You are on page 1of 17

Journal of the Air & Waste Management Association

ISSN: 1096-2247 (Print) 2162-2906 (Online) Journal homepage: https://www.tandfonline.com/loi/uawm20

A Fuzzy Composting Process Model

Xiaosheng Qin , Guohe Huang , Guangming Zeng , Amit Chakma & Beidou Xi

To cite this article: Xiaosheng Qin , Guohe Huang , Guangming Zeng , Amit Chakma & Beidou Xi
(2007) A Fuzzy Composting Process Model, Journal of the Air & Waste Management Association,
57:5, 535-550, DOI: 10.3155/1047-3289.57.5.535

To link to this article: https://doi.org/10.3155/1047-3289.57.5.535

Published online: 29 Feb 2012.

Submit your article to this journal

Article views: 428

Citing articles: 29 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=uawm20
TECHNICAL PAPER ISSN:1047-3289 J. Air & Waste Manage. Assoc. 57:535–550
DOI:10.3155/1047-3289.57.5.535
Copyright 2007 Air & Waste Management Association

A Fuzzy Composting Process Model


Xiaosheng Qin and Guohe Huang
Sino-Canada Center of Energy and Environmental Research, North China Electric Power
University, Beijing, People’s Republic of China; and Environmental System Engineering Program,
Faculty of Engineering, University of Regina, Regina, Saskatoon, Canada

Guangming Zeng
College of Environmental Science and Engineering, Hunan University, Changsha, People’s
Republic of China

Amit Chakma
Department of Chemical Engineering, University of Waterloo, Waterloo, Ontario, Canada

Beidou Xi
Chinese Research Academy of Environmental Sciences, Beijing, People’s Republic of China

ABSTRACT membership functions. The results indicated that the un-


Composting processes are normally complicated with a certainties projected in input parameters will result in
variety of uncertainties arising from incomplete or impre- significant derivations on system predictions; the pro-
cise information obtained in real-world systems. Previ- posed FCPM can generate satisfactory system outputs,
ously, there has been a lack of studies that focused on with less computational efforts being required. Analyses
developing effective approaches to incorporate such un- on degree of influence of system inputs were also pro-
certainties within composting process models. To fill this vided to describe the impacts of uncertainties on system
gap, a fuzzy composting process model (FCPM) for simu- responses. Thus, suitable measures can be adopted either
lating composting process under uncertainty was devel- to reduce system uncertainty by well-directed reduction
oped. This model was mainly based on integration of a of uncertainties of those high-influencing parameters or
fractional fuzzy vertex method and a comprehensive to reduce the computational requirement by neglecting
composting model. Degrees of influence by projected un-
those negligible factors.
certain factors were also examined. Two scenarios were
investigated in applying the FCPM method. In the first
scenario, model simulation under deterministic condi- INTRODUCTION
tions was conducted. A pilot-scale experiment was pro- Composting is an effective approach for converting or-
vided for verifications. The result indicated that the pro- ganic substrates to stable humus-like products via biolog-
posed composting model could provide an excellent ical degradation. Its process is affected by a variety of
vehicle for demonstrating the complex interactions that biochemical and physical factors (e.g., constituents, pH,
occurred in the composting process. In the second sce- and water content of solid wastes) and operating condi-
nario, application of the proposed FCPM was conducted tions (e.g. C/N ratio, aeration rate, and reaction temper-
under uncertainties. Six input parameters were considered ature). Although effects of these factors have been stressed
to be of uncertain features that were reflected as fuzzy by many experiments, it is often difficult to gain an in-
depth understanding into their interactions and the asso-
ciated process dynamics. Analytical and/or numerical
IMPLICATIONS simulation models have been widely used for handling
The research outputs demonstrated a number of deviations such complexities. In addition, composting processes are
between model simulation and real composting reactions also complicated with a variety of uncertainties arising
because of existence of a variety of uncertainties arising
from incomplete or imprecise information obtained in
from incomplete or imprecise information. This implies that
real world systems. For example, a measure through ex-
the simulation outputs that are normally obtained through
deterministic composting models (with crisp-value param- periments may be imprecise because of the existence of
eters) can hardly be used to describe all of the possible human-induced, systematic, and/or random errors; some
system responses. The proposed method showed better parameters can hardly be quantified as deterministic
applicability in dealing with such problems and could gen- numbers because of insufficient information sources,
erate satisfactory outputs with less computational efforts complex on-site conditions, and/or varying human pref-
being required.
erences. These lead to difficulties in generating a complete
spectrum of potential modeling results that are crucial for

Volume 57 May 2007 Journal of the Air & Waste Management Association 535
Qin, Huang, Zeng, Chakma, and Xi

proper design, control, and operation of composting sys- regarded as an effective framework for tackling uncertain-
tems. The reasons are as follows: (1) converting the un- ties. 14 With this approach, uncertainties associated with
certainties into recognizable mathematical expressions is the modeling inputs are described by probability distribu-
influenced by availability and characteristic of the uncer- tion, such that the modeling outputs can be characterized
tain information; (2) mapping the uncertainties into sys- as probabilistic information. Most of the previous efforts
tem responses is complicated with process dynamics and focused on the technique of Monte Carlo simulation.16,17
system nonlinearities; (3) controlling a composting pro- Such a method involved repeated generations of pseudo-
cess without considering the uncertainties may lead to values for uncertain parameters drawn from known prob-
deviated or even false operating conditions. Thus, effec- ability distributions, so as to produce probability and cu-
tive approaches that can incorporate such uncertainties mulative distribution curves. In many engineering
within composting process models are desired. practices, there is a shortage of information for establish-
In the past decades, a number of efforts were made in ing probability density functions,18 leading to limited
developing various composting models. Two main ap- applicability of the stochastic methods.
proaches were generally developed: analytical models and Another approach based on fuzzy set theory has been
numerical models. The analytical models were primarily proved to be more practical. It is a method that facilitates
based on mass balance and energy conservation princi- the analysis of systems with uncertainties being derived
ples and/or experiential biodegradation kinetics. For ex- from vagueness or fuzziness rather than randomness
ample, Keener et al.1 developed an analytical model to alone. Thus, uncertainties can be handled in a direct way
examine the interactions among biological and physical without a large number of realizations. The fuzzy model-
factors based on kinetic expressions, where impacts of ing approach has been successfully applied in many fields,
temperature, depth of composting bulk, and biological such as risk assessment, decision analysis, engineering
degradation rate on system performances were investi- design, and optimization.19 –23 For example, Hanss24 pro-
gated; Kaiser2 analyzed the thermodynamics and kinetics posed a so-called fuzzy transformation method and ap-
of composting processes and proposed a model that re- plied it to the simulation and analysis of a model for the
flected mass transfer, heat transfer, and biotransforma- friction interface between the sliding surfaces of a bolted
tion of organic materials; Stombaugh and Nokes3 devel- joint connection under uncertainties. Li et al.17 proposed
oped a model based on Monod microbial growth kinetics; a modified fuzzy vertex (FV) method to reflect parameter
this model was presented as differential equations for uncertainties in simulating contaminant flow and trans-
describing microbial, substrate, moisture, and tempera- port in porous media to examine the fate of petroleum
ture profiles in composting processes. Mohee and White4 contaminants in groundwater.
developed a dynamic composting model to examine bio- Generally, the previous applications of fuzzy arith-
degradation processes based on the knowledge of physical metic were mostly based on the FV method. Uncertain
and chemical reactions. More recently, Bari and Koenig5 parameters were tackled as fuzzy sets, and the subsequent
studied a kinetics analysis of forced aeration composting arithmetical operations were implemented through dis-
processes operated under different aeration modes. Xi et cretizing the membership domain. Uncertainties of the
al.6 developed a two-stage modeling approach for describ- modeling results could then be represented as fuzzy sets.
ing inoculation effects on composting processes. Numer- However, the number of simulation runs could become
ical approaches can also be found in Nakasaki et al.7 and extremely high when many uncertain parameters and/or
Das and Keener.8 Most of the previous studies achieved ␣-cut levels were investigated; this could lead to high
successes in modeling specific portions of the composting computational requirements and hinder applicability of
processes from either a biological or physiochemical such methods. In addition, the traditional fuzzy methods
point of view. In fact, the composting process generally could hardly quantify respective effects of each uncertain
has significant temporal variations because of the dy- parameter to the overall modeling outputs.
namic features of many process components and their Therefore, as an extension to previous efforts and an
interactions. Design of a cost-effective composting system attempt to tackle the above problems, the objective of this
should, thus, be based on modeling efforts from a more research is to develop a fuzzy composting process model
comprehensive point of view. (FCPM) through the advancement of a fractional FV
Another complexity of the composting process is the method and the formulation of an integrated composting
existence of uncertainties in many modeling parameters process model. The FCPM is capable of handling system
and their interrelationships. However, almost all of the dynamics, nonlinearities, and uncertainties. More impor-
previous models were deterministic ones. The main ap- tantly, it is useful when the related information is not
proaches for process simulation under uncertainty in- readily available for supporting uncertainty analyses of
clude interval, stochastic, and fuzzy models.9 –12 In the the complexities through conventional stochastic meth-
interval analysis approach, each of the uncertain param- ods. The objective entails the following: (1) the develop-
eters is assumed to have upper and lower limits without a ment of a fractional FV method for addressing uncertain-
probability structure. This method is especially useful ties represented in deterministic model parameters; (2)
when system uncertainties were highly unknown. It has the attempt to integrate biological kinetics, mass conser-
been widely used in optimization problems of water qual- vation, water/energy balances, and oxygen content into a
ity planning and solid waste management.13–15 Probabi- general framework, under uncertain system conditions,
listic methodology is another measure to address uncer- which will result in the development of an innovative
tainties. It has been widely used in environmental integrated, dynamic composting-process model; (3) the
modeling systems during the past decades and has been development of a FCPM based on integration of proposed

536 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

fuzzy approaches and dynamic composting models; and ␣-cuts and variables. The total number of function that
(4) the verification of the proposed modeling system needs to be evaluated for n variables and m ␣-cuts can be
through a pilot-scale laboratory experiment. estimated based on the following equation:

EXPERIMENTAL WORK

m
With the fuzzy approaches, the uncertain parameters in PN ⫽ 共2j ⫺ 1兲 n (2)
the compost model can be quantified as fuzzy sets, lead- j⫽1
ing to a FCPM. The whole system integrates the fractional
FV (FFV) method, kinetics of composting processes, water
balance, and mass/energy conservations. where PN is the number of evaluation, and j is the jth
␣-cut level. Obviously, the number of function evalua-
tions will become unmanageable for problems with
FFV Method
large numbers of uncertain variables or ␣-cuts. Mean-
In fuzzy set theory, a membership function ␮A (x) is used
while, if an insufficient number of ␣-cuts is used, the
to represent the degree to which an element belongs to
interior extremum might still be missed. Therefore, a
the set in question.25 The membership function of any
method that can reduce the number of function evalu-
fuzzy set A may conveniently be expressed for all x⑀X in
ations but still ensure the solution containing the ex-
canonical form19:
tremum of the function within the bounds of the fuzzy
variables is desired. In addition, influence of uncer-


f A 共x兲 when x⑀关a,b兲 tainty on model outputs from an individual input pa-
1 when x⑀关b,c兴 rameter was not well handled through the FV method.
␮ A 共x兲 ⫽ g 共x兲 when x⑀共c,d兴 (1) Over the last decades, improvement of this method has
A
0 otherwise already been an objective of a number of studies.27 The
transformation method was such an effort that has
been proved to be successful in many applications.24 In
where a, b, c, d ⑀ X and a ⱕ b ⱕ c ⱕ d, fA is a real-valued
this study, a modified version of the vertex method,
function that is increasing and right continuous, and gA
FFV, is developed based on fuzzy transformation and
is a real-valued function that is decreasing and left
degree-of-influence (DI) determination.
continuous.
There are many approaches for generating member-
ship functions; among them, the Gaussian membership Fuzzy Transformation
function is commonly used by many researchers.22 A For a fuzzy-parameterized simulation model, each fuzzy
Gaussian fuzzy set can be defined by specifying three parameter (xi) can be decomposed into m ␣-cuts, leading
numbers: (1) the most credible value, (2) the lowest pos- to PAi ⫽ {xi(1), xi(2), . . . , xi(m)}. The xi( j) (j ⫽ 1, 2, . . . , m)
sible value, and (3) the highest possible value. These num- represents a fuzzy interval at the jth ␣-cut level, denoted as
bers can be acquired through approaches such as inter- xi( j) ⫽ [ ai( j), bi( j) ]. The ai( j) and bi( j) are the lower and upper
views, round-table meetings, and questionnaire surveys. bounds at the jth ␣-cut level, respectively. Obviously,
The most credible value is assigned a membership value of when j ⫽ 1, then ai(1) ⫽ bi(1) . Assume an equal distance
1, and any number that falls short of the lowest possible over ␮ axis between each pair of neighboring ␣-cut levels,
value or exceeds the highest possible value will get a the membership grade ␮j at ␣-cut j can be given by ␮j ⫽
membership grade of 0. The intermediate membership (m ⫺ j)/(m ⫺ 1). In the conventional FV method, the
grades can be obtained by interpolation based on the input-value vector (cij, i ⫽ 1, 2, . . . , n; j ⫽ 1, 2, . . . , m)
Gaussian distribution curve. consists of interior points at the jth ␣-cut level. Thus the
The ␣-level cut (defined as A␣ ⫽ {x ␮A(x) ⱖ ␣}) is a set ith input variable can be given as follows:17
of elements that belong to fuzzy set A at least to degree ␣,
and this degree is also called the degree of confidence or c ij ⫽ T i,j
1 2
, T i,j k
, . . . ,T i,j 2j ⫺ 1
, . . . , ,T i,j ,
the degree of plausibility. The FV method was first pro- k ⫽ 1, 2, . . . , 2j ⫺ 1, (3)
posed by Dong and Shah.26 The basic idea is to divide
input membership domain into a series of equally spaced
k
␣-cuts; for each ␣-cut, the upper and lower points, as well where Ti,j is the kth element of cij, representing a parameter
as the interior points of each fuzzy variable, are selected; for the kth interior point, which is defined as follows:
through model simulation, with different combinations
of those points being used as input values, output mem-


a i共j ⫺ k ⫹ 1兲 1⬍k⬍j⫺1
bership domain can be given by the maximum and min-
共1兲 共1兲
imum values of model results at each ␣-cut levels. Dong k
T i,j ⫽ a i or bi k⫽j (4)
and Shah27 noted that the maximum or minimum value 共k ⫺ j ⫹ 1兲
b i j ⫹ 1 ⱕ k ⱕ 2j ⫺ 1.
of the response function might not occur at the bounds of
the interval. The extremum of the function within the
interval must be included, or an incorrect result will be According to definitions in eq 4, the number of
obtained. In the FV method, all of the possible combina- model evaluation at the jth ␣-cut level would be (2j ⫺ 1)n.
tions of the interior points from all of the ␣-cuts are When the number of input variables is large, the resulting
evaluated. However, the number of function evaluations computational requirement would then limit the applica-
of the FV method increases with the numbers of both bility of the FV method.

Volume 57 May 2007 Journal of the Air & Waste Management Association 537
Qin, Huang, Zeng, Chakma, and Xi

Figure 1. Selections of interior points for (a) FV and (b) FCPM methods.


Based on the concept of fuzzy transformation n
⳵f
method that proposed by Hanss,24 interior extremum can df ⫽ 共p៮ 兲dx i , (7)
also be obtained based on a more efficient selection of ⳵x i
i⫽1
internal points. For such an improved selection scheme,
the input-value vector c1j (i ⫽ 1, 2, . . . , n; j ⫽ 1, 2, . . . , m)
where df is the total differential, which can be interpreted
consisted of interior points at the jth ␣-cut level, and ith
as an approximation of the overall change rate of the
input variable can be given as follows:
function value f(p៮ ) when the input parameters xi are
changed by dxi around x៮ i. Each input change dxi individ-
c 1j ⫽ 共I i,j
1 2
, I i,j r
, . . . , I i,j j
, . . . , I i,j 兲 共r ⫽ 1, 2, . . . , j兲, (5) ually contributes dfi to the overall change rate df. If the
change rate dxi of the ith input parameters is assumed to
be a constant percentage c of its corresponding peak
r
where Ii,j is the rth element of cij, defined by the following value, dxi can be replaced by cx៮ i. Thus, eq 10 converts as
equations:28 follows:

冘冋 册

a i共j兲
n
r⫽1 ⳵f
df ⫽ c x៮ i 䡠 共p៮ 兲 (8)
r⫺1 ⳵x i
⫽ 共I ⫹I 2ⱕrⱕj⫺1
r r
I i,j i,j ⫺ 1 i,j ⫺ 1 /2 (6) i⫽1

b i共j兲 r⫽j
Based on eq 10, the normalized change rate ␳i of each
input parameter is given as follows:
Figure 1 shows the selection schemes of interior points of
both FV and FFV methods. Based on eq 6, the number of
x៮ i 䡠 兩⳵f/⳵x i 兩 x ⫽ p៮
interior points at the jth ␣-cut level for FFV method is j. ␳i ⫽
Thus, the number of model evaluation at the jth ␣-cut

n

level for the FFV method becomes jn. 关x៮ i 䡠 兩⳵f/⳵x i 兩 x ⫽ p៮ 兴 (9)
i⫽1

DI Determination
The influence measure is critical for quantifying the where ␳i is interpreted as the DI for input parameter i,
proportion to which the uncertainty of a single model
n
satisfying ¥i⫽1 ␳i ⫽ 1. It is used for quantifying the influ-
parameter contributes to the overall uncertainty of the ence of the ith varying parameter (xi) on the overall vari-
model output. It can provide more valuable information ation (df) of the problem output, that is, f(x1, x2,.., xn).
in describing system uncertainties. If the model function
f is available in analytical form, its total differential df can Solution Algorithm
be easily determined and used to compute the degrees of The proposed FFV method was based on a reduced num-
influence for each fuzzy input parameter.24 ber of interior-point combinations, showing a more effi-
Let xi (i ⫽ 1, 2,.., n) be one of the fuzzy input param- cient algorithm and, thus, a better applicability in real
eters, with peak value at x៮ i. Based on differential calculus, applications. Figure 2 shows a general framework of the
the total differential at point p៮ ⫽ (x៮ 1, x៮ 2, . . . , x៮ n) can be proposed method. Detailed pseudocode can be described
described as follows: as the following steps:

538 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

biological reactions of the composting system. Associated


thermal, physical, and chemical reactions should also be
accounted for in constructing an effective and compre-
hensive modeling system. The Monod equation that has
been widely used expresses the microbial growth rate as a
function of nutrient that limits microbial growth.29 Ex-
pression of this equation is of the same form as the
Michaelis-Menton equation for enzyme kinetics but was
derived empirically.30 The limiting nutrient can be sub-
strate, nitrogen, phosphorous, or any other similar elec-
tron acceptors. As an extension to Monod equation, the
Contois equation of growth is more effective in describing
the process of high-density cell fermentations.31 Thus, the
Contois equation of growth is used in this study for de-
scribing microbial kinetics.32 It can be written as follows:

␮ s ⫽ ␮ smax 冉 SX
K sX ⫹ S 冊
⫺ bX, (10)

where ␮s is the specific growth rate (1/h); X is the biomass


concentration (kg/m3); S is the substrate concentration
Figure 2. Block diagram of FCPM method. (kg/m3); ␮smax is the maximum specific growth rate (1/h);
Ks is the half saturation constant (value of S at which ␮s
First, build fuzzy numbers A1, A2, . . . ., An for uncer-
equals a half of ␮smax); and b is endogenous decay con-
tain input parameters x1, x2, . . . , xn and denote an element
stant (1/h).
of Ai by xi, i ⫽ 1, 2, . . . , n, where n is the total number of
Substrate utilization is described by the ratio of the
uncertain input parameters. The output y can be related
specific growth rate to yield coefficient, Yx/sec, which can
to input x1, x2, . . . , xn through the model simulator (i.e.,
be determined by experiment. Thus, the rate of substrate
mapping y ⫽ f[x1, x2, . . . , xn]).
utilization can be expressed by the following equations:
Second, discretize the range of membership grade
[0,1] into a finite number of ␣-cuts (␣1, ␣2, . . . , ␣m), which
are evenly distributed.
Third, arrange combinations of the m ␣-cut level for n rs ⫽
dS
dt
⫽⫺
rx
Yx/s
⫽⫺
␮smax SX
Yx/s KsX ⫹ S冉⫹
b
Yx/s
X. 冊 (11)
input parameters. Interior points on fuzzy intervals at
each ␣-cuts can be defined based on eqs 8 and 9. where rs is the rate of substrate utilization (kg/m3䡠h) and rx
Fourth, run the simulation model for each of the is the rate of biomass growth (kg/m3䡠h). Item ␮s max/Yx/s
combination and obtain all possible values for the output in eq 10 is often denoted as k⬘, that is, the maximum
y, denoted as yk (k ⫽ 1, 2, . . . , p). specific substrate-utilization rate (1/T); and b/Yx/s is called
Fifth, analyze these results regarding the same output kd, that is, microbial maintenance coefficient (1/h). Thus,
membership value based on the same ␣-cuts (␣1, ␣2, . . . , formula 10 can be derived as follows:
␣m) as those for input parameters. The upper and lower
bounds of the output value at each ␣-cut level can be
given by the maximum and minimum values of yq (q ⫽ 1,
2, . . . , j) (which is obtained in step 4 at the same ␣-cut
r s ⫽ ⫺k⬘ 冉 SX
KsX ⫹ S 冊
⫹ kdX. (12)
level; j is the number of interval and boundary points at
the jth ␣-cut level). According to Haug,33 constant k⬘ is represented by
Sixth, the final solution to the fuzzy number B can be multiplicative subfactors of temperature, oxygen, free air
obtained based on analysis of upper and lower bounds of space, and moisture constant as follows:
the output values at different ␣-cut levels.
Last, is the determination of DI based on a total
differential method. k⬘ ⫽ k 䡠 k T 䡠 k moisture 䡠 k FAS 䡠 k O 2, (13)
Generally, a specific simulation model is required be-
fore the proposed fuzzy approach can be used. In this study, where kT is the temperature correction, kmoisture is the
a comprehensive composting model that is capable of ad- moisture content correction, kFAS is the free airspace cor-
dressing gradients of substrate, water, oxygen, and biomass, rection, and kO2 is the oxygen concentration correction.
as well as fluctuation of temperature levels, will be devel- These corrections are described in detail as follows.
oped. Such a modeling system is based on the knowledge of
microbial kinetics, mass, and energy conservations. The Temperature Correction kT
The relationship between microbial-specific growth rate
Kinetics of Composting Processes ␮s and temperature in compost bulk is described by equa-
The complex and dynamic interactions within bioconver- tion as follows33:
sion are fundamental components for properly describing when T ⱕ TM

Volume 57 May 2007 Journal of the Air & Waste Management Association 539
Qin, Huang, Zeng, Chakma, and Xi

kT ⫽
␮s
␮ ss
⫽ exp ⫺ 再 冉
EA 1

1
RA T ⫹ 273 Ts ⫹ 273 冊冎 (14)
Therefore, Kparticle should be incorporated in the oxygen
correction equation, with an empirical relationship being
given as follows:
when TM ⱕ T ⱕ TL
Vol%O 2
k O2 ⫽ 䡠K , (19)
KO ⫹ Vol%O2 particle
␮ TL ⫺ T
kT ⫽ ⫽ (15)
␮ sm T L ⫺ T M
where Kparticle is particle size coefficient (ranging from 0 to
1). The value of Kparticle can be determined based on
and when T ⱖ TL, ␮s ⫽ 0 and kT ⫽ 0, experiment.
where ␮ss and ␮sm the microbial specific growth rates
(1/h) at the preference temperature Ts and medium
Water Balance and Mass/Energy Conservation
threshold temperature TM (°C), respectively; T is the tem-
Mass Balance. The mass balance of substrate, biomass,
perature of the composting bulk (°C); TL is the maximum
and water during time t ⫺ 1 to t can be described as
allowable temperature for microbial activities; EA is the
follows:34
activate energy of composting bulk (J/mol), and R is the
universal gas constant (J/mole䡠K)
dPI
⫽ PI t ⫺ 1 ⫺ PI t , (20)
The Moisture Content Correction kmoisture dt
Based on experimental results, the relationship between
microbial maximum specific growth rate and water con- where PI is a general term representing S (mass of organic
tent in compost bulk is identified as follows: when water substance, kg), X (mass of microorganisms, kg), and W
content w is lower than critical value wa, which is critical (mass of water in the reactor, kg), respectively; t is the
for microbial growth, then ␮s ⫽ 0 and kmoisture ⫽ 0; when time period (h).
w is greater than wa, we have the following:
Water Balance. Water content in the composting bulk is
␮s w ⫺ wa correlated to evaporation rate and airflow rate. The con-
k moisture ⫽ ⫽ . (16) servation equation of water content is given as follows31:
␮ smax K a ⫹ w

dW W
When w is greater than the suitability index w1 (empiri- ⫽ ⫺␭jq , (21a)
cally set as 60%), we have the following: dt M

␮s w ⫺ wa w2 ⫺ w 18 ps 273
k moisture ⫽ ⫽ 䡠 j⫽ 䡠 䡠 , (21b)
␮ smax K a ⫹ w w 2 ⫺ w 1
(17) 22.4 p 0 ⫺ p s T ⫹ 273

where q is the flow rate of air supply (m3/h); ␭ is the satura-


where w2 is the maximum allowable index for water con-
tion ratio of vapor; M is the total mass of compost bulk,
tent (empirical value 80%).
equals S ⫹ X ⫹ W ⫹ U (U is the nondegradable content); j is
the saturated water vapor content (kg/Nm3); ps is the satu-
The Oxygen Concentration Correction kO2 ration vapor pressure (Pa); po is the air pressure (Pa); and T is
The oxygen concentration is limited by diffusion of the the temperature of composting bulk (°C). The relationship
particle matrix. A Monod-type expression can be used to between saturation vapor pressure and air pressure can be
simulate the oxygen limitations, given as follows: found in Mohee and White4 and Xi et al.35

Vol%O 2 Energy Conservation. Energy conservation is addressed


k O2 ⫽ , (18)
KO ⫹ Vol%O2 through the thermal balance within the composter, de-
scribed as follows:
where Vol%O2 is the percentage of oxygen in the incom-
ing air. If the substrates are well mixed, the oxygen levels
in the FAS between composting particles are assumed to C cM
dT
dt
⫹ h1冉dS dX
dt

dt
⫽ h2 冊
dW
dt
⫹ 共T a ⫺ T兲

冋 冉 冊册
be identical. The half velocity coefficient Ko is calculated
through the relationship between the flow rate and oxy- dW dS dX
⫻ qC a ⫹ KF ⫺ C w ⫺ Cs ⫹ (22)
gen concentration in the exhaust gas. Empirical value is dt dt dt
generally set to be 2%.31
In real composting reactions, oxygen concentrations where Cc is the heat capacity of compost bulk [kJ/(kg䡠°C)];
in the exhaust air are normally maintained above 6%, h1 is the heat quantity generated by unit dry organic
preventing the composting bulk from becoming faculta- (kJ/kg); h2 is the potential heat of water evaporation (kJ/
tive or anaerobic. However, particle sizes also exert diffu- kg); T is the temperature of compost bulk (°C); Ta is the
sion resistances in composting bulk. Significant errors will temperature of inflow air (°C); q is the rate of air supply
be generated without considering this critical effect. (m3/h); Cw is the heat capacity of water [kJ/(kg䡠°C)]; Ca is

540 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

the heat capacity of air [kJ/(kg䡠°C)]; Cs is the heat capacity Suppose there are n independent fuzzy-valued input
of volatile organic; K is the thermal conductivity coeffi- parameters xi with the membership function ␮j(xi) at the
cient of compost facilities (kJ/m2䡠h䡠°C); and F is the total jth ␣-cut level, i ⫽ 1, 2, . . . , n, j ⫽ 1, 2, . . . , m. The fuzzy
thermal dispersion area of compost facility (m2). modeling system for composting process (FCPM) can be
On the left hand side of eq 22, CcMdT/dt is the heat formulated as follows:
change caused by temperature change of compost bulk;
h1(dS/dt ⫹ dX/dt) is the energy variation for microbial cell 关共y 1共1兲 , y 1共2兲 , . . . , y 1共m兲 兲, . . . , 共y 共1兲 共2兲
k , yk ,
synthesis and substrate degradation. On the right hand
side, h2dW/dt is the energy loss by water evaporation; (Ta . . . , y 共m兲 共1兲 共2兲 共m兲
k 兲, 共y p , y p , . . . , y p 兲
⫺ T)[qCa ⫹ KF ⫺ CwdW/dt ⫺ Cs(dS/dt ⫹ dX/dt)] is the ⫽ ␾兵关共x 1共1兲 , x 1共2兲 , . . . , x 1共m兲 兲, . . . , 共x i共1兲 , x i共2兲 ,
energy variation because of temperature change of air,
water, and volatile organic, as well as system heat loss. . . . , x i共m兲 兲, 共x 共1兲 共2兲 共m兲
n , x n , . . . , x n 兲兴, P, t其, (26)
The thermal conductivity coefficient of compost facility
can be estimated by equations provided by Xi et al.35 where yk is the kth element among p model outputs.
The composting process is often affected by a variety
Oxygen Concentration in Exhaust Air. The oxygen concen- of biochemical and physical factors, as well as their inter-
tration in exhaust air is related to oxygen uptake rate, actions. During simulations, they are generally described
airflow rate, and initial oxygen concentration. It can be through a number of interrelated differential equations.
calculated as follows: Consequently, the partial derivatives of model function
on specific inputs are not always available in single ana-
lytical forms. This leads to difficulties in calculating the
a v 273 ⫹ T ⌬S
z ⫽ z0 ⫹ 䡠 䡠 , (23) DI of input parameters through total differential method.
q 273 ⌬t However, the partial derivative of model function fj
(x1, x2, . . . , xn) (j ⫽ 1, 2, . . . , m) on variable xi can often
where z0 is the initial oxygen concentration (%); q is the be approximated based on the following equations:
airflow rate (m3/h); and av is the required oxygen amount
(under standard conditions) for decomposing unit mass ⳵f i
of substrate (Nm3/kg). 共x៮ ,x៮ , . . . , x៮ n 兲
⳵x i 1 2
兩f j 共x៮ 1 ,x៮ 2 , . . . , x៮ n 兲 ⫺ f⬘j 共x៮ 1 ,x៮ 2 , . . . , x៮ n )兩
Fuzzy Composting Process Model ⬇ (27)
Combination of the FFV method and the dynamic com- ⌬x i
posting model leads to formation of an FCPM. It is com-
plicated with many differential equations, where a num- where xi is one of the fuzzy input parameters (i ⫽ 1, 2, . . .
ber of parameters are highly interrelated. Generally, the , n); fj is one of the system outputs (j ⫽ 1, 2, . . . , 6); x៮ i is
interactions between system inputs and outputs can be the peak value of parameter xi; ⌬xi is a small change of
conceptualized into the following expression: parameter xi; fj⬘ is the change of model output generated
by the change of parameter xi; n is the number of input
parameters; and m is the number of output items.
Y OU ⫽ ␸共X IN , P IN , t兲, (24)
Thus, the total differential at the peak point P៮ ⫽
(x៮ 1,x៮ 2, . . . , x៮ n) can be given as follows:
where XIN is the vector of system inputs/parameters that
considered to be uncertain, such as initial water content
冘冋 册
n
and yield coefficient; PIN is the deterministic system in- 兩f j 共x៮ 1 ,x៮ 2 , . . . , x៮ n 兲 ⫺ f⬘j 共x៮ 1 ,x៮ 2 , . . . , x៮ n 兲兩
df j ⫽ 䡠 c 䡠 x៮ i
puts such as airflow rate; t is the composting time; and i⫽1 ⌬x i
YOU is the vector of system outputs such as temperature
(28)
and substrate concentration.
The objective of the FCPM method is to generate a set
of fuzzy-valued outputs. The manipulation process entails DI can then be derived by the following equation:
fuzzification of input parameters, simulations under dif-
ferent scenarios, and generation of fuzzy membership 兩f j 共x៮ 1 ,x៮ 2 , . . . , x៮ n )⫺f⬘j (x៮ 1 ,x៮ 2 , . . . , x៮ n )兩䡠x៮ i /⌬x i
functions of various model outputs. For composting pro- ␳ ij ⫽


n
cesses, the fuzzy input parameters are assumed to have [兩f j (x៮ 1 ,x៮ 2 , . . . , x៮ n )⫺f⬘j (x៮ 1 ,x៮ 2 , . . . , x៮ n )兩䡠x៮ i /⌬x i ]
symmetric quasi-Gaussian distributions, given as follows: i⫽1


(29)
␮共x兲 ⫽ e ⫺ 共x ⫺ m៮ 兲 for兩x ⫺ m
៮ 兩 ⱕ 3␦
2 /2␦ 2

␮共x兲 ⫽ 0 for x ⬎ m
៮ ⫹ 3␦ or x ⬍ m
៮ ⫺ 3␦, where ␳ij is the degree of influence of input parameter i that
contributes to the overall uncertainty of the model output j.
(25) Generally, the FCPM method is not only capable of
handling uncertainties associated with composting pro-
where m៮ and ␦ denote the mean value and the standard cess but also having lower computational requirements. A
deviation of the Gaussian distribution. quantitative analysis in terms of the impacts of different

Volume 57 May 2007 Journal of the Air & Waste Management Association 541
Qin, Huang, Zeng, Chakma, and Xi

through a fine mesh screen near the bottom of the reac-


tor. The leachate from the composting bulk is collected
and recycled through drum rotations, aiming at preventing
the composting bulk from drying out too fast. Thermocou-
ple sensors are used for temperature measurements. An ox-
ygen analyser (MD-520E) is used for oxygen detection.
Composting raw materials were collected from local
communities. Initial moisture content of the compost
mixtures was adjusted to ⬃52.4 –57.5% (g H2O/g wet sol-
ids); initial substrate of raw materials was measured at
⬃35%. The airflow was controlled at ⬃0.020 – 0.033 m3/
h䡠kg. During experiment, ⬃150 g of solid waste samples at
three different locations were collected for measuring sub-
strate and water. Water content was obtained by calculat-
ing weight loss of the sample after drying in the oven at
101 °C for 24 hr. Volatile solids content was determined
by baking samples at 550 °C for ⱖ6 hr in a muffle furnace.

Model Development
Adaptations and assumptions for the modeling system are
listed as follows. First, based on the experimental results
of testing samples, the composting bulk takes ⬃3 days to
reach temperature peak and 15 days to reach mature.
Thus, 360 hr (15 days) are handled as the temporal do-
main. Second, the interactions of particles and moisture
are beyond the scope of this study; thus, FAS was assumed
to be constant. Third, airflow rate has significant impacts
on the composting performance. Proper levels should be
maintained to provide sufficient oxygen concentration, as
Figure 3. Schematic diagram of experimental reactor. well as to prevent undesired loss/gain of water and heat.
Based on the results of test runs, the airflow rate with a range
factors can also be investigated through this method. from 0.02 to 0.05 m3/h䡠kg is considered to be suitable.
Thus, effective precautions can then be taken to either A number of experiments are conducted for parameter
decrease the influence of sensitive factors or neglect those estimation under different factorial designed scenarios, for
that are insensitive. example, under varied levels of initial moisture content,
initial biomass, and airflow rate. A list of the estimated
RESULTS coefficients is listed in Table 1. Other coefficients and pa-
Pilot-Scale Composting System rameters are estimated from the experimental results of
A pilot-scale composter was constructed from a cylindri- other researchers or from related literature. 33 For example,
cal polyvinyl chloride tube with an effective volume of ␮smax is estimated from laboratory composting experiments
⬃250 L. Detailed schematic diagram of the experimental assuming logarithmic growth.3 Y is estimated by regression
reactor is presented in Figure 3. The air is flowing upward analysis of experimental results under different levels of

Table 1. Coefficients and parameters used in the application.

Parameters Notation Default Unit

Maximum specific growth rate ␮smax 0.18 h⫺1


Half velocity constant for degradable substrate Kc 24 g/kg
Yield coefficient Y 0.50 kg/kg
Activate energy of compost bulk EA 29 kJ/mol
Universal gas constant RA 8.314 J/(mol 䡠 K)
Critical value of water content wa 15%
Half velocity coefficient of oxygen correction Ko 0.066 g/kg
Moisture correction coefficient Ka 4.0%
Flow rate of air supply q 0.033 m3/(h 䡠 kg)
Heat productivity generated by unit dry organic h1 1.76 ⫻ 104 kJ/kg
Latent heat of water evaporation h2 2.44 ⫻ 103 KJ/kg
Heat capacity of water Cw 4.2 kJ/(kg䡠°C)
Heat capacity of air Ca 1.4 kJ/(kg䡠°C)
Heat capacity of volatile organics Cs 2.1 kJ/(kg䡠°C)
Oxygen requirement per unit substrate av 0.891 m3/kg

542 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

Figure 4. Comparisons between simulation and experimental results: (a) total mass, (b) substrate content, (c) microbial content, (d) oxygen
content, (e) temperature, and (f) water content.

moisture, temperature, and airflow rate based on the study and moisture content. In comparison, results obtained
result of Xi et al.31 Heat quantity generated by unit organic from the pilot-scale experiment under the same condi-
in compost bulk, i.e., h1, is ⬃17.4 MJ/kg for hydrocarbon tions are also presented.
compounds, 23.4 MJ/k for proteins, and 39.3 MJ/kg for fat The profiles of different composting indicators in Fig-
(based on the composition of solid waste used in pilot scale ure 4, a–f, describe a typical composting process. From
model, h1 for organic substrate in this study is estimated as Figure 4, a and b, the total mass and substrate content
17.6 MJ/kg). Values for potention thermal of water evapo- decrease continuously over the composting time horizon,
ration and heat capacity for water, compost bulk, and air, showing a typical pattern of biodegradation reaction.
Cw, Cc, and Ca, respectively, were found in Haug33 (detailed Generally, the growth of microorganisms consists of three
description of these coefficients and parameters are also typical stages, that is, exponential, stationary, and death
shown in Table 1).
stages. Dissimilarly, the result in Figure 4c (biomass con-
Two scenarios will be investigated by applying the
tent) demonstrates a monotonously increasing trend.
proposed FCPM. The first scenario is initiated to investi-
This is because of the fact that the content of biomass is
gate interactions among total mass weight, substrate con-
also affected by the total weight of the composting bulk,
tent, microbial biomass, oxygen concentration, tempera-
which is continuously decreasing throughout the study
ture level, and water content under deterministic
conditions. Experimental data under the same condition period. Figure 4d presents the oxygen concentration in
will be used for verifying the results from simulations. The the outlet gas from the composting reactor. It is indicated
second scenario aims to examine the impacts of uncer- that a rapid decrease of oxygen concentration occurs in
tainties in both model parameters and initial conditions the first 48 hr; afterward, it returns rapidly to the ambient
through the FCPM method. level of 20.9%. The fluctuation of oxygen level is consis-
tent with that of the biological activities over different
DISCUSSION composting stages. Temperature profiles are shown in
Model Simulation under Deterministic Figure 4e. At the early stage (in 48 hr), the temperature
Conditions level increases rapidly until a peak level is reached; then,
Figure 4 shows the simulated outputs under a determin- the high temperature levels last for ⬃3 days and decline
istic condition (w0 ⫽ 55%, X0 ⫽ 1%, and q ⫽ 0.033 gradually to the ambient level. The profile of water con-
m3/h䡠kg) through the proposed composting model. The tent is shown in Figure 4f, indicating a deceasing pattern
results include profiles of total mass, substrate concentra- similar to that of substrate degradation. This is because of
tion, biomass, oxygen concentration, temperature level, the direct water loss through evaporation.

Volume 57 May 2007 Journal of the Air & Waste Management Association 543
Qin, Huang, Zeng, Chakma, and Xi

The results in Figure 4 also indicate that the simu- Six parameters that are critical to composting processes are
lated outputs are consistent with those from the experi- selected as fuzzy input numbers; they include the maximum
mental measurements. For example, on day 15, the re- specific growth rate ␮smax, half velocity constants for de-
duced total mass and degraded substrate mass are ⬃57.5 gradable substrate Kc, yield coefficient Y, initial biomass
and 25.8 kg, respectively; in comparison, the measured concentration X0, initial water content W0, and initial tem-
levels are 52.7 and 25 kg, respectively. The relative deri- perature T0. Their membership functions are assumed to
vations are only 0.08 and 0.031, respectively. Results in have quasi-Gaussian distributions, with six ␣-cut levels (i.e.,
Figure 4d indicate that the minimum oxygen concentra- 0, 0.2, 0.4, 0.6, 0.8, and 1) being considered. The detailed
tion obtained from simulation (15.13%) is close to that membership functions are shown in Figure 5. Six system
from experiment (13.67%). Similarly, the peak tempera- outputs will be investigated, that is, reduced total mass,
ture level from simulation (63 °C) is also in good accor- substrate, biomass, and water content, as well as tempera-
dance with that from experiment (64 °C). However, a ture level and oxygen concentration.
number of unconformities are also observed. For example, Figure 6 shows the simulated fuzzy outputs over five
the oxygen concentration obtained from experiment time points (i.e., days 3, 6, 9, 12, and 15), with different
takes a few more hours to reach the minimum level com- fuzzy membership functions being demonstrated. It is
pared with that obtained from simulation. This is because indicated that the output values under various ␣-cut lev-
of the fact that the real composting reactions are normally els can be obtained from analyzing the derived fuzzy
subjected to more heat loss (caused by imperfect heat membership functions. For example, on day 9, the water
insulation) than the ideal ones, leading to reduced bioac- content is varying from 22.1% to 43.39% under ␣-cut
tivities and delayed oxygen consumptions. This also re- level of 0.4; the interval will decrease to 26.13% to 36.66%
sults in a delay of temperature to reach the ambient level when the system become more deterministic (under ␣-cut
at the final stage of the composting process. For example, level of 0.8); the water content will converge into 31.04%
it takes 15 days for the simulated temperature to reach the when the ␣-cut level turns into 1. This indicates that the
ambient level, whereas the observed one takes only 13 degrees of uncertainties of the outputs (i.e., ␣-cut values)
days. In addition, the simulated moisture contents also are directly related to those of the input parameters. The
indicate large deviations compared with the experimental results corresponding with the ␣-cut level at 1 are identi-
data. As shown in Figure 4f, an apparent intersection is cal to the results that one would get if only the mean
observed on day 9. Before the time of this point, the values are considered as crisp settings for the input pa-
measured water content is much higher than the simu- rameters. Inversely, the result obtained under ␣-cut level
lated one; afterward, it becomes much lower. at 0 corresponds with the highest system uncertainties.
Results from Figure 4 demonstrate the ability of the For example, Figure 6c shows the biomass growth at dif-
proposed model to predict substrate decomposition and ferent time points; the water content on day 6 under the
biomass growth, as well as the interactions among tem- most uncertain condition (␣ ⫽ 0) is ranging from 5% to
peratures, oxygen concentrations, and water contents. 25%, whereas it becomes deterministic (18%) under the
Most of the predicted outputs are consistent with the most ideal condition (␣ ⫽ 1).
results obtained from experiment. It also implies that The results in Figure 6 also demonstrate that the
different composting parameters are highly interrelated. output curves are in irregular shapes. For example, the
For example, an increasing biomass growth normally cor- fuzzy membership functions of the oxygen outputs on
responds with decreasing substrate content; a rapid sub- day 3 (Figure 6d) and temperatures on days 3 and 6
strate degradation rate generally occurs in high tempera- (Figure 6e) demonstrate strong nonsymmetric shapes
ture range and corresponds with a high oxygen uptake compared with the quasi-Gaussian shapes of input pa-
rate; a peak temperature corresponds with a minimum rameters. However, the fuzzy membership functions of
oxygen concentration. temperature output on days 9, 12, and 15 are still in fine
Results in Figure 4 also demonstrate a number of symmetric quasi-Gaussian shapes. This is because of the
deviations between model simulation and real reactions nonlinearities of the composting models. Figure 6 also
because of the existence of a variety of uncertainties. For demonstrates a time-varying effect in terms of the fuzzy
example, in composting experiment, the accuracy of tem- membership functions. As shown from Figure 6, a–f,
perature prediction is often affected by measuring errors the degrees of uncertainties in terms of total mass,
(e.g., improper human operations or instrumental errors). oxygen concentration, and temperature levels are de-
The initial reaction conditions and many model parame- creasing over time, whereas the degrees of uncertainties
ters usually exhibit rather high degrees of uncertainties of substrate, biomass, and water contents exhibit in-
arising from incomplete or imprecise information. Con- creasing patterns. For examples, the fuzzy interval of
sequently, the simulation outputs that are normally ob- oxygen concentration on the third day ranges from
tained through deterministic models (with crisp-value pa- 14.49% to 19.65% under ␣-cut level at 0; in compari-
rameters) can hardly be used to describe all of the possible son, the interval reduces to the range from 20.02% to
system responses. To bridge this gap, the FCPM method
20.90% on day 15. The results in Figure 6 also imply
will be proposed, with detailed analysis being provided in
that the change rate of the individual output is consis-
the following scenario.
tent with the distance between the curve peaks at dif-
ferent time points. For example, the water content in
Simulation under Uncertainty Figure 6f changes rapidly (⬃9.15% per day) between
In the second set of simulations, impacts of uncertainties days 3 and 6, indicating a high biodegradation rate;
will be investigated through the proposed FCPM method. however, the rate reduces to 5.41% per day from day 6

544 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

Figure 5. Membership functions of input parameters: (a) initial water content, (b) initial temperature, (c) initial microbial content, and
(d) maximum growth rate.

to day 9 and reaches a minimum level (2.47% per day) less uncertain conditions (␣ ⬍ 0.4) generally exert ⬎50%
after day 12. derivations compared with the corresponding mean val-
The results in Figure 6 demonstrate that the uncer- ues. Particularly, the minimum possible rate of total mass
tainties projected in model input parameters and initial reduction (shown in Figure 6a) under ␣-cut levels at 0 on
conditions will significantly impact the predicted out- day 6 is ⬃84.97% lower than the level of the mean value;
puts. As shown in Figure 6, the fuzzy outputs under the the minimum temperature level (shown in Figure 6f)

Volume 57 May 2007 Journal of the Air & Waste Management Association 545
Qin, Huang, Zeng, Chakma, and Xi

Figure 6. Simulation results through FCPM method: (a) total mass reduction, (b) mass percentage of substrate, (c) microbial mass percentage,
(d) output oxygen concentration, (e) temperature, and (f) water content.

546 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

Table 2. Comparisons between outputs with FCPM and FV methods in day 3.

␣-Cut Level

Output Items Methods 0 0.2 0.4 0.6 0.8 1.0 Relative Error

Total mass reduction (%) FCPM (0.10, 32.6) (2.63, 26.75) (4.48, 24.37) (6.32, 22.33) (8.74, 20.11) (15.02, 15.02) 0
FV (0.10, 32.6) (2.63, 26.75) (4.48, 24.37) (6.32, 22.33) (8.74, 20.11) (15.02, 15.02)
Substrate content (%) FCPM (29.80, 36.76) (30.96, 34.82) (31.37, 34.21) (31.70, 33.82) (32.04, 33.44) (32.73, 32.73) 0
FV (29.80, 36.76) (30.96, 34.82) (31.37, 34.21) (31.70, 33.82) (32.04, 33.44) (32.73, 32.73)
Microbial content (%) FCPM (2.17, 8.31) (2.78, 7.46) (3.15, 7.09) (3.55, 6.75) (4.09, 6.36) (5.40, 5.40) 0.0012
FV (2.17, 8.32) (2.78, 7.46) (3.15, 7.09) (3.55, 6.75) (4.09, 6.36) (5.40, 5.40)
Temperature (oC) FCPM (28.67, 67.71) (38.81, 66.01) (44.53, 65.36) (50.18, 64.84) (56.96, 64.31) (63.22, 63.22) 0.0005
FV (28.66, 67.71) (38.81, 66.02) (44.53, 65.36) (50.18, 64.84) (56.96, 64.31) (63.22, 63.22)
Water content (%) FCPM (41.77, 56.45) (45.44, 55.34) (46.81, 54.81) (47.90, 54.28) (49.01, 53.56) (51.27, 51.27) 0
FV (41.77, 56.45) (45.44, 55.34) (46.81, 54.81) (47.90, 54.28) (49.01, 53.56) (51.27, 51.27)
Outlet oxygen content FCPM (14.68, 19.65) (14.99, 18.70) (15.10, 18.03) (15.60, 17.27) (15.44, 16.23) (15.79, 15.79) 0.0035
FV (14.49, 19.65) (14.93, 18.70) (15.05, 18.03) (15.38, 17.27) (15.44, 16.23) (15.79, 15.79)

under ␣-cut levels at 0.2 on day 3 is ⬃69.97% lower than the two methods. The relative derivations of biomass
the mean level; and the maximum possible level of water content, temperature, and oxygen concentration are
content (shown in Figure 6f) under ␣-cut levels at 0.4 on 0.12%, 0.05%, and 0.35%, respectively. Comparisons for
day 15 is ⬃60.07% higher than the level of the mean the other time points also demonstrate similar relations.
value. Therefore, neglecting uncertainties in simulation These results imply that the FCPM has a satisfactory ac-
will often lead to a lack of information regarding system curacy level and is, thus, applicable to practical problems.
responses. This is especially true in situations when ap-
plying models under deterministic considerations. In
Analysis on DI
comparison, the FCPM method could predict system out-
puts under different possibilistic levels, with more useful DI is provided to describe the impacts of uncertainties of
information being provided. Therefore, it could effec- model input parameters on system responses. However,
tively communicate implicit knowledge into dynamic the DI levels are different over time because of the dy-
simulations and, thus, obtain reliable and applicable namic features of composting systems. Figure 7 shows the
model outputs. results of DI for the input parameters on day 3. It is
indicated that their DI levels are significantly different.
For example, DIs of the initial water content (W0) dem-
Comparison between FCPM and FV Methods
onstrate moderate influences on the outputs of total mass
The traditional fuzzy modeling approach (FV) generally
reduction (␳ ⫽ 27.6%) and water content (␳ ⫽ 35.5%);
has high computational requirements. The proposed
however, it shows a much higher level of influence on
FCPM approach is developed based on reduced interior-
substrate content (␳ ⫽ 67.6%) and negligible influences
point distributions, leading to significantly reduced cal-
on biomass, temperature, and oxygen concentration. The
culations. The reduction percentage can be estimated by
the following equation: DIs of the yield coefficient (y) indicate significant impacts
on most of the system outputs (from 16.8% to 42.4%); in
comparison, the initial biomass content has a much lower

冘 冘
n n
influence (e.g., DI on temperature levels is only 0.69%).
共2i ⫺ 1兲 m ⫺ jm This is because of the fact that the microorganisms have
i⫽1 j ⫽ 1 excellent reproductive capacities, leading to less signifi-
r reduced ⫽ , (30)


n cant influences on biodegradation reactions.
共2i ⫺ 1兲 m The results in Figure 7 demonstrate that the proposed
i⫽1 FCPM method is effective in quantifying the proportion
to which the uncertainty of each single model parameter
where r is the reduction rate for calculation numbers; n is contributes to the overall uncertainty of the model out-
the number of ␣-cut levels; and m is the number of fuzzy put. Consequently, those sensitive parameters that may
input parameters. In this study, n ⫽ 6 and m ⫽ 6, so the lead to errors or unreasonable predictions on system out-
calculation number is 2.43 ⫻ 106 for the traditional FV puts can be determined. Thus, special efforts should be
method and 6.72 ⫻ 104 for the FCPM method where the made to accurately define those sensitive parameters be-
rreduction is as high as 97.24%. fore any simulations and result analysis be conducted. For
The accuracy of the FCPM method was verified by example, the initial water content and yield coefficient
comparing its results with those obtained through the FV could lead to significant derivations on predictions of
method. Table 2 shows the system outputs (of day 3 under substrate content (DI levels are 67.7% and 16.8%, respec-
different ␣-cut levels) from the FCPM and FV methods. tively); if inaccurate input value or biased fuzzy distribu-
The results of total mass reduction, substrate content, and tions are defined before simulation, significant deriva-
water content under all of the ␣-cut levels are identical for tions will occur. Therefore, practical arrangements can be

Volume 57 May 2007 Journal of the Air & Waste Management Association 547
Qin, Huang, Zeng, Chakma, and Xi

Figure 7. The DIs of uncertain parameters on model outputs in day 3.

made to efficiently reduce system uncertainties by a well- insensitive fuzzy parameters. For example, when a sensi-
directed reduction of the uncertainties of those parame- tive parameter (e.g., W0) is not considered fuzzy, a signif-
ters showing high DI levels. icantly different curve would be generated; conversely,
Another remarkable conclusion that can be drawn when an insensitive parameter (e.g., X0) is neglected, no
from Figure 7 is that the negligible influential parameters significant variation would occur.
may be neglected to reduce the labor of calculation. For
example, in Figure 7, extravagant expenses for a prefera- CONCLUSIONS
bly exact identification of the parameter X0 can be con- In this study, an FCPM was developed for simulating the
sidered as a waste of effort, because the influence of its composting process under uncertainty. This model was
variation on most of the outputs (especially for tempera- mainly based on integration of an FFV method and a
ture prediction) can be neglected; without this factor, the comprehensive composting model. DIs by projected un-
calculation number could reduce ⬃69.46%. certain factors were also investigated.
To further demonstrate the advantage of the pro- Two scenarios were investigated in applying the
posed method in reducing the computational require- FCPM method. In the first scenario, model simulation
ment, system responses under seven scenarios are inves- under deterministic conditions was conducted. A pilot-
tigated. These scenarios are established for examining the scale experiment was provided for verifications. The result
fuzzy outputs of substrate contents on day 3, with the indicated that the proposed composting model could pro-
respective fuzzy inputs being defined as follows1: (1) all of vide an excellent vehicle for demonstrating the complex
the six parameters are fuzzy; (2) initial water content (W0) interactions that occurred in the composting process. In
is deterministic, whereas the others are fuzzy; (3) yield the second scenario, application of the proposed FCPM
coefficient y is deterministic, whereas the others are fuzzy;
(4) initial temperature T0 is deterministic, whereas the
others are fuzzy; (5) maximum degradable rate ␮s is de-
terministic, whereas the others are fuzzy; (6) half velocity
of degradable substrate Kc is deterministic, whereas the
others are fuzzy; and (7) initial biomass content X0 is
deterministic, whereas the others are fuzzy.
The solutions are shown in Figure 8, indicating sim-
ilar results as compared with those in Figure 7. The system
response under scenario 1 shows the largest difference
compared with that under scenario 2; scenario 4 ranks the
second in term of the level of difference. The responses
under scenarios 3, 5, 6, and 7 are relatively close to that
under scenario 1. Quantitatively, in reference to the result
of scenario 1, the average relative deviations [average rel-
ative deviation ⫽ (result of comparing scenario ⫺ result of
reference scenario) ⫻ 100%] under scenarios 1 to 7 are
3.79%, 0.77%, 0.50%, 0.92%, 0.43%, and 0.13%, respec-
tively. These results are consistent with those generated
through the total differential method.
The results in Figure 8 demonstrate different system Figure 8. Results of substrate content under different scenarios in
responses under different combinations of sensitive and day.

548 Journal of the Air & Waste Management Association Volume 57 May 2007
Qin, Huang, Zeng, Chakma, and Xi

was conducted under uncertainties. Six input parameters ␮ss, ␮sm ⫽ microbial specific growth rates (1/h) at
(the maximum specific growth rate, half velocity for de- temperatures Ts and TM (oC),
gradable substrate, yield coefficient, initial biomass con- respectively
centration, water content, and temperature) were consid- Ts ⫽ preference temperature (oC)
ered to be of uncertain features (reflected as fuzzy TM, TL ⫽ medium threshold temperature (oC) and
membership functions in quasi-Gaussian shapes). The re- maximum allowable temperature (oC)
sults indicated that the uncertainties projected in input EA ⫽ activate energy of composting bulk (J/
parameters will result in significant derivations on system mol)
predictions. R ⫽ universal gas constant (J/mole䡠K)
Comparisons between FCPM and FV methods dem- w, wa ⫽ moisture content and critical moisture
onstrated that the FCPM can generate satisfactory system content
outputs, with less computational efforts being required. w1, w2 ⫽ suitability index and maximum allow-
Analyses on degree of influence of system inputs were also able index for water content
provided to describe the impacts of uncertainties on sys- Kparticle ⫽ particle size coefficient
tem responses. Thus, suitable measures can be adopted W ⫽ mass of water in the reactor (kg)
either to reduce system uncertainty by well-directed re- PI ⫽ a general term representing S, X, and W
duction of uncertainties of those high-influencing param- t ⫽ time period (h)
eters or to reduce the computational requirement by ne- ␭ ⫽ saturation ratio of vapour
glecting those negligible factors. M ⫽ total mass of the compost bulk (kg)
The developed FCPM is valuable in evaluating the j ⫽ saturated water vapor content (kg/Nm3)
dynamic derivations of composting parameters under un- ps, p0 ⫽ saturation vapor pressure (Pa) and air
certainty. The acquired information would be especially pressure (Pa)
useful for managers to determine whether fluctuations of Cc ⫽ heat capacity of compost bulk (kJ/ kg䡠°C)
the composting performances are because of unavoidable h1 ⫽ heat quantity generated by unit dry or-
system uncertainties or human-induced factors. Accord- ganic (kJ/kg)
ingly, proper controls can be taken for mitigating any h2 ⫽ potential heat of water evaporation (kJ/kg)
adverse effects. Ta ⫽ temperature of inflow air (°C)
Cw, Ca, Cs ⫽ heat capacity of water, air, and volatile
NOMENCLATURE organics (kJ/kg䡠°C)
K ⫽ thermal conductivity coefficient of the
composting facility (kJ/m2䡠h䡠°C)
F ⫽ total thermal dispersion area of com-
␮A (x) ⫽ fuzzy membership grade of fuzzy set A
post facility (m2)
fA ⫽ a real-valued function that is increasing
z0 ⫽ initial oxygen concentration (%)
and right-continuous
av ⫽ required oxygen amount
gA ⫽ a real-valued function that is decreasing
XIN ⫽ vector of uncertain system inputs/
and left-continuous
parameters
PN ⫽ number of evaluations for FV method
PIN ⫽ vector of deterministic system inputs
PA ⫽ fuzzy ␣-cut set
YOU ⫽ vector of system outputs
xi(j) ⫽ a fuzzy interval at the jth ␣-cut level for
m៮ , ␦ ⫽ mean value and the standard deviation
the ith fuzzy parameter
of Gaussian distribution
ai(j), bi(j) ⫽ lower and upper bounds of input values
P៮ ⫽ total differential at the peak point
at the jth ␣-cut level
␳ij ⫽ DI of input parameter i that contributes
ci ⫽ input-value vector at the jth ␣-cut level
j
to uncertainty of the model output j.
and ith input variable
Ti,j ⫽ the kth element of cij in FV method
k

r
Ii,j⫽ the rth element of cij in FFV method ACKNOWLEDGMENTS
␳i ⫽ DI for input parameter i This research was supported by the Major State Basic
␮s ⫽ specific growth rate (1/h) Research Development Program (2005CB724201 and
X ⫽ biomass concentration (kg/m3) 2005CB724207) and the Natural Science and Engineer-
S ⫽ substrate concentration (kg/m3) ing Research Council of Canada. The authors thank the
␮smax ⫽ maximum specific growth rate (1/h) anonymous reviewers for their helpful comments and
Ks ⫽ the half saturation constant suggestions.
b ⫽ endogenous decay constant (1/h)
rs ⫽ rate of substrate utilization (kg/m3䡠h) REFERENCES
rx ⫽ rate of biomass growth (kg/m3䡠h) 1. Keener, H.M.; Marugg, C.; Hansen, R.C.; Hotink, H. Optimizing the
Efficiency of the Composting Processes; In: Science and Engineering of
k⬘ ⫽ the maximum specific substrate-utiliza- Composting; The Ohio State University: Columbus, OH, 1993; pp 59-
tion rate (1/T) 94.
2. Kaiser, J. Modelling Composting as a Microbial Ecosystem: a Simula-
kd ⫽ microbial maintenance coefficient (1/h) tion Approach; Ecol. Model. 1996, 91, 25-32.
kT, kmoisture ⫽ temperature and moisture corrections, 3. Stombaugh, D.P.; Nokes, S.E. Development of a Biologically Based
respectively Aerobic Composting Simulation Model; Trans. ASAE 1996, 39, 239-
250.
kFAS, kO2 ⫽ free-airspace and oxygen-concentration 4. Mohee, R.; White, R.K. Simulation Model for Composting Cellulosic
corrections (Bagasse) Substrates; Compost Sci. Util. 1998, 6, 82-92.

Volume 57 May 2007 Journal of the Air & Waste Management Association 549
Qin, Huang, Zeng, Chakma, and Xi

5. Bari, Q.H.; Koenig, A. Effect of Air Recirculation and Reuse on Com- 25. Zadeh, L.A. The Concept of a Linguistic Variable and Its Application to
posting of Organic Solid Waste; Resour. Conserv. Recy. 2001, 33, 93- Approximate Reasoning; J. Inform. Sci. 1975, 8, 199-249.
111. 26. Dong, W.; Shah, H.C. Vertex Method for Computing Functions of
6. Xi, B.D.; Huang, G.H.; Qin, X.S.; Liu, H.L. Two Stages Kinetics of Fuzzy Variables; Fuzzy Set. Syst. 1987, 24, 65-78.
Municipal Solid Waste Inoculation Composting Processes; J. Environ. 27. Wood, K.L.; Otto, K.N.; Antonsson, E.K. Engineering Design Calcula-
Sci. 2004, 16, 520-524. tions with Fuzzy Parameters; Fuzzy Set. Syst. 1992, 52, 1-20.
7. Nakasaki, K.; Kato, J.; Akiyama, T.; Kubota, H. A New Composting 28. Hanss, M.; Klimke, A. On the Reliability of the Influence Measure in
Model and Assessment of Optimum Operating for Effective Drying of the Transformation Method of Fuzzy Arithmetic; Fuzzy Set. Syst. 2004,
Composting Materials; J. Fermentation Technol. 1987, 65, 441-447. 143, 371-390.
8. Das, K.; Keener, H.M. Numerical Model for the Dynamic Simulation of 29. Molz, F.J.; Widdowson, M.A.; Benefield, L.D. Simulation of Microbial
a Large Scale Composting System; Trans. ASAE 1997, 40, 1179-1189. Growth Dynamics Coupled to Nutrient and Oxygen Transport in
9. Isukapalli, S.S. Uncertainty Analysis of Transport-Transformation Models; Porous Media; Water Resour. Res. 1986, 22, 1207-1216.
Ph.D. dissertation; The State University of New Jersey: New Brunswick, 30. Nelson, D.L., Cox, M.M. Lehninger Principles of Biochemistry, 3rd ed.;
NJ, 1999. Worth Publishers: New York, NY, 2000.
10. Guo, H.C.; Liu, L.; Huang G.H. A Stochastic Water Quality Forecasting 31. Xi, B.D.; Liu, H.L.; Zeng, G.M.; Huang, G.H.; Bai, Q.Z. Composting
System for the Yiluo River; J. Environ. Inform. 2003, 1, 18-32. MSW and Sewage Sludge with Effective Complex Microorganisms; J.
11. Qin, X.S.; Huang, G.H.; Jiang, X.Y.; Xi, B.D.; Liang, Z.W.; Christine, Environ. Sci. 2002, 14, 264-268.
W.C. Fuzzy Approach for Dynamic Simulation of Composting Process 32. Yilmaz, A.H. Modelling of the Anaerobic Decomposition of Solid
under Uncertainty. Trans. Nonferrous Met. Soc. China 2004, 14, 18-24. Wastes; Energy Sources 2003, 25, 1063-1072.
12. Hu, Z.Y.; Chan, C.W.; Huang, G.H. Simulation of the Fate of Contam- 33. Haug, R. T. Development of Simulation Models. In The Practical Hand-
inant in Groundwater Under Uncertainty Using Optimized Linear book of Compost Engineering; Lewis Publishers: New York, NY, 1993; pp
Interpolation; J. Environ. Inform. 2005, 6, 93-102. 385-436.
13. Huang, G.H.; Moore, R.D. Grey Linear Programming, Its Solving Ap- 34. Xi, B.D.; Huang, G.H.; Qin, X.S.; Liu, H.L. Two Stages Kinetics of
proach, and Its Application; Int. J. Systems Sci. 1993, 24, 159-172. Municipal Solid Waste Inoculation Composting Processes; J. Environ.
14. Luo, B.; Yin, Y.Y.; Huang, G.H.; Huang, Y.F. Uncertainty Analysis for Sci. 2004, 16, 520-524.
Distribution of Greenhouse Gases Concentration in Atmosphere; J. 35. Xi, B.D.; Wei, Z.M.; Liu, H.L. Dynamic Simulation for Domestic Solid
Environ. Inform. 2004, 3, 89-94. Waste; J. Am. Sci. 2005, 1, 34-45.
15. Yang, X.H.; Yang, Z.F.; Shen, Z.Y. GHHAGA for Environmental Sys-
tems Optimization; J. Environ. Inform. 2005, 5, 36-41.
16. Lahkim, M.B.; Garcia, L.A;. Stochastic Modeling of Exposure and Risk
in a Contaminated Heterogeneous Aquifer, 1: Monte Carlo Uncer-
tainty Analysis; Environ. Eng. Sci. 1999, 16, 315-328.
17. Li, J.B.; Huang, G.H.; Chakma, A.; Zeng, G.M.; Liu, L. Integrated
Fuzzy-Stochastic Modeling of Petroleum Contamination in Subsur-
face; Energ. Source. 2003, 25, 547-563.
About the Authors
18. Wood, K.L.; Antonsson, E.K. Modeling Imprecision and Uncertainty Guohe Huang is a research professor and an environmental
in Preliminary Engineering Design; Mech. Mach. Theory 1990, 25, scientist (Canadian Research Chair, Tier 1) at Environmen-
305-324. tal System Engineering Program, Faculty of Engineering,
19. Klir, G.J. The Role of Constrained Fuzzy Arithmetic in Engineering. In
Uncertainty Analysis in Engineering and Sciences: Fuzzy Logic, Statistics,
University of Regina. Xiaosheng Qin is the Ph.D. candidate
and Neural Network Approach; Ayyub, B.M.; Gupta, M.M., Eds.; Kluwer at Environmental System Engineering Program, Faculty of
Academic Publishers: Norwell, MA, 1997; pp 1-19. Engineering, University of Regina. Guangming Zeng is the
20. Sasikuman, K.; Mujumdar, P.P. Fuzzy Optimization Model for Water research professor (head of department) at the College of
Quality Management of a River System; J. Water Resour. Plng. Mgmt.
1998, 124, 79-88.
Environmental Science and Engineering, Hunan University.
21. Bosma, W.J.P., Bellin, A.; Rinaldo, A. Linear Equilibrium Adsorbing Amit Chakma is a research professor and an environmen-
Solute Transport in Physically and Chemically Heterogeneous Porous tal/chemical scientist at the Department of Chemical Engi-
Formations, 2, Numerical Results; Water Resour. Res. 1993, 29, 4013- neering, University of Waterloo. Beidou Xi is an environ-
4043.
22. Dou, C.; Woldt, W.; Bogardi, I.; Dahab, I. Steady State Groundwater
mental researcher majoring in solid waste management at
Flow Simulation with Imprecise Parameters; Water Resour. Res. 1995, the Chinese Research Academy of Environmental Sciences.
31, 2709-2719. Address correspondence to Guohe Huang, Faculty of Engi-
23. Zeng, Y.; Trauth, K.M. Internet-Based Fuzzy Multicriteria Decision neering, University of Regina, Regina, Saskatchewan, Canada
Support System for Planning Integrated Solid Waste Management; J.
Environ. Inform. 2005, 6, 1-15.
S4S 0A2; phone: ⫹1-306-585-4095; fax: ⫹1-306-585-4855;
24. Hanss, M. The Transformation Method for the Simulation and Anal- e-mail: huangg@env.uregina.ca.
ysis of Systems with Uncertain Parameters; Fuzzy Set. Syst. 2002, 130,
277-289.

550 Journal of the Air & Waste Management Association Volume 57 May 2007

You might also like