Icsser extent by
thermal diffusion and heat capacity effects. As the particle is heated from its initial value, three stages of evolution
are observed. First, an initial rapid heating during which only negligible reactions occur. This is followed by
a period of primary pyrolysis which occurs at relatively constant effective pyrolysis temperatures determined by
the relative endothcrmicity. lle final regime is reached when the particle mass has been reduced sufficiently to
allow the inward them~al diffusion to overcome endothermic heat absorption. The remaining particle mas is then
rapidly heated until pyrolysis completion occurs, generally at temperatures less than the reactor tcmpcraturc. Both
the mean particle temperature and the heating rate are observed to be almost identical when time is normalized
by the final conversion time. In ail cases considered, the three pyrolysis regimes correspond to approximately
20!4., 60% and 20!40 of the total conversion time, respectively.
The simulated results show that models which neglect the exterior particle thermal boundary layer can sub-
stantially over predict both the pyrolysis rate and the experimentally attainable tar yields. Variations of either
the initial porosity or the initial particle temperature primarily affect the conversion time, whereas only relatively
minor deviations in the total char yields are obwr-vcd. An incrc,ase in the mnvcrsion time with increasing porosi-
tics is attributed to t.hc parallel thermal conduction model employed for the multi-phase heat transfer within the
particle. This trend needs further validation through future exq)crimcnts. In almost all of the particle simulations

24
considered, the kinetically controlled assumption proved to be invalid,
A qualitative comparison with experimental results from several investigations is provided. Ilesc comparisons
indicate that the CCI1U1OSC kinetics schcmc provides a relatively accurate prediction of char yields and conversion
times over a langc range of reactor tcmpcraturcs. Ilc wood pyrolysis kinetics display fhir agrccmcnt with ex-
periments for very low reactor temperatures. However, at higher reactor tcrnpcratures the wood kinetics schcmc
substantially over predicts the char yields and is inconsistent with the expenmcntrd me.asuremcnts. It is su~gested
that signitic.ant improvements be made to the wood pyrolysis kinetics schcmc before any practical predictions arc
attempted. Although the experiment.al data were performed for very small particle sizes in order to exhibit kinet-
ical ly controlled rates, comparisons with the present particle model have indicated a good qualitative agreement
for diffusion limiting particle dynamics. Future work will be aimed at expanding the model in order to portray a
variety of reactor conditions, including turbulent diffusion, wall effects, and free stream convection. In addition,
a kinetics scheme capable of reproducing experimental wood pyrolysis results is presently under investigation.

ACKNOWLEDGNfEN~’S

This research was conducted at the Jet Propulsion Laboratory (JPL) and sponsored by the U.S. Department of Energy (DOE),
with Mr. Neil Rossmcissl (DOE Headquartcm) and Mr. D. Hooker (DOF, Golden Center) serving as contract monitom, under
an agreement with the NatiorM Aeronautics and Space Administration, Computatioml resources are provided by the super
computing facility at Jt%. T’hc authors wish to thmrk Mr. Jim DicboId and Dr. Estaban Chornet of the National Renewable
Energy 1,aborwory for helpful discussions, and also Dr. Kenneth Harstad of JPL for advice on the numcncal method.

25
REFERENCES

Antal, M. J. and Mok, W. S. L. Review of methods for improving the yield of charcoal from biomass. Energy
and ~uek, 4(3):22 1–225, 1990.
Antal, M. J. and Varhegyi, G. Cellulose pyrolysis kinetics: The current state of knowledge. M Eng. Chem.
Res., 34(3) :703-7 17, 1995.
Bear, J. Dynamics of l“Iuids in Porous Media. American Elsevier, Ney York, New York, 1972.
Bcllan, J. and Cuffcl, R. A theory of nondilute spray evaporation based upon rnultiplc drop interactions.
Comb. ond J+lome, 51:55-67, 1983.
13hatia, S. K. and Per]mutter, D. D. A random pore model for fluid-solid reactions: I. Isothcm~al, kinetic
control. AIChE Journal, 26(3):379-386, 1980.
13hatia, S. K, and Perlmutter, D. D. A random pore mode] for fluid-solid reactions: 11. Diffusion and transport
effects. AICM< Journal, 27(2):247–254, 1981.
Bilbao, R,, Millcra, A. and Munllo, M. B. Heat transfer and weight loss in the thermal decomposition of
large wood particles. ]n A. V. Bridgwater, editor, Advances in Thermochemical Biomass Conversion, volume 2,
pages 833-845. Blackie Academic and Professional, New York, New York, 1992.
Bradbury, A, G., Sakai, Y. and Shafizadch, F. A kinetic model for pyrolysis of cellulose. J. App. Polymer
Ski., 23:3271-3280, 1979.

Guam, H. S. and Amundson, N. R. Diffusion and reaction in a stagnant boundary Iaycr about a cwbon
particle. lnd Dng. Chem. Mmdamentals, 16:171, 1977.
Chan, W. R., Kelbon, M. and Kneger-Brockct~ B. Correlations of reaction products with process conditions.
lnd. Eng, Chem. Ales., 27;2261-2275, 1988.

Di Blasi, C. Modeling of trrmspofi phcnonlcna and kinetics of biomass pyrolysis. In A. V. Bridgwatcr, editor,
Advances in 7hernlochemicaI Biomass Conversion, volume 2, pages 906921. Blackie Academic and Professional,

Ncw York, Ncw York, 1992.

Di Blasi, C. Analysis of convection and se~ndary reaction effects within porous solid fuels undergoing
pyrolysis. Combust. Sci. and lech., 90:315-340, 1993a,
Di Blasi, C. Modeling and simulation ofconlbustion proccssx Of charring and non-ch,arring solid fllcls. }’rog.
Mcrgy Combusf. Sci., 19:71-104, 1993b.
l)i I~lasi, C. Numerical simulation of CC1lUIOSC pyrolysis. Biomass and 13ioener~, 7:87–98, 1994.

IIi Blasi, C. Personal communication, Feb. 1996a.

26
 Di Blasi, C. Kinetic and heat transfer control in the slow and flash pyrolysis of solids. Ind. Eng. Chem &s.,
35:3746, 1996b.
Dicbold, J. P. Ablative pyrolysis of macroparticles of biomass. In l+oc. SPC. Workshop Ihst Pyrolysis of
Biomass, pages 237–25 1, Copper Mountain, Colorado, Oct. 1980. Solar Energy Research Institute.
Dicbold, J. P. and Power, A. Engineering aspects of the vortex pyrolysis reactor to produce primary py-
rolysis oil vapors for usc in resins and adhesives. In A. V. Bndgwater and J. L. Kuester, editors, R?search in
lhermochemicol Biomuss Conversion, pages 609-628. Elsevicr Applied Science, Ncw York, New York, 1988.
Ciwal,as, G. R. A random capilla~ model with application to char gasification at chemically controlled rates.
AIChE Journal, 26(4):577-585, 1980.
Gavalass, G. R. Analysis of char combustion including the effect of pore enlargement. Comb Sci. and ?imhnol.,
24:197-210, 1981.
Ilarstad, K. and Bellan, J. A model of the evaporation of binary-fuel clusters of drops. Alom. and Sprays,
1:367-388, 1991.
Kothari, V. and Antal, M. J. Numerical studies of the flash pyrolysis of cellulose. Ftiel, 64:1487--1494, 1985.
Ledc,J., panagopoulos, J., Li, H. Z. and Villermaux, J. Fast pyrolysis of wood: Direct measurement and
study of ablation rate. lhel, 64:15 14–1520, 1985.
Lim, K. S., Zhu, J. X. and Grace, J. R. Hydrodynamics of gas-solid fluidization. ~iiel, 21(Suppl.):141-193,
1995.
Locwcnberg, M., Bellan, J. and Gavalas, G. R. A simplified description of char combustion. Chem. Eng.
Comm., 58:89-103, 1987.
Magnaterra, M., Fusco, J. R., Ochoa, J. and Cukiem~an, A, L. Kinetic study of the reaction of different
hardwood sawdust chars with oxygen. Chemical and structural characterizzrtion of the samples. In A. V. Bridgwater,
editor, Advances in Ihermochemicul Biomass Conversion, volume 1, pages 116130, Blackic Academic and
Professional, Ncw York, New York, 1992.
Maschio, G., Lucchcsi, A. and Koufopanos, C. Study of kinetic and transfer phenomena in the pyrolysis
of biomass particles. In A. V. Bridgwatcr, editor, Advances in thermochemical Biomass Conversion, volume 2,
pages 746-759. 131ackie Academic and Professional, New York, New York, 1992.
MoI], E. and Amundson, N. R. Diffusion and reaction in a stagnant boundary layer about a carbon particle .
2. An extension. lnd. Dng. Chem. )imdamentals, 17:313, 1978.
Narayan, R. and Antal, M. J. qlcrmal lag, fusion, and the compensation effect during biomass pyrolysis.
lnd. l{ng. Chem. &s., 1996. submitted.
Reed, T. B., Diebo]d, J. P. and Desrosicrs, R, Perspectives in heat transfer requirements and mcch<anisms for

27
f,ast pyrolysis. In Ymc. Spec. Worhhop Fast Pytdysis of Biomass, pages 7–20, Copper Mountain, Colorado, Oct.
1980. Solar Energy Research lnstitutc.
Richard, J. P. and Antd, M. J. l’hcrmogravimctric studies of charcoal formation from cellulose at elevated
pressures. in A. V. Bridgwater, editor, Adwmces in lhermochemicol Biomass Conversion, vohmc 2, pages
784-792. E31ackie Academic and Professional, Ncw York, Ncw York, 1992.
Scott, D. S., Piskorz, J., Bergougnou, M. A., Gr-sham, R. and Ovcrend, R. P. The role of temperature in the
fast pyrolysis of cellulose and wood. Ind. Eng. Chem. Res., 27:8-15, 1988.
Shafizadch, F., Furneaux, R. H., Cochran, T. G., Scholl, J. P. and Sakai, Y. Production of lcvoglucosan and
ghrcosc from pyrolysis of cellulosic materials. J. Appl. Polym. Sci., 23:3525-3539, 1979.
Simmons, G. M. and Gentry, M. Particle size limitations due to heat transfer in determining pyrolysis kinetics
of biomass. J Anal. and Appl. Pyrolysis, 10:117– 127, 1986.
Sotirchos, S. V. and Amundson, N. R. Diffusion and reaction in a porous char particle and in the surrounding
gas 2. l,imiting model. lnd. Dng. Chem. Jimdamentols, 23:180, 1984a.
Sotirchos, S. 1’. and Amundson, N. R. Diffusion and reaction in a prous char puticie ami in the surrounding
gas. a continuous model. Ind. Dng. Chem. Jimdamen[als, 23:191, 1984b.
Sundaresan, S. and Amundson, N. R. Diffusion and reaction in a stagnant boundary layer about a carbon
particle 5. Psucdostcady state structure and parameter sensitivity. Ind. Ilng. Chem. Fundamentals, 19:344, 1980.
Thurncr, F. and Mann, U. Kinetic investigation of wood pyrolysis. Ind, Eng. Chem. Process Des. Dev.,
20:482-488, 1981.
Varhcgyi, G. and Antal, M. J. Kinetics of the thermal decomposition of cellulose, hemicellulose, and sugar
c.anc bagasse. Energy and Fuels, 3:329–335, 1989.

28
TABL13S

. . .
—-. -.—
R e a c t i o n - ‘ A~ [ 1 / s ] El [kJ/k~tol] A hi [k.l/kg]--
.. —.—.
K, 2.8 X 103T 2.424 X 1-~ o
K2 1.3 x 1010 1.505 x 107 -418
K3 3.28 X 1014 1.965 X I@ -418
K4 4.28 X 106 1.08 X 105 -t42
KS 1.43 x 10 4
8.86 x 104 -418
KG 4.13 xlo~ 1,127 X I@ -418
K~ 7.38 X 105 1.065 X I@ -418
—. ]<S
. . . . —.— 1.OX I@ 1,08 X 105 +42
—.. — —.———— ——
Table 1: Reaction parameters.

—
—....———..——.-——.——-——
Spccics (1 - - eo)p [~fi–-~ [~$1 ~ [R!$%l_
–cellulose 420 2.3 2.426 X 1 0- r
active 420 2.3 2.426 X 1 0-4
char (420,650)” 1.1 1.046 x10-4
wood 650 2.3 1.046 x10-4 —
age— —.. —— __ ._. .D.,m n -.=-,-.—_mv. _—
Table 2: Property values for the solid phase species. ~le superscript * indicates assumed values.

-—— ____
.——_— -. ——
S p e c i e s &f~ [*] CV [~--] ~) @r#- ‘“
.._2:__.___P [:9;1 = ~ [*RI
--5
H20 18.016 2.20 2,9 X 10-5 7.8 “—X 1 0 1.1 x IF-”
gas 18.016’ 1,1 3.0 X 1 0 2 . 5 7 7 X 1 0 1.1 X 1 0- ” 4 ’
-5 -5

tar 18.016* 2.5 3.0 X 1 0-5 2 . 5 7 7 X 1 0-5 1.1 X 1 0- 4*

~hble 3: Property values for the gas phase species. The steam values are given for 1’ == 800K
and p == 100kl’u, and the superscript * indicates assumed values.

~.. . — . . -—
Reaction char limit conversion limit [s] -
-
cellulose 0.35~z(Kz +- K3)=T---’ -- IOg(o.ool )(K2 + K3~-~— ‘“
baseline: 6.4 x 10 -3 baseline: 5.3 x 10’”3s
wood K7 4 KG ~~A5} ( K5 -t KG -{ K7)- 1 -. log(0.001)(K5 + ]{6 + 1{7)- 1
{
L— —. .- baseline: 0.29 baseline: 3.9s . .
~hblc 4: Kinetic litnifi for char yields and conversion times. Ilc baseline values correspond to
limits evaluated at 7’= !300K.

29
FIGURE CAPTIONS

F’igurc 1: Reaction schcmcs: (a) CMUIOSC pyrolysis, (b) wood pyrolysis.

Figure 2: Temporal evolution of the partial char density for the baseline case simulations, (a) CC1IU1OSC, t == 53s,
116s, ]8~S, 254s, (b) wood, t = 89s, 181s, 268s, 348s.

Figure 3: Temporal evolution of the tar mass fraction for the baseline case simulations, (a) CC11 U1OSC, t == 53s,

1 ]6S, 187s, 254s, (b) wood, t == 89s, 181s, 268s, 348s.

F’igurc 4: ~ernporal evolution of the temperature for the baseline case simulations, (a) cellulose, t = 0s, 53s,
116S, 18~S, 254s, (b) wood, t = os, 89s, 181s, 268s, 348s.

Figure 5: Temporal evolution of the gas phase velocity for the baseline case simulations, (a) CCI1U1OSC, t == 53s,
116s, 187s, 25~4s, (b) wood, t = 89s, 181s, 268s, 348s.

l’igurc 6: Char yield and conversion time as a function of the reactor temperature for &o == 500~, ~~~ = 5%,0,
1<3,,0 =. 5mni and e. == 0.7.

Figure 7: Kinetic limits for char yield and conversion time as a function of the reactor temperature.

l;igurc 8: Kinetic efficiency for char yield and conversion time as a function of the reactor temperature for
111, = 5RP,0, l~,o == 5mm, co = 0.7 and 7~,o =- 500K.

FiSurc 9: Mass averaged particle temperature < 7; > as a function of tinlc normali~cd by & for ~~,o = 500K,
1/1 = 51+,,., l~,o == t%nrn and e. =. O. 7; (a) cellulose, (b) wood, lhc reactor ternpcraturc is varied from
700K < 7}t < 1200K in increments of100K.

Figure 10: Wmporal denvativc ofthc mass averaged particle temperature < l; > as a function of time normalized
by t, for 7~,o == 500K, 1{7 == 51~,o, l~,o == 5mm and EO = 0.7; (a) CCHU1OSC, (b) wood. ~hc reactor tewcraturc
is varied from 700K < 1}{ < 1200K in increments of 100K.

Figure 11: M~~imun~ normalized radial position at which the tar fraction has decayed to 5?4. and the maximum tar
fraction as a function of the reactor temperature for both CCIIU1OSC Md wood ~d for ~~,o = ~OOK, ~~~ = 5~%o,

30
Figure 12: Char yield and conversion time as a function of the normalized thermal radius for 7’R == 900K,
14,,0 = 5nM)L, C(j = 0.7 and 7;,0 = 5 0 0 K .

Figure 13: Char yield and conversion time as a function of the initial particle radius for 7k = 900K, RT = 5~,0, }
&o =-- 0.7 and ~~,o = 5001<.

Figure 14: Char yield and conven.sion time as a function of the initial porosity for 7;{ = 900K, 1/1 =- 51$,.,
l~,o = 5nmi and TP,o = 500K.

Figure 15: Char yield and conversion time as a function of the initial particle temperature for 7}? == 900K,
~~~ .- 51$,., ]&,. == 5nim and FO = 0.7.

Figure 16: Comparison of final char yields as a function of the reactor temperature with experiments. Both the
kinetic limits and simulation results arc shown (}lT == 5}+,., 1+,0 == 5mnl, E. =- 0.7 and 7~,o = 500 K): Poplar
wood (()) and Olive husk (.) from Maschio et. al. (1992), Avicell CC1IU 1 OSC (U) and Eastern red maple (@)from
Scott ct. al. (1 988), Oak sawdust (A) from ‘IIumcr and Mann (1981) and cellulose (0) from Shafizadch et. al.
(1979).

31
 K
TAR ~ GAS

K3
I
K]
CELLULOSE——————— ACTIVE

I K2

CHAR + GAS
GAS

CHAR
 CHAR DENSITY [kg/m3]
0
0
0 I
200

150

100

50
,\
time

() 0.0——LA_A4
0.5 1.0 1.5 2*()
rfl?p ~
 TAR MASS FRACTION
o 0 0 0 0 0
0 “+ N- w 4A Cn

I?
TAR MASS FRACTION

I 1
 \ /1//

b 500 J

r/Rp ~
900 L

800

700

600

500 r

400” ~
o12345
r/Rp , ~
5

2
\ time
1

0
o 2 3
r/Rp ~
1.5

1.0

0.5

0.0
0 1 2 3
.

L-)
0.4
c. time w. yield n
:/ o
./
I
1 z
0.3
3
10
0.2 \ “\
\\
\ ~,
\ \ ● \, w. time
\\ . .
●*
● ‘. --* x“
0.1 .* ‘.. -- --- --- -------- -
--- --- -----
>
c. yield I , I
O.$OO , I * 1 , m
10
2

750 1000 1250 1500

REACTOR TIWIPERATIJRE [K]
 6
0.4 10
5
10
4
10
0.3 103
102
101
0.2
10°
1
10-
2
0.1 10-
3
10-
4
10-
5
0.0 10-
500 750 1000 1250 J 500
REACTOR TEMPERATURE [K]
1.0 10°

0.8 10-
1
\ t
I \
\

\\
1
1 \ c. time
0.6
\
\
\\ \\
\\/. \\ 10“2
1- \ \\
0.4 \ \
\
.. \ w. time J
\ \
\ \ /
\
\ \ 3
\
\
\
\
\
10-
0.2 \
\
\
:/\ \\\ \

c. yield “~, \
\
- - ” 10-4
o.% O O ’ ’ ’ ’ ’ ’ - ’ ’ ’ ’ ’ ’ ’ ’ ’ ’ - 1500
750 1000 1250 ‘
REACTOR TEMPERATURE [K]
550 “

I
500
0.0 0.2 O*4 O*6 O*8 1’ 0
NORMALIZED TIME
&’ 800 ~
750 :

700 :

650 :

600 :

550 :

500
0.0 0.2 0.4 0.6 0.8 1.0
NORMALIZED TIME
10

0
0.0 0.2 0.4 0.6 0.8 1..0
NORMALIZED TIME
 [l .l. 11, Al,,,,.,,,,,,,,,,,
0.0 0.2 0.4 0.6 0.8 loo
NORMALIZED TIME
 I
1. I
I
s

8m
I c. frac. 0.5
s8 I
88
III 0.4
IIII \
0.3

*. *. -. 0.2
:. .
● .-.
- ‘--~x%w+---- --,
/ 0.1
w. radius
L
\
o 1 m1 ,
*
10.0
1 , 1 I

500 750 1000 1250 1500

REACTOR TEMPERATURE [K]
0.4 400
w. time
>+..
*.-
-.--------”--------”
4
4
0.3 0
w. yield /
#
300
1#
8

---------- ----
I) I ~# ● --- ..- ----
@-
0.2 #4 ~ }
c. time 200
tt

,c. yield
0.1 100
-
m

0.0 0
O 1 23 45 67 8 910
THERMAL RADIUS: R../R.I .W() 1
0.4

0.3
800

0.2 500

400
0.1
200

O.OO 0 n
● 0.2 0.4 0.6 0.8 1.0 “
INITIAL PARTICLE RADIUS [cm]
0.4 w. time 400
----,
----
------ ----
\
---- ------ ---

0.3 w. jield 300

---- *,
------
---- ------ ----- -----
+
0.2 ‘c. time 200

c. yield
0.1 / 100

, 1 1 I
0.0 m , , 1 , I I ,
0
0.4 0.5 0.6 0.7 0.8 0.9
INITIAL POROSITY
0.4 500

1
w. yield
-.
\\
-. *--
--- ---- 400
0.3 / ----
w. time -. -.*
‘.. ---
------- - ------ --- . 300
oa~ ----- -.. ---
/ ----
c. time
c. yield
0.1 /
i
0.0
300 400 500 600 “
INITIAL PARTICLE TEMPERATURE [K]
 0.4

.-% - - - - - - ., - - wood
0.3
la8 ●
08
--’ --

4
I
w.’ limit
------- -m

t c. limit
u 0.1 Ilu

0.0
600 700 800 900 1000 1100
REACTOR TEMPERATURE [K]