-
Kinetic Modeling of Cellulosic Biomass Processing Featuring
Enzymatic Hydrolysis with Anticipation of
Incorporation into a CFD Framework (Revised title)
* Corresponding Author
Funding from grant No. 60NANB1D0064 from the National Institute
of Standards and Technology
Xiongjun Shao, Zhiliang Fan, Colin HebertLee R. Lynd*, Charles
E. Wyman*
Thayer School of EngineeringDartmouth College
André BakkerFluent Inc.
Presented at the 2003 AIChE Annual MeetingSan Francisco, CA
Nov 21, 2003
-
Project RationaleBiological Conversion of cellulose biomass to
commodity products
� Sustainable resource supply� Energy security� Rural economic
development
Desirable because of potential benefits with respect to
Cost of overcoming the recalcitrance of cellulosic biomass
� Most costly process step� Least technically mature� Enzyme,
microbially-based processes have outstanding potential
Scale-up
� No experience with full-scale facilities� Limited fundamental
understanding
� Computational fluid dynamics (CFD) is powerful tool for scale
up analysis� A collaborative project with FLUENT inc. recently
initiated
Bottlenecks
01
-
System definition
Cellulose Cellobiose Glucose EthanolCellulase β-Glucosidase
Yeast
Experimental data
� Pretreated wood, peptone yeast extract growth media, 37 ºC�
Genencor CL cellulase supplemented with Novozyme 188 β-
glucosidase� Yeast (Saccharomyces cerevisiae), strain D5A� 1 L
working volume
Simultaneous Saccharification & Fermentation
02
South et al. (1995), batch & continuous feeding
� As above except paper sludge (Fraser Papers Mill, Gorham, NH)
wasprocessed in a lean medium containing 0.15% (v/v) corn steep
liquor and 0.25mM MgSO4.
This study, intermittent feeding
-
� All three features deviate from classical kinetics for soluble
substrates� While various mathematical forms can be used to
describe these phenomena, all
three must be addressed for any broadly applicable model
Essential features of enzymatic hydrolysis models (South et
al)
1) Rate saturation with respect to either substrate or
enzyme(e.g. using Langmuir adsorption, but not Michaelis-Menten
kinetics)
2) Declining reaction rate per adsorbed enzyme with increasing
conversion
3) Particle reactivity changes with concentration &
conversionParticle population model (For a well-mixed, fully
continuous steady state reactor)
RTD
CPDM (Loescher et al)
dttExx t∫ ×=∞0
),()( τ )exp(1),( τττttE −×=
dxxnxn
x ∫ ××=1
00
)(ˆ1 dxxnn ∫=
1
00)(ˆ
03
+−=
s
S
dx
rd
r
xn
dx
xnd
τ0ˆ
ˆ)(ˆ)(ˆ
cmxkxk +−×′= )1()(
-
Outline of South et al. model (perfect mixing assumed)
Rate equations
Cellulose:
Cellobiose:
Glucose:
Cells:
Ethanol:
PSKPPSK
CSKCBCSK
S
CEcmxkSr
/][/
/
/][))1((+
×+
××+−×′−=σ
][)][
1(
][][056.1
/
CBK
GK
BCBKrr
GC
m
CSCB
++×
××−×−=
)][
1(][
][][
/
max
PXG
C
K
P
KG
GXrXc −×+
××=
µ
GXY
rrrr XcCBSG
/
053.1)056.1( −×−×−=
GX
GP
Y
Yrr XcP
/
/×=
Conservation equations
Cellulose:
Lignin:
Cellulase:
S
CESS f σ
][][][ +=
L
LELL f σ
][][][ +=
][][][][ LECEEET ++=
Material balance
Batch:
Continuous:
ir
dt
id =][
])[]([1][
0 iiir
dt
id −×+=τ
i = Cellulose, Cellobiose, Cells, Glucose, Ethanol
04
-
Batch SSF
□: 5 U/go: 10 U/g∆: 15 U/g◊: 20 U/g
Enzyme loading
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (days)
Con
vers
ion
SubstrateDilute acid-pretreatedwood
Curves generated using best-fit parameters to
cmxkxk +−×′= )1()(
k’ = 2.8625 /hm = 5.30c = 0.18125 /h
05
-
Continuous SSF (steady state)
• Predictions based on parameter values obtained by fitting
batch data without adjustment
• Consideration of changing reactivity over the time a particle
spends in the reactor is absolutelyrequired to get agreement with
experimental data for the continuous system
- Experimental data - CSTR prediction
Range of CSTR data, runs 1 through 7
Feeding concentrations: 12.9-foldEnzyme loadings: 2.1-fold
Residence time: 5.6-fold
06
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
# 1 # 2 # 3 # 4 # 5 # 6 # 7
-
Solution Algorithm of South et al model
� Set derivatives = 0 (assumed steady state)
� Simultaneous solution of 5 non-linear algebraic equations
using multiple iterative loops
Limitations
General� Not readily adapted to intermittent feeding� Perfect
mixing assumed (not readily adjusted to imperfect mixing)
Specific to CFD (limited to solving ~50 dynamic equations per
element per time step)
� Particles modeled as ~100,000 discrete populations� Extensive
iterations (~30)� Excessive computational requirements when
implemented on a
distributed basis in a CFD framework (~10,000 computational
elements)
07
In light of these limitations, the solution of South et al. has
to be modifiedto be compatible with CFD analysis
-
Modifications to Kinetic model
Change Equilibrium Enzyme Adsorption to Dynamic Enzyme
Adsorption
08
Accommodate Intermittent Feeding
Dynamic enzyme adsorption
Equilibrium enzyme adsorption
Model prediction: Batch SSF (enzyme loading = 10 U/g)
-
Accommodate Discrete feeding
Particle conversion, xp(i)
)]([
)]([)(
1)]([
)(0
0
iS
iSiR
iSi
px
×=
−
[S0(i)] = g cellulose/L, population i, fed to the reactor [S(i)]
= g cellulose/L, population i, in the reactor @ t
)(iR Fraction of particles of population iremain in the
reactor
=
xp changes due to reaction only
][
)]([[
0
1]0
S
iSSx
n
i∑−
= =
Reactor conversion
Define particle conversion
[S0] = g cellulose/L, fed to the reactor [S(i)] = g cellulose/L,
population i, in the reactor @ t
Changes due to
1. Reaction2. Exit of substrate
x
cmxkxk ipip +−×′= )1()( )()(
xp(i) rather than , is appropriate to use for the conversion
dependent rate constantx
09
-
Accommodate Discrete feeding (continued)
� Particles fed at a given time are modeled as a discrete
population (i)
� Total enzyme = Free enzyme + sum (enzyme bound to each
population)
� Track particle populations until they are highly converted
10
, )]([
]))(1([S
mp
iCEcixkir σ
×+−×=)]([
)]([)(
1)]([
0
0
)(iS
iSiR
iS
x ip
×=
−
Hydrolysis rate
]/[))]([([ 01
]0 SiSSxn
i∑−==
Reactor conversion
Material balancefor component J )]([
)()]([
)()]([0 iJf
tOiJ
f
tIirdt
iJd ×−×+=
=0
1)(tI
t = t0 (original feeding)
i: index of individual particle population
n: total number of particle populationsf : feedings/residence
time
R(i): remaining fraction of particle pop i
=0
1)(tO
removal time (feeding time)
at all other times at all other times
J = Substrate, enzyme, lignin, cellobiose, glucose, cells,
ethanol
-
Comparison of predicted CSTR conversion using the South Solution
Algorithm and the Discrete Solution Algorithm
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
# 1 # 2 # 3 # 4 # 5 # 6 # 7
- Experimental data - CSTR prediction - Discrete model
prediction ( f = 200)
11
-
f (feedings/residence time)
Fully continuousResults( f= )∞
Tau = 96 hrs, Enzyme loading = 10 U/g
Discrete SSF, Steady State (end of cycle) cellulose
conversionvs. feeding frequency (enzyme loading = 10 U/g, = 4 days
)τ
Batch SSF, t = 4 days
12
10
Experiments in this region
-
Experimental system
13
Paper sludge as substrate
-
0.85
0.86
0.87
0.88
0.89
0.9
0.91
0.92
0.93
0.94
0.95
0.96
0.97
0.98
1 2 3 4 5 6 7 8 9 10 11
Feeding frequency, f
Con
vers
ion
Prediction, E = 10.5 U/g Prediction, E = 20 U/g
Data, E = 10.5 U/g Data, E = 20 U/g
Prediction vs. experimental data for Paper sludge
� Lowering f allows conversion to remain constant while reducing
enzyme loading� Model predictions based on parameter values for
pretreated wood� Experimental data obtained with paper sludge
14
Anticipated trends:Lower enzyme cost Lower cost for mixing, heat
transfer
-
CFD analysis of continuous systems requires that the number of
equation solved per element (N) be limited (Current limit: 50
equations)
Number of equations (N)
Equations for n discrete particle population
Additional equations
• Cellulose concentration, i th population, [S(i)]•
Cellulose-enzyme complex concentration, i th population,
[CE(i)]
• Lignin concentration, [L]• Lignin-enzyme concentration, [LE]•
Cellulase enzyme concentration, [E1]• Cellobiose concentration,
[CB]• Cell concentration, [Xc]• Ethanol concentration, [P]• Glucose
concentration, [G]• Carbon dioxide concentration, [CD]
2n + 8
2n
8
15
-
ε(n) = 1 -Quasi-steady state reactor conversion with n particle
populations tracked
Quasi-steady state reactor conversion with >> n particle
populations tracked
Fractional error
N depends on the degree of accuracy required
16
2482810401687.23%20
14318522788.73%8
919112293.63%2
4
48205825783577.10%20
2482810361478.36%8
12214316482.74%2
2
NnNnNn
ε(n) = 2.0%ε(n) = 1.0%ε(n) = 0.2%f(days)
Enzyme loading = 15 U/g
τ x
Equations limit: ~50
For many scenarios, N falls within the practical range for
CFD
-
Status of Modeling Work in Relation to Complexity
17
Not likely to be practical w/ CFDmanyImperfectContinuousN
(staged)10
SolvablefewImperfectIntermittent15
Solvable
In progress
In progress
Not likely to be practical w/ CFD
Done
Done (South)
Done*
Done (South)
Status/Solution expected
fewImperfectIntermittentN (staged)9
fewPerfectIntermittentN (staged)8
manyPerfectContinuousN (staged)7
manyImperfectContinuous16
fewPerfectIntermittent14
manyPerfectContinuous13
1ImperfectBatch12
1PerfectBatch11
PopulationMixingFeeding# of reactorsScenario
Intermittent feeding is advantageous in terms of both
application and computational feasibility
* Presented at 25th Symposium on Biotechnology for Fuels and
Chemicals
-
Summary
18
• Combining kinetic and CFD models for biocommodity processesis
a promising approach for scale-up analysis that has received little
prior attention previously
• SSF model of South et al. has been reformulated to be
compatiblewith requirements for analysis via CFD, reducing the
number of particle populations tracked from 100,000 to < 30 with
little error
• Model results indicate that reduced feeding frequency allows
high conversion to be realized at ~ 2-fold lower enzyme loading
• Experimental results with paper sludge confirm predicted
trend
• Good agreement between experimental and predicted data is
obtainedalthough parameter values obtained for a different
substrate
• Continued development and application of combined kinetic and
CFD models is underway