Lactic Acid Recovery Process by Ion-exchange Resin: Modeling€¦ · acid recovery is proliferating [2-8]. Currently, commercial recovery of lactic acid by ion-exchange resin exists

1

Lactic Acid Recovery Process by Ion-exchange Resin:

Modeling

Archw Promraksab) and Chairat Siripatanaa)

School of Engineering and Resources, Walailak University, 80161, Nakhon Si Thammarat, Thailand

a)Corresponding author: [email protected]

b)[email protected]

Abstract. A very flexible model for studying lactic acid adsorption by ion-exchange resin was developed. The model is based on generalized volumetric dispersion formulation (VDF) for fixed-bed adsorption and desorption. The Crank-Nicolson finite difference technique was employed to obtain a numerical solution for different resins, feed conditions, column’s geometry, packing compactness etc. The effects of resin selectivity (sorption isotherm), in-particle diffusion mechanism (mass transfer parameters), liquid phase dispersion (Peclet number), initial bed condition and external boundary conditions were included in the model. Furthermore, the model flexibility extends its applicability to cover all steps of ion-exchange process including acidification, lactic acid adsorption by basic sorbents, lactic acid desorption and resin regeneration, although suitable parameters are to be determined experimentally. The results of simulation for a few

number of cases demonstrate the potential of its application in studying the kinetics of lactic acid adsorption ion-exchange resin, optimizing the processes and scaling up to commercial scales. Furthermore, it is anticipated that the model could be used for describing solute mass transfer in supercritical extraction for fixed-bed configuration.

Keywords—Lactic acid; Ion-exchange; Adsorption; Desorption; Modeling

INTRODUCTION

Biodegradable plastics become more important alternatives to fossil-based plastics as the world is more

concerned on the environment. Among the fermentation-derived products, lactic acid, either in D- or L- form, has a

high potential for multi-million dollar market as monomers for producing the biodegradable plastics namely,

polylactic acid (PLA). The challenge and the key of success of the product hinges on the cost reduction in

fermentation and purification of lactic acid from the broth. It was estimated that the recovery and purification alone

attribute to almost 50% of the final product cost [1]. Although, lactic acid can be recovered by a few methods such as liquid extraction, electrodialysis and others, adsorption seems to be generally suitable due to its low cost and

simple in operation. With the requirement of well-purified product for production of high quality PLA, adsorption

followed by crystallization in the form of lactate salt is feasible.

Adsorption is a unit operation suitable for recovery of organic acids in dilute and complex aqueous solution such

as fermentation broth as indicated by numerous related publications. Recently, interest on its application for lactic

acid recovery is proliferating [2-8]. Currently, commercial recovery of lactic acid by ion-exchange resin exists

mainly for demineralization of thin crude lactic acid. The real use for fermentation broth is under development.

Evangelista and et al. [1] evaluated weak-, moderate-, and strong-basic resins for sorption capacities of lactic acid

from solution with different pHs. Langmuir isotherms and breakthrough curves indicated that the resin sorption

capacities, for all resins evaluated, decreased considerably as pH of the feed exceeded the pKa of lactic acid as a

result of the decrease of undissociated lactic acid concentration. Of the weak-base sorbent, VI-15 had a very good capacity although its potential disadvantage is the excessive swelling and shrinking during the process cycle. Later

in 1996, Evangelista and Nikolov [9] studied sorption process cycle of lactic acid recovery from fermentation broth

using weak base sorbent and the following scheme. Firstly the clear broth was acidified by weak-acid cation

exchanger (Dualite C-464) and went through the weak-base absorption column until the sorbent was exhausted.

Then the unbound component of the broth was removed by rinsing with pure water before eluting with methanol or

2

5% NH₄OH to recover absorbed lactic acid from the sorbents. However, as pointed by the authors that cation

exchanger employed was not strong commercial practicality and this step still produced waste salt during

regeneration of sorbent. Furthermore the resin (Vl-15) is not physically stable due to excessive swelling and

shrinking. More importantly, the broth components were not removed sufficiently during the rinse step, but eluted

with lactic acid during the desorption step. Therefore, broth pretreatment or polishing steps would be necessary to

improve the purity of the product.

In depth understanding of adsorption kinetics requires careful experiment and realistic mathematical models.

Mathematical models for fixed bed mass transfer were developed to use in many publications, which include the in-solid diffusion, solid geometry and longitudinal dispersion [10-15]. Zheng and et al. [16] used an exact solution for

fixed-bed adsorption given by [11] to fit breakthrough data for lactic acid adsorption at low concentration (< 10

mg/mL) where isotherm was essentially linear. Good agreement between data and model was observed. However,

all analytical solutions show similar restricted applicability due to the assumption of linear isotherm, uniform initial

concentration in both phase and constant mass transfer and dispersion parameters. To obtain a practical model for

studying the real sorption processes with more complicated isotherm and initial non-uniform concentration, more

involved models and numerical solution are inevitable.

The problem in lactic acid recovery by sorption process poses two challenges: firstly, the selection of suitable

resin and their sequences of operation; secondly, in depth understanding, control and process optimization require a

flexible and realistic model. In this paper, we formulate a general model, give numerical solutions, and compare

with a few data given by [1, 9].

MODEL DEVELOPMENT

The purpose of this work is to develop a very flexible model for dealing with a range of absorption and

desorption processes. Similar models have been developed and proposed by a number of authors however most of

the derivation were based on linear velocity in continuous phases which implied that the void fraction in the column

is uniform [16-19]. However in many cases such as supercritical extraction, the solid particles containing the solute to be extracted along the column change their size with time [20]. Moreover, during model development we keep in

mind that our model could be used to explain fixed-bed desorption in supercritical conditions such as extraction of

vitamins from palm oil and avocado oil by supercritical CO2 [21-22]. However, this paper only provides the

numerical solutions for lactic acid adsorption on basic resins which show good performance as investigated by

previous authors [1, 9, 23].

In contrast to traditional formulation, our differential element is based on volume rather than a differential length

in the column [24]. This is chosen to remove the inherit assumption of a constant linear velocities of each phase.

This means the column may have non-uniform cross-sectional and non-uniform void fraction.

Consider a mass balance of solute-solvent phase-adsorber subsystem k (k = 1, 2, 3, … , p) in a differential

volume ,L SdV dV as illustrated in Fig. 1

3

FIGURE 1. Mass balance in a differential volume ,L SdV dV

Here S and L are the volumetric flow rate (m3/s) of the solid and liquid phase respectively, kx and ky are the

solute (sorbate) concentration (kg/m3) in solid and liquid phase for each subsystem k respectively. Let,x kD and

,y kD

are the volumetric dispersion coefficient (m2/s) in solid and liquid phase respectively.

By standard analysis of differential control volume, the final differential equations for each pair of interacting components in both phases are obtained, as follows:

Liquid phase: ( ), *

,k

y kk k kL k k

L L L s

Dy y ydAL K y y

V V L V dV t

− − − − =

(1)

Solid phase: ( ),

,

x kk k kL k k k

s s s s

Dx x xdAS K y y

V V s V dV t

+ + − =

(2)

Here ,L kK is overall liquid phase mass transfer coefficient (m/s) of solute in subsystem k, A is mass transfer area

(m2), and t is processing time (s).

Introducing dimensionless variables t = whereas L L = is the average retention time of liquid

phase,L is the total volume (m3) of liquid phase retained in the column at any time.

L Lz V= is a dimensionless

column position, and , , ,

LL k L k a k

L

dAT K K

L dV

= = where

,a kK is mass transfer coefficient (s-1), ,a k L

L

dAK K

dV=

L

L S

dV

dV dV =

+

is fraction of void volume available for liquid flow through the column, and Peclet numbers (kP and

kR ) are defined for each phase as follow.

,

S

k

S k

SP

D

= ,

,

L

k

L k

LR

D

= ,

in another form of void fraction, it can be written as 1S

L

V

V

−=

After some manipulation, for mobile phase we have

,

1k k k kL k k

k k

y y x yT y

z z R z m

− + + − =

(3)

and for solid phase, fixed-bed sorption equation can be written as

,1

k kL k k

k

x xT y

m

− =

− (4)

with more simplified boundary conditions

,

1 kk in k

k

yy y

R z

= −

at z = 0 and

, , 0kk out k

yy y

z

= =

at z ≥ 1

and initial condition ( = 0)

( ),0k ky y z= , ( ),0k kx x z= and also ( ), ,k in k iny y t= can vary with time

The partition coefficient (m) depends on sorbate-sorbent interaction and other physiochemical factors.

Evangelista and et al. [1] used Langmuir model to describe the competition between lactic acid and n molecules of

water for a basic site on the sorbent. The following relation was obtained. It must be emphasized that K ,HLaC ,

mq ,

thus q and m vary with time () and position (z) within the bed.

4

( )1

HLa

m HLa

KCq

q KC=

+

(5)

where q = composition uptake of sorbate (mg/g)

mq = total sorbent capacity for lactic acid (mg/g)

K = association constant of sorbent-lactic acid complex (mL/mg)

HLaC = equilibrium concentration in the bulk fluid (mg/mL)

and

( )1

m s

HLa HLa

Kqx qm

y C KC

= = =

+ (6)

where s is the density of resin (g/mL).

ANALYSIS OF SOLUTIONS

The parameters mq and K which are used for resins which potentially used in lactic acid recovery to simulate

the breakthrough curves in this work are listed in Table 1. Crank-Nicolson finite difference was used according to

our previous work [24].

TABLE 1. Properties of selected basic adsorbents used by Evangelista and et al. [1]

Resin Type/Matrix Functional

group

apK

mq

(mg/g dry resin)

K (mL/mg)

VI-15 Gel/methylene-bis-Acrylamide

Imidazole 6.9 280 2.2

MWA-1 Macro porous/SDVB 30-amine 8.8 365 8.4

Stability Constraints

The simulations were stable with excellent accuracy for the following practical range: z ≤ 1/50, t ≤ 0.001/, 0

≤ TS ≤ 100, 0.01 ≤ R ≤ 100, 0.01 ≤ m ≤ 50. However, there restrictions will not limit its applicability in most

practical cases.

Analysis of Simulated Breakthrough Curves

Breakthrough curves were obtained by retrieving the concentration in mobile phase at the outlet versus time. The

results a specific case (p = 1) were summarized below.

The Effect of Association Constant

For the resins that strictly follow Langmuir isotherm, high association constant ( K ) means m approaches

m HLaq C which is strongly dependent on HLaC Thus, in general, partition coefficient m (the relative absorption

capacity of the resin) changes with time and position. High K means higher adsorption capacity and more delayed

and steeper breakthrough curve as shown in Fig. 2. Higher K has a similar effect with slightly different behavior.

This demonstrates the importance of resin isotherm characteristics as they interact with different mobile phases

during various steps in process cycle. In general, if the isotherm for each step, e.g. acidification, adsorption, rinsing,

resin regeneration, is established, prediction of breakthrough curve is practical and process optimization is feasible.

5

FIGURE 2. Simulated breakthrough curves showing the effect of association constant (K)

( = 12 mL, = 0.3, = 15 min, qm = 280 mg/g, TL = 10, R =10)

The Effect of Longitudinal (Axial) Dispersion

In general, axial dispersion reduces the efficiency of the sorption process by lowering concentration gradient

without the loss of resin capacity. For higher dispersion (lower Peclet number, R) the tail of breakthrough curve

becomes less steep although its starting position is affected slightly as shown in Fig. 3. In practice, axial dispersion

is related to flow rate and physical properties of the mobile phase, column size and geometry, surface and packing

characteristics of resin particles. These combined effects must be determined experimentally for each case.

Currently, no data available to predict the Peclet number for system under our consideration.

The Effect of Mass Transfer Coefficient and Liquid Retention Time

Fig. 4 demonstrates the effect of

LT which a combination of mass transfer coefficient and the average retention

time. In practice can be specified independently whereas aK is a complex function of interface area per unit

volume of resin, flow characteristic of mobile phase and in solid diffusion mechanism of solutes adsorbed and desorbed.

aK , thus, must be determined experimentally. As shown in Fig. 4, higher aK or lower flow rate of mobile

phase (increase retention time) enhances the efficiency of sorption process as seen by the steeper breakthrough

curve. However, the change in LT does not affect relative position of breakthrough curve, thus the resin capacity, as

indicated by a fixed turning point.

6

FIGURE 3. Simulated breakthrough curves showing the effect of longitudinal dispersion

( = 12 mL, = 0.3, = 15 min, qm = 280 mg/g, TL = 10, K = 2.2 mL/mg)

FIGURE 4. Simulated breakthrough curves showing the combined effect (

LT ) of mobile phase retention time and the effective

mass transfer coefficient

( = 12 mL, = 0.3, = 15 min, qm = 280 mg/g, R =10, K = 2.2 mL/mg)

The Effect of Void Volume

Void volume is the fraction of column space available for liquid flow through. Clearly, the average retention

time () is directly related to the void fraction () at a specific flow rate of mobile phase. Thus the effect of

increasing is essentially equivalent to increasing the retention time. However it should be realized that indirect

consequence of changing is to alter the liquid flow within the bed, thus affecting the degree of intra-particle

dispersion and interfacial mass transfer. Fig. 5 illustrates the effect of on the breakthrough curves. As expected,

lower void fraction associated with longer average retention time and less steep curves due to dispersion effect and

the breakthrough curves shift to the right.

7

FIGURE 5. Simulated breakthrough curves showing the effect of void fraction ()

( = 12 mL, = 15 min, qm = 280 mg/g, TL = 10, R =10, K = 2.2 mL/mg)

Represented Breakthrough Curves for VI-15 and MWA-1

Based on isotherms and operating condition given in [1] together with rough fit additional parameter estimated

here for illustration, we obtain the breakthrough curves in Fig. 6. Clearly, the results of simulation compare well

with their data. Unfortunately, it was not possible to use their data for detailed comparison because of limited

information related to basic resin and flow parameters. Nevertheless, the model can explain the effect of different

isotherm quite clearly as discuss previously. The comparison also indicates that the approximateLT and R are the

same order of magnitude with moderate mass transfer capacity.

FIGURE 6. Model versus experimental breakthrough curves for V1-15 and MWA-1 [1]

(VI-15: = 12 mL, = 0.3, = 10 min, K = 2.2 mL/mg, qm = 280 mg/g, TL = 12, R =12)

(MWA-1: = 12 mL, = 0.7, = 10 min, K = 8.4 mL/mg, qm = 365 mg/g, TL = 8, R = 8)

The Effluent Profile for Adsorption-Washing-Desorption Cycle (MWA-1 Resin, Washed by

Water and Desorbed by Methanol)

8

Consider an adsorption-washing-desorption cycle having three components in liquid phase namely; lactic acid

(solute), water and methanol. Assuming that no transfer of water and methanol across solid-liquid interface occurs,

there exist three solute-mobile-adsorber subsystems (p = 3), which resulting in six partial differential equations.

However, based on no-transfer assumption of water and methanol across phases as stated above, after some manipulation and simplification, the following final equations were obtained.

Liquid phase: 1

L

y y x yT y

z z R z m

− + + − =

(7)

Solid phase: 1

L

x xT y

m

− =

− (8)

Liquid phase dispersion: 1w w w

z z R z z

− + =

(9)

Here, x and y are the solute concentration (g/mL) in solid and liquid phase respectively, w is the concentration

(g/mL) of water in liquid phase at any time and position, m is the average partition coefficient calculated

dynamically from the following relation:

1

m s L HLam

L HLa HLa L HLa

Kq w Cwm m

C KC C

− −= +

− + −

(10)

where mm is partition coefficient (dimensionless) for lactic acid-methanol–resin subsystem, and L , s are the mass

density in (g/mL) of liquid and solid (resin) phases respectively.

The effluent profile in Fig. 7 was obtained by using (7) to (10), with suitable boundary and initial conditions and

Crank-Nicolson finite-difference method [24]. The results of simulation follow the trend of the data from [9] very

well. Unfortunate currently detailed comparison was not possible due to the lack of required model parameters.

FIGURE 7. Simulated effluent profile for adsorption, washing, and desorption

( = 12 mL, = 0.7, = 10 min, K = 8.4 mL/mg, qm = 365 mg/g, TL = 2, R =2)

CONCLUSION

A robust volumetric dispersion model was developed and its numerical solutions were in good agreement with a

represented published data of previous work [1]. Most parameters and variables can realistically vary with operating

time and column position, including sorption isotherm, mass transfer related parameters ( ),L LdA dV K , operating

9

variables ( ), , L , and degree of longitudinal dispersion ( ),R P . Finite difference technique allows arbitrary initial

concentration distributions in both mobile and sorbent phases, non-uniform column cross-section areas, variable

void fraction and even section of different resins packed in the same column. The flexibility of the model with

suitable required parameters extends the possibility of applying the solution for the whole cycle of sorption process,

e.g. acidification by cation exchanger, lactic acid adsorption by basic sorbents, lactic acid desorption and resin

regeneration. The obstacle of applying to model to real processes is the lack of publish design parameters for

specific cases. In practice, thus is must be done experimentally by batch sorption experiment or model-fitting with

the data of sorption processes. Once sufficient data are available, the model will simplify the optimization, control

and scale-up of the sorption processes. More importantly, the model allows deeper understanding of sorption

kinetics due to the underlining founded theoretical basis. Furthermore, it is also anticipated that the model could be

used for describing solute mass transfer in supercritical extraction for fixed-bed configuration.

ACKNOWLEDGMENT

This research was carried out under the financial support from Walailak University with contract number

WU59114.

REFERENCES

[1] R. L. Evangelista, A. J. Mangold, and Z. L. Nikolov, “Recovery of lactic acid by sorption: Resin evaluation,” Appl. Biochem. Biotechnol., 45/46, pp. 131-144, 1994.

[2] M. J. Dethe, K. V. Marathe, and V. G. Gaikar, “Adsorption of lactic acid on weak base polymeric resins,” Sep. Sci. Technol., 41, pp. 2947-2971, 2006.

[3] S. S. Bayazit, I. Inci, and H. Uslu, “Adsorption of lactic acid from model fermentation broth onto activated carbon and Amberlite IRA-67,” J. Chem. Eng. Data, 56, pp. 1751-1754, 2011.

[4] J. Quintero, A. Acosta, C. Mejia, R. Rios, and M. Torres, “Purification of lactic acid obtained from a fermentative process of cassava syrup using ion exchange resins,” Rev. Fac. Ing-Univ. Antioq., 65, pp. 139-151, 2012.

[5] W. Sodsai and T. Sookkumnerd, “Modeling of lactic acid adsorption isotherm by anion exchange resin Amberlite IRA-96,” KMITL Sci. Technol. J., 13, pp. 82-86, 2013.

[6] M. Bishai, S. De, B. Adhikari, and R. Banerjee, “A platform technology of recovery of lactic acid from a fermentation broth of novel substrate Zizyphus Oenophlia,” 3 Biotech, 5, pp. 455-463, 2015.

[7] T. Rampai, S. Thittiprasert, W. Boonkong, K. Kodama, V. Tolieng, and N. Thongchul, “Improved lactic acid productivity by simultaneous recovery during fermentation using resin exchanger,” Asia-Pac. J. Sci. Technol., 21, pp. 193-199, 2016.

[8] A.Yousuf, F. Bonk, J. R. B. Oyanedel, and J. E. Schmidt, “Recovery of carboxylic acids produced during dark fermentation of food waste by adsorption on Amberlite IRA-67 and activated carbon,” Bioresour. Technol., 217, pp. 137-140, 2016.

[9] R. L. Evangelista and Z. L. Nikolov, “Recovery and purification of lactic Acid from fermentation broth by adsorption,” Appl. Biochem. Biotechnol., 57/58, pp. 471-480, 1996.

[10] A. Rasmuson and I. Neretnieks, “Exact solution of a model for diffusion in particles and longitudinal dispersion in packed beds,” AIChE J., 26, pp. 686-690, 1980.

[11] D. M. Ruthven, “Principles of Adsorption and Adsorption Processes,” Wiley, New York, 1984. [12] K. Shams and A. Fayazbakhsh, “Dynamics of reactive chromatographic columns of inert core/hollow/film

coated spherical packing: An analytical solution and applications,” J. of Chromatogr. A, 1370, pp. 93-104, 2014.

[13] W. Lemlikchi, N. Drouiche, N. Belaicha, N. Oubagha, B. Baaziz, and M. O. Mecherri, “Kinetic study of the adsorption of textile dyes on syntetic hydroxyapatite in aqueous solution,” Ind. Eng. Chem., 32, pp. 233-237, 2015.

[14] O. Kitous, N. Abdi, H. Lounici, H. Grib, N. Drouiche, E. H. Benyoussef, and N. Mameri, “Modeling of the adsorption of metribuzin pesticide onto electro-activated granular carbon,” Desalin. Water Treat., 57, pp. 1865-1873, 2016.

[15] S. Qamar, N. Akram, and A. S. Morgenstern, “Analysis of general rate model of linear chromatography considering finite rate of adsorption and desorption steps,” Chem. Eng. Res. Des., pp. 13-23, 2016.

[16] Z. Huang, X. H. Shi, and W. J. Jiang, “Theoritical models for supercritical fluid extraction,” J. Chromatogr. A, 1250, pp. 2-26, 2012.

10

[17] A. Rai, K. D. Punase, B. Mohanty, and R. Bhargava, “Evaluation of models for supercritical fluid extraction,” Int. J. Heat Mass Tran., 72, pp. 274-287, 2014.

[18] G. Sodeifian, S. A. Sajadian, and N. S. Ardestani, “Experimental optimization and mathematical modeling of the supercritical fluid extraction of essential oil from Eryngium billardieri: Application of simulated annealing (SA) algorithm,” J. supercrit. Fluids, 127, pp. 146-157, 2017.

[19] S. C. Kupski, E. J. Klein, E. A. Silva, F. Palu, R. Guirardello, and M. G. A. Vieira, “Mathematical modeling of supercritical CO2 extraction of hops (Humulus lupulus L.),” J. supercrit. Fluids, 130, pp. 347-356, 2017.

[20] S. Samadi and B. M. Vaziri, “Two- structured solid particle model for predicting and analyzing supercritical extraction performance,” J. Chromatogr. A, 1506, pp. 101-108, 2017.

[21] M. A. Lima, D. Charalampopoulos, and A. Chatzifragkou, “Purification of supercritical-fluid carotenoid-rich extracts by hydrophobic interaction chromatography,” Sep. Purif. Technol., 203, pp. 1-10, 2018.

[22] S. C. Corzzini, H. D. Barros, R. Grimaldi, and F. A. Cabral, “Extraction of edible avocado oil using supercritical CO2 and a CO2/ethanol mixture as solvents,” J. Food Eng., 194, pp. 40-45, 2017.

[23] Y. Zheng, X. Ding, P. Chen, C. W. Yang, and T. Tsao, “Lactic acid fermentation and adsorption on PVP,” Appl. Biochem. Biotechnol., 57/58, pp.627-632, 1996.

[24] C. Siripatana, H. Thongpan, and A. Promraksa, “A generalized volumetric dispersion model for A large class of

two-phase separation/reaction: Finite difference solutions,” J. Phys. Conf. Ser., 820, p. 012015, 2017.

11

Monod-Type Two-Substrate Models for Batch Anaerobic

Co-Digestion

Nirattisai Rakmak1, b), Laddawan Noynoo1), Sunwanee Jijai2) and Chairat

Siripatana1,a)

1Biomass and Oil-Palm Excellence Center and School of Engineering and Resources, Walailak University, Nakhon

Si Thammarat, Thailand 2Faculty of Science Technology and Agriculture, Yala Rajabhat University, Yala, Thailand

a) Corresponding author: [email protected]

b) [email protected] c) [email protected] d) [email protected]

Abstract. This paper attempts to provide a solution to the problems occurred in interpreting batch anaerobic co-digestion data using Monod approach by extending the simple Monod model to cover two/multiple substrates having distinct

characteristics and microbial preference. The ultimate aim is to obtain kinetic parameters that can be related reactor design and anaerobic digestion (AD) process performance in pilot and production scales. An assessment was carried out on the effect of pig manure and food waste ratio on the anaerobic co-digestion (ACoD) process, in batch reactors with a hydraulic retention time of 30 days. The experimental data will be fitted and described by 3 models including Simple Monod kinetics (SM), Simple Monod two-substrate model (SMTS), and Monod two-substrate (MTS) model with intermediate (MTSI). The results concluded that MTSI model is better for performance evaluation in ACoD process.

Keywords— anaerobic co-digestion; batch anaerobic digestion; monod two-substrate model; biogas kinetics

INTRODUCTION

Anaerobic co-digestion (ACoD) provides a few more degree of freedom to control and optimization of anaerobic

digestion process [1]. If properly use, it becomes a valuable tool to manipulate the process because of the following

reasons: provide deficit nutrients, synchronize the growth the microbial consortium, help in pH regulation and

reduce the effect of toxic substances [2-4]. However from the modeling point of view, co-digestion increases the

complexity of anaerobic digestion (AD) processes and thus more elaborated models are needed for process

simulation, parameter estimation, stability analysis, process control and optimization [5-6]. In conventional setup for

co-digestion studies, researchers often start with many batch experiments to determine biochemical methane

potential (BMP) and specific methane activity (SMA) for various co-digestion conditions. The accumulated methane

vs time curves are then plotted and the data sets are fitted to some models, most probably the modified Gompertz

equation. Then BMP and/or SMA is calculated from its parameter the ultimate methane produced. The Gompertz

equation is expresses the methane generated vs time as follows:

(1)

Where and are methane production rate and time lag respectively.

12

Gompertz equation has become an empirical model for microbial product accumulation data although a

theoretical basis for AD is provided by Gompertz [7] and other researchers [8-13]. The main advantage of Gompertz

equation is it can empirically represent large amount of BMP data very well, particularly for BMP experiments with

single and easily digestible substrates. For ACoD with more complex substrate the fitting is often not satisfactory

due to non-smoothing curves occurring when the microbes switch to different substrates after the preferred substrates are exhausted. Similarly, simple Monod model [14] were used satisfactorily, although not as widely as the

Gompertz counterpart because of the difficulty in model fitting and thus parameter estimation. However, Monod-

type models is more interpretive and more suitable for propose of deeper insight, AD process design and operation,

control and optimization. However, simple Monod model, which is based on single limiting substrate, suffers from

the same problems as Gompertz equation does if applied to ACoD data.

This article attempts to provide a solution to the problems occurred in interpreting batch ACoD data using

Monod approach by extending the simple Monod model to cover two/multiple substrates having distinct

characteristics and microbial preference. The ultimate aim is to obtain kinetic parameters that can be related reactor

design and AD process performance in pilot and production scales.

TYPES OF ACCUMULATED BIOGAS EVOLUTION (ABE) CURVES

In this article we target four types of accumulated biogas evolution (ABE) curves as depicted in Fig. 1. The

basis of these four different curve types can be summarized as follows:

Type I: Single substrate consumed by single group of microorganisms.

Type II: Multiple substrates consumed in parallel or sequentially by one or multiple groups of microorganisms.

Type III: Multiple substrates but can be simplified by three categories: easily degradable, slowly degradable and

intermediates.

Type IV: Multiple substrates with complex chain of consumption by groups of microorganisms.

Time

Acc

um

ula

ted

Bio

ga

s

Type I

Type IIType III

Type IV

FIGURE 1. Four type of accumulative biogas curves (ABE)

Simple Monod Kinetics (SM)

Simple Monod kinetic (SM) model assumed that the microbial growth is limited by single substrate without any

kind of inhibition and endogenous metabolism. Its development and solution were provided by Reference [15]. The

governing ordinary differential equations (ODEs) are:

(2)

(3)

(4)

13

Here are the concentration of biomass, concentration of a limiting substrate, and product (biogas)

respectively. are the maximum specific growth rate, saturation constant, biomass yield and product

yield factors respectively. All four parameters are assumed constant.

With proper initial condition, these systems of ODE can be integrated analytically and the following solution, in

term of accumulative biogas is obtained (add citations):

(5)

where , is hypothetical biogas generated due to previous growth up to the starting experimental

time and is the ultimate amount of biogas generated when the digestible substrate is completely consumed.

This simplest Monod model was used as an alternative to Gompertz equation albeit much less often presumably

due to difficulty in applying equation (5) to fit experimental data and to express explicitly in term of time . The

main advantage of this simplest form of Monod model is its mechanistic meaning, making it suitable from design,

control and optimization of AD processes. Like the Gompertz counterpart, to fit equation (5) to experimental data

only ABE curves are needed, avoiding laborious biomass and time-course COD analysis. By fitting it with biogas

data, we can also determine and directly (if is known).

Simple Monod Two-Substrate Model (SMTS)

SM model is only suitable for describing type I of ABE curves because of its underlining assumptions. For

ACoD whereby multiple substrates are consumed in parallel from a previous work [16] extended SMTS model by

dividing substrates into two distinct entities: easily and slowly degradable substrates (ED and SD). They made the same assumptions as SMTS models but considered substrates as two types with one group of microbes acting on

them. The model assumed that while it grows on ED substrate it only hydrolyses SD into ED without growing on

SD substrate. The ODEs for this model are summarized as follow.

Product formation: (6)

Easily degradable substrate: (7)

Slowly degradable substrate: (8)

Where the concentration of ED substrate is, is the concentration of SD substrate. are hydrolysis

rate constant, saturation constant of SD substrate and ED substrate respectively.

This model requires information about the amount of initial substrates in order to obtain the solution. It was

used successfully in describing ABE data for co-digestion of wastewater from pig farm and domestic organic waste

which was characterized as type-II curves. However, there were some experimental data, which exhibit two plateaus (type III), did not agree well with this model. The problem will address in our proposed model.

Monod Two-Substrate Model with Intermediate (MTSI)

Type III and IV ABE curves can be addressed in multiple-substrate approach using switching or preference

function g which is introduced into the model to describe how each group of microbes deals with

multiple substrates according to its preference. Intuitively, substrate preference is a function of other

physico-chemical and biochemical conditions. However, to keep the model simple, we will consider only the cases

where we can assume that g is a function of or or or only. In addition, we relax a few

assumptions restricted in SM and SMTS, they are:

14

(1) Now endogenous metabolism is included.

(2) We add a so-called intermediate obtained from in hydrolysis step, waiting to be consumed by the

microbes .

(3) There are two groups of microorganisms (consumes and ) and (grows only on ).

Based all these assumptions, the following ODEs can be written.

(9)

(10)

(11)

(12)

(13)

(14)

where , and are the specific death rate of and

respectively. are the corresponding yield

coefficient as specified by the subscripts, and and are conversion factors for and

respectively.

In this article we consider 5 variances of preference function, they are:

Model 0 (MTSI): parallel/independent consumption g (15)

Model I (MTSI-I): g is only a function of time g (16)

Model II (MTSI-II): g is a function of g (17)

Model III (MTSI-III): g is a function of g (18)

Model IV (MTSI-IV): g is a function of g (19)

Model V (MTSI-V): g is a function of (20)

where is the substrate concentration ratio and c is its critical value (a parameter to be estimated).

All variant of g can be visualized in Fig. 2.

15

FIGURE 2. Graphical representation of g

These variants of preference functions give slightly different results (not shown in this article to save space)

when applied to the ABE data demonstrated in this paper. Thus we will only show how well equation (20) fit the

current data. In addition, equation (20) and (17) are equivalent because the relation . So it can be

used interchangeably.

METERIAL AND METHOD

400-ml-working-volume serum bottles were used as reactors. Batch experiments were carried out at different

ratio of pig manure and food waste with initial pH adjustment to 7+0.2 by added NaOH. The N2 gas is used in

flushing over the headspace thus remove the trace of oxygen to ensure anaerobic condition. The serum bottles were

covered with the rubber stoppers and sealed with aluminum caps. All experimental was analysis following our

experimental previous work [16].

The experimental data will be fitted using Monod approach by extending the simple Monod model (SMTS and

MTSI) to cover two/multiple substrates having distinct characteristics and microbial preference. The models are

implemented by Python language coupled with least square optimization routines to estimate kinetic parameters that

can be related reactor design and AD process performance in pilot and production scales. Moreover, the experimental results of Reference [17] will be fitted and described by MTSI model too.

RESULTS AND DISCUSSION

For illustration, the first set of ABE data was taken from our experimental data [16] where pig manure was co-

digested with food waste in a series of batch experiments. The experiments were designed to optimize the digestion

ratio between pig manure (M) and food waste (W) at room temperature (approximately ) and initial

neutral pH. The experimental conditions and parameters estimated for SMTS and MTSI are summarized in Table 1

and Table 2 respectively. SMTS and MTSI model predicted versus experimental results of accumulated biogas is

showed in Fig. 3 and Fig. 4 respectively.

TABLE 1. The experimental conditions and parameters estimated from SMTS model

Parameter Ratio of pig manure (M) and food waste (W)

0:100 25:75 50:50 75:25 100:0

( )1/m d 0.700 0.750 1.200 0.800 0.490

( )/SeK mg L 30,000 30,000 30,000 30,000 5,000

( )'

0P mL 40.0 400.0 400.0 5.0 40.0

( )/removedCOD mg L 26,000 47,400 41,200 137,700 58,600

f (no unit) 0.760 0.600 0.800 0.760 0.580

( )1/k d 0.035 0.030 0.055 0.025 0.053

16

2500

2000

1500

1000

500

0

0 5 10 15 20 25

Time (days)

Acc

um

ula

ted

bio

ga

s vo

lum

e (m

L)

75M:25W

0M:100W

100M:0W

50M:50W

25M:75W

FIGURE 3. SMTS model predicted vs experimental results of accumulated biogas from different M:W ratio.

(PSY = 0.02 ( )/ /mL mg L ,

SsK = 4,000 ( )/mg L )

Firstly, in most cases SMTS fit the experimental data very well and better than MTSI (Fig. 4) except for

75M:25W ratio where the ABE curve exhibit type-IV behavior: three-substrates in appearance. When the ABE

curves fell into type I and II (50M:50W and 100M:0W), SMTS is sufficient to describe the ABE curves. In this case,

based on the model assumption for SMTS there are two groups of substrate present in the digestate and both

consumed by only one entity of microbe. While the first substrate is consumed and produced methane as the main

product, the second substrate is converted to first substrate and fed into consumption pipeline. However, SMTS

does not take into account the preference of one substrate over the other so that it cannot represent the ABE curves

with exhibit more than one plateaus as appeared in ratio of 0M:100W, 25M:75W and 75M:25W.

According to the explanation of Reference [16] pure pig manure (100M:0W) gave lowest methane yield

although they had highest initial COD, but the COD removed is the smallest one. This can be explained by high

percentage of easily digestible substrate it contained (> 0.9) from which the acid accumulation was so high, causing

pH to drop quickly and greatly slow down the activities of methanogens. In spite of having fraction of easily

digestible substrate included in the model, its estimation is not accurate because we cannot

observe clear plateaus in type-II ABE curves.

TABLE 2. The experimental conditions and parameters estimated from MTSI model

Parameter Ratio of pig manure (M) and food waste (W)

0:100 25:75 50:50 75:25 100:0

( )1

me d − 1.000 1.000 1.150 0.550 0.820

Sef (no unit) 0.940 0.440 0.940 0.650 0.750

( )/SeK mg L 25,000 25,000 19,800 30,000 30,000

( )/SiK mg L 15,000 10,000 15,000 10,000 10,000

( )/SsK mg L 10,000 10,000. 10,000 8,000 8,000

( )( )/ /PSeY mL mg L 0.018 0.023 0.026 0.022 0.022

( )( )/ /PSiY mL mg L 0.023 0.015 0.022 0.015 0.020

( )/COD mg L 58,600 47,400 40,700 137,700 24,150

Ssf (no unit) 0.230 0.360 0.230 0.230 0.230

( )0 /X mg L 150 1195 1300 300 300

cf (no unit) 1.000 1.002 1.001 1.001 0.900

R2 0.9864 0.9439 0.9494 0.9848 0.9895

17

3000

2500

2000

1500

1000

500

00 5 10 15 20 25 30

Time (days)

Acc

um

ula

ted

bio

gas

volu

me

(mL

)

75M:25W

0M:100W

100M:0W

50M:50W

25M:75W

FIGURE 4. MTSI model predicted vs experimental results of accumulated biogas from different M:W ratio.

(ief = 1.000,

edk = 0.043 d-1,

SsXf = 0.700, isf = 1.000,

XeSeY =XeSiY = 0.152,

XeSsY = 0.215, PSeSsY = 1.000,

Xsf = 0.100, =

100)

When there were more than one observed plateaus (eg. for 0M:100W, 25M:100W, 75M:25W, it was relatively

easy to estimate 1e sf f= − but only MTSI model can represent type-III and -IV ABE curves, thus it is more suitable

for estimating ef in these cases. Using MTSI for 0M:100W, it was found that 0.75ef in food waste. From this set

of batch experiments, the role of slowly degradable fraction is very apparent. The success of 75M:25W batch can be

described as a proper balance of essential nutrients as well as the balance between and acidogenesis of ED substrate

(which produces VFA) and methanogenesis of VFA (which increase alkalinity). As SD substrate is high (but not too

high) and the hydrolysis rate of macromolecules are much slower than the acid production rate, hydrolysis helped to

control VFA by limiting the level of ED substrate while methane producing by methanogens consumes VFA and bring pH up. This optimal synchronization seemed to help 75M:25W batch produce methane in the highest amount.

However, one should also note that, in this case, C/N ratio for pig manure and food waste were 11-13 and 17-20

respectively. Co-digestion will bring C/N ratio closer to the neighborhood of its optimal value.

Unfortunately models at this level of difficulty can neither include the mechanisms explaining how pH would

evolve during the AD process nor can it predict how C/N ratio would affect the AD performance. Furthermore, ABE

curves alone do not show the microbial activities directly although we can roughly infer the overall/net results of the

microbial activities at different time. However, for engineering purpose and with the simplicity of the experimental

setup and measurement, ABE curves with suitable (mechanistic) models like SM, SMTS and MTSI provide optimal

solution for design, scale-up, process prediction and operational adjustment of AD processes.

For illustration let us refer to AD batches in Fig. 4. again. Firstly, we must use our intuition to visually identify

the type of each AD batch. Then we apply a suitable model to the corresponding data to estimate the model

parameters. Finally, we use the chosen models and their parameters to describe the ABE data and infer the models to get the insight as far as the models allow.

Let us try to describe 75M:25W batch in Fig. 4. For this case, we chose MTSI model to represent the

experimental data and try to interpret the model representation accordingly. We accepted the two-substrate (with

intermediate) assumption as posed by the model although it would be more suitable to approximate as three

substrates rather than two substrates. Here, the model describes the ABE curve as long lag (about 5 days), followed

by a moderate methanogenic activity 10.55me d −= but reached the highest level of accumulated methane as

compared to the others. To explain why this was so happened, we need to look as the so-called lag period for this

case in more detail. Here, one striking fact reveals which was a moderate methanogenic activity in this lag period.

So, the high COD removal (> 90% of initial COD) for 75M:25W batch was interpreted as follows.

(1) The main reason for low CH4 yield in this AD system was the accumulation of VFA which rapidly brought down

pH to the level that methanogenic activity was essentially stopped.

18

(2) In case of 75M:25W batch, moderate production of CH4 helped to stabilize the pH of digestate so that

methanogenic activity can prolong further, resulting in highest COD removal and biogas yield.

(3) When ED substrate was exhausted, the microbes need time (approximately 3-5 days, according to the ABE

curve) to switch to what we designated as “intermediate”, then the methane production shot up again until SD

substrate was used up.

In fact, we can even discuss more detail using MTSI model and its parameters in Table 2. and have more

insightful discussion. However, it would not be suitable for this limited space. This warrants future work and

analysis.

Other case taken for discussion is due to Reference [17]. The experimental conditions and parameters estimated

for MTSI is summarized in Table 3 and some details are depicted in Fig. 5. Here the inoculum from two sources

were used to obtain ABE curves from POME: PP inoculum: a mixture composed by an anaerobic sludge coming

from the municipal Wastewater Treatment Plant and pig manure obtained from a pig processing plant in a

proportion 1:1 (v/v); LP inoculum: mixing between anaerobic sludge coming from the oxidation ponds from the

same palm oil mill industry. All of these data are of type-III so that as expected they can be represented very well

using MTSI model (Fig. 5). According to model fitting, the fraction of ED substrate was between

0.5 0.7ef which ensured slow release of the ED substrate and intermediate, sustaining good system stability.

However, high portion of SD substrate means longer time for COD removal (and CH4 evolution to complete).

This was very clear for LP inoculum (two-high curves with no appearing plateau on the second period) where the

COD removal was continuing even after 30 days, giving higher CH4 yields. Furthermore, it is clear from the ABE

data and parameters tabulated in Table. 3 that LP inoculum were much more active, having almost no lag and gave

higher final COD conversion. PP inoculum on the other hand, obtained from sub-optimal sources, although finally

could adjust to new environment did not show relatively high overall COD conversion (and so CH4 yield). The

second plateau only told us that no further COD removal was achieved after 30 days.

TABLE 3. The experimental conditions and parameters estimated from MTSI model due to Reference [17].

Parameter Inoculums

LP pH4.8 LP pH7 PP pH4.8 PP pH7

( )1

me d − 0.700 1.500 1.000 0.600

Sef (no unit) 1.000 0.300 1.000 0.750

( )/SeK mg L 3,000 7,000 4,000 5,000

( )/SiK mg L 2,200 10,000 3,000 7,000

( )/SsK mg L 2,200 7,000 3,000 5,000

( )( )/ /PSeY mL mg L 0.170 0.270 0.190 0.270

( )( )/ /PSiY mL mg L 0.100 0.270 0.210 0.214

Ssf (no unit) 0.300 0.440 0.300 0.500

( )0 /X mg L 32.00 1000.00 80.00 1000.00

Xsf (no unit) 0.500 0.100 0.500 0.100

cf (no unit) 0.985 1.002 0.990 1.000

R2 0.9884 0.9903 0.9956 0.9936

19

1000

800

200

00 5 10 15 20 30

Time (days)

Acc

um

ula

ted

bio

gas

volu

me

(mL

)

PP – pH 7

LP - pH 7

LP – pH 4.8

PP – pH 4.8

25

600

400

FIGURE 5. SM2SI model predicted versus experimental results of accumulated biogas from Reference [17].

(ief =1.000,

edk =0.043 d-1,

SsXf =0.700, isf =1.000,

XeSeY =XeSiY =0.152,

XeSsY =0.215, PSeSsY =1.000, COD =3,333 mg/L, =100)

CONCLUSION

ABE curves particularly from co-digestion experiment have a lot of insight to be explored. With a suitable

model (MTSI in our current article) and its well-fit parameters, we can interpret the curves in a present and

insightful way, even without other supplementary data. This will help us to obtain more information from BMP and

SMA experiment. Hopefully, this approach (Monod kinetics for AD batch experiments) will enable the engineers in

this field to obtain or relate these parameters to design and operational parameters. Ultimately, more confidence will

be achieved when they predict the performance of the commercial-scale biogas plant even using ABE data (with a

suitable model) alone.

ACKNOWLEDGMENT

This research was carried out under the financial support from Walailak University.

REFERENCES

[1] S. Xie, F. I. Hai, X. Zhan, W. Guo, H. H. Ngo, W. E. Price, and L. D. Nghiem, “Anaerobic co-digestion: A critical review of mathematical modelling for performance optimization,” Bioresource Technology, vol. 222, pp. 498-512, 2016.

[2] Md. N. I. Siddique and Z. Ab. Wahid.,”Achievements and perspectives of anaerobic co-digestion: A review,”

Journal of Cleaner Production, vol. 194, pp. 359-371, 2018.

[3] M. Islas-Espinoza, A. De las Heras, J. Vazquez-Chagoyán, and A. Salem, “Anaerobic cometabolism of fruit and vegetable wastes using mammalian fecal inoculums: fast assessment of biomethane production,” Journal of Cleaner Production, vol. 141, pp. 1411–1418, 2017.

[4] H. M. Jang, J. H. Ha, M.S. Kim, J.O. Kim, Y.M. Kim, and J.M. Park, “Effect of increased load of high-strength food wastewater in thermophilic and mesophilic anaerobic co-digestion of waste activated sludge on bacterial community structure,” Water Reseach, vol. 99, pp. 140–148, 2016.

[5] L. Jahn, T. Baumgartner, K. Svardal, and J. Krampe, “The influence of temperature and SRT on high -solid digestion of municipal sewage sludge,” Water Science and Technology, vol. 74, no. 4, pp. 836–843, 2016.

[6] M. Kouas, M. Torrijos, P. Sousbie, J. Harmand, and S. Sayadi, “Modeling the anaerobic co-digestion of solid

waste: From batch to semi-continuous simulation,”Bioresource Technology, vol. 74, pp. 33-42, 2019.

[7] C. Hu, B. Yan, K. Wang, and X.Xiao, “Modeling the performance of anaerobic digestion reactor by

the anaerobic digestion system model (ADSM),” Journal of Environmental Chemical Engineering, vol. 6, no. 2,

pp. 2095-2104, 2018.

20

[8] S. Sarto, R. Hildayati, and I. Syaichurrozi., “Effect of chemical pretreatment using sulfuric acid

on biogas production from water hyacinth and kinetics.” International Journal of Renewable Energy Research,

vol. 132, pp. 335-350, 2019.

[9] A. A. Rajput, Zeshan, C. Visvanathan., “Effect of thermal pretreatment on chemical composition, physical

structure and biogas production kinetics of wheat straw,” Journal of Environmental Management, vol. 221, pp. 45-52, 2018.

[10] M. X. Zheng, L.C. Schideman, G. Tommaso, W. T. Chen, Y. Zhou, and K. Nair. “Anaerobic digestion of

wastewater generated from the hydrothermal liquefaction of Spirulina: toxicity assessment and minimization,”

Energy Convers Manage., vol. 141, pp. 420–8, 2017. [11] J. Shen, H. Yan, R. H. Zhang, G.Q. Liu, and C. Chen. “Characterization and methane production of different

nut residue wastes in anaerobic digestion,” Renewable Energy., vol. 116, pp. 835–41, 2018. [12] D.D. Nguyen, S.W. Chang, S.Y. Jeong, J. Jeung, S. Kim, W. Guo, and H.H. Ngo. “Dry thermophilic semi-

continuous anaerobic digestion of food waste: performance evaluation, modified Gompertz model analysis, and energy balance. Energy Conversion and Management, vol. 128, pp. 203–210, 2016.

[13] M. Zwietering, I. Jongenburger, F. Rombouts, K. and Van’t Riet. “Modeling of the bacterial growth curve,” Apply and Environmental Microbiology., vol. 56, pp.1875–81, 1990.

[14] M. Zamanzadeh, W. J. Parker, Y. Verastegui, and J.D. Neufeld. “Biokinetic and molecular studies of methanogens in phased anaerobic digestion systems,” Bioresource Technology., vol. 149, pp. 318–326, 2013.

[15] S. Jijai, and C. Siripatana. “Kinetic Model of Biogas Production from Co-digestion of Thai Rice Noodle Wastewater (Khanomjeen) with Chicken Manure,” Energy Procedia.,vol. 138, pp. 386-392, 2017.

[16] R. Thongnan, H. Thongpan, N. Rakmak, and C. Siripatana. Modeling of anaerobic co-digestion of pig manure and domestic organic waste,” Jurnal Teknologi.,vol. 78, 5–6, pp. 117–124, 2016.

[17] N. Debora-Alcid. A. B. Ligia-Patrici, S. H. Diana-Milen, and D. Niño-Bonill. “Biogas production by anaerobic digestion of wastewater from palm oil mill industry,” Ciencia, Tecnología Y Futuro., vol. 5, no. 2, pp. 73-84, 2013.

21

Gompertz-Type Two-Substrate Models for Batch Anaerobic

Co-Digestion

Laddawan Noynoo1, Sunwanee Jijai2, Khunakon Phayungphan1, Nirattisai

Rakmak1, 3 and Chairat Siripatana1, 3, a)

1School of Engineering and Resources, Walailak University, Nakhon Si Thammarat, Thailand. 2Faculty of Science Technology and Agriculture, Yala Rajabhat University, Yala, Thailand.

3Biomass and Oil-Palm Excellence Center, Walailak University, Nakhon Si Thammarat, Thailand.

a) Corresponding author: [email protected]

Abstract. This paper attempts to address on how to improve the application of Gompertz’s approach. The approach is used to describe and represent accumulative biogas (or bio methane / bio hydrogen) evolution curves (ABE curves) obtained from batch anaerobic co-digestion as frequently used in biochemical methane/hydrogen potential (BMP/BHP)

experiments. The authors proposed four types of ABE curves typically encountered in practice based on the substrate complexity. Type I is shown up when the substrate can be represented by single entity. Type II-VI is found when dealing with complex substrates or multiple substrates. A Gompertz two-substrate model (GTS model) was developed based on a corrected form of Gompertz equations and a preference (or switching) function. The resulting equation is more versatile than the original Gompertz equation and can represent most of typical ABE curves very well. Its parameters are intuitive, easy to interpret and give meaningful description of ABE curves. The authors recommended that, for batch anaerobic co-digestion, GTS model should be used instead of conventional modified Gompertz equation, particularly when ABE curves exhibit non-smooth characteristics due to complex microbial growth and substrate consumption patterns.

Keywords— Gompertz model; batch; anaerobic; co-digestion; ABE curves

INTRODUCTION

The Gompertz function, is a type of mathematical model for a time series and is named after Benjamin

Gompertz. The Gompertz model is well known and widely used in many aspects of biology. Originally, it has been

frequently used to describe the growth of animals and plants, as well as the number or volume of bacteria and cancer

cells [4]. In the context of anaerobic digestion, Gompertz function particularly, its modified form is widely used to

represent the accumulative biogas data [2]. The main reasons for these are that its parameters can be interpreted

easily, and the non-linear regression can be used to fit the model to experimental data in a straightforward manner.

Furthermore, the fitting its normally very satisfactory for single anaerobic digestion. Recently, AD co-digestion has been found as an effective mean to increase the biogas yield and it has appeared

in large amount in the literature. To represent and describe the accumulative biogas data, modified Gompertz model

is used most often. It is also very helpful in determining the biochemical methane and hydrogen potential by

removing subjective judgment regarding the ultimate biogas that could be obtained for specific set of conditions.

However, the accumulative biogas data obtained from AD co-digestion often too complex for Gompertz-type and

Schnute models to represent the data sets satisfactorily. This problem occurs for these models because the single

substrate assumption is shared by all models of this type. For example, in Fig. 1 we show a set of AD co-digestion

data of chicken manure, Thai noodle wastewater and rice husk. If we use Gompertz or Schnute model to fit this set

of data by trying to fit the initial period, the middle period will be off. Conversely, if we try to fit the data in

intermediate region, the initial period will be off as show in Fig. 1.

22

FIGURE 1. Failure of Gompertz and Schnute equations to represent the ABE curve often obtained from co-digestion batch experiments

Classification of Batch Anaerobic (Co-) digestion curves

Model development in this article is based on the accumulative biogas evolution (ABE) curves as shown in Fig.

2. By observing large amount of ABE curves in the literature. ABE curves (bio-methane and bio-hydrogen) can be

groups into four types based on the curve morphology and number of substrates grouping. The background behind

the different types may not be described uniquely. However, here we will describe their differences as attributed by

the distribution of substrates as available to the main groups of microorganisms. Here we propose that there are

more than two classes of substrate having different degree of digesting difficulty and microbial specificity. In the light of substrate digestibility, we describe four hypothetical types of ABE curves as follow.

Type I: Single substrate which is consumed by single group of microorganisms. Typically, these curves can be

represented well with Monod kinetics and Gompertz model. Most biogas data can be represented well with these

models.

Type II: Multiple substrates consumed in parallel or sequentially by one or multiple groups of microorganisms.

Type III: Multiple substrates but they can be simplified by three categories: easily degradable, slowly degradable

and intermediates.

Type IV: Multiple substrates with complex chain of consumption by groups of microorganisms.

FIGURE 2. types of accumulative biogas evolution

Two approaches for describing biogas data are popular among researchers, namely: Gompertz and Monod

approaches. The first approach postulates that the specific growth rate, µ is a function of time whereas the other

relates µ to the substrate concentration and other physical and chemical factors such as pH and the presence of

inhibitors. Monod approach is more fundamental and more flexible and widely used for modeling and simulation of anaerobic digestion process as well as many other fermentation processes. However, Gompertz model, particularly

in the modified form, has become popular for representing biogas data from batch or BMP (biochemical methane

potential) experiments because of its simple interpretation and easy parameter estimation. More importantly, it

generally fit the biogas data very well, presumably most of data fall into Type I and type II.

In this article, the authors attempt to develop a flexible and robust model by extending the conventional

Gompertz model to cover ABE curve of type I, II and III, so that the AD co-digestion data can be described

sufficiently and effectively. The resulting two-substrate model will cover most cases in normal co-digestion

experiments.

MODEL DEVELOPMENT

Jijai and Siripatana used Schnute postulation and rewrote the specific biogas production rate as a function of biomass and biogas generated as follow.

23

( )dP'/ dt P', d /dt=-= + (1)

where P´ = P + is the total accumulative biogas generated , is the amount of biogas produced by active biomass before the experiment starts and P is the observed biogas after the AD process starts. µ is specific growth rate of micro-organisms. α and β are parameters of Schnute kinetics.

Integrating Eq. (1), we obtain

( ) ( )1/ 1/' ' t / t

0P P ' P P e e 1 −

= − = − − −

()

Where γ=βµ0/(α+βµ0), µ0 is the initial specific growth rate and where P∞ is the final (t→∞) biogas accumulated.

In Eq. (1) if we set 0 = , Gompertz equation is obtained. There are three forms of Gompertz equation in use,

namely: original form modified form and corrected modified [9]

( ) ( )( )0P P exp / exp t= − − ()

( )( )( )( )mP P exp exp R e / P t 1 = − − + ()

( ) ( ) ( )( )( )( )( )' ' '

0 0 m 0P P / P P exp exp R e / P P t 1 + + = − + − + ()

In the Gompertz two-substrate model, we postulate that the specific growth rate of microbes consuming (µ) can

be represented by the following time function.

( )e st t

e se g t e− −

= + ()

In equation, there are two terms time functions for the specific biogas production rates for easily (subscribe e) and

slowly (subscribe s) digestible substrates respectively. Here α is the corresponding Gompertz’s parameters. And (t)

is the derivative of switching or preference function which describes how the microorganisms switch from one

substrate to another.

The solution of the model is as follows:

( ) ( )' ' ' ' '

e s e e0 e e s0 sP P P P P g(t) P P P = + = + + − + ()

Where g(t) is a chosen switching/preference function and

' ' ' ' ' '

0 e e e0 s s s0P P P ,P P P ,P P P= + = + = + ()

Here P´, and are the total accumulated biogas that contributed by easily and slowly digestible substrates at

time (t) respectively. and are hypothetical biogas generated from easily and slowly digestible substrates

before starting the batch experiment.

( )( )( )( ) ( )( )( )( )' '

e me e e s ms s sexp exp R e / P t 1 , exp exp R e / P t 1 = − − + = − − + ()

Where θ and Rm are fractional conversions and biomethane/biogas production rate for the corresponding

degradable substrate respectively.

In this article, g(t) was proposed as follows and its graphical representation is depicted in Fig. 3.

( ) ( ) ( )( ) ( )( )1

sg t 1/ tan t / 2−= − + ()

This means

( ) ( ) ( )( ) ( )( ) ( )( )( )( )s2t1 2

s s0 sg t 1/ tan t / 2 e / t 1−−= − + + − + ()

where αs = Rmse/(P∞-Pe∞) and preference gain (κ) which describe how the presence of first substrate was an effect

on the consumption rate of the second one.

As we can see, (t) is more complex than g(t) because it changes with time. However, if κ>1 general the second

term of Eq. (13) is small in comparison to the first term and thus (t) ≈ g(t) in most cases for co-digestion data.

Higher κ is associated with a higher preference for the first substrate while lower κ means less relative preference.

That is in the latter case, the consumption of the first and the second substrate is relative independent. The behavior

of g can be visualize as shown in Fig. 3.

24

FIGURE 3. Graphical representation of g

Seeking a suitable function for g(t) is tricky. Three alternatives are chosen for the following development:

Model I: set g(t) = 1, this is equivalent to independent substrate consumption and biogas production from two

sources. In this case the solution of GTS model is

( ) ( )' ' ' ' '

e s e e0 e e s0 sP P P P P P P P = + = + + − + ()

Model II: The full forms of Eq. (7) and (11) are used for curve fitting.

Model III: It was observed that model II contains both the switching time or τs (plus preference gain, κ) and the

Gompertz’s lag-time λs both parameters have similar function - characterizing the delay response. However, it is

more flexible to use the combination of τs, κ and set λs = 0. We call this “Model III” which is essentially a subset of

model II.

In the following section we will illustrate the power of Gompertz two-substrate models in characterizing ABE

data obtained from AD co-digestion experiments. We will also explore three possibilities in choosing preference

functions and draw some recommendation based on our experiences in dealing with data of these kinds.

MATERIAL AND METHOD

In this work, we show three experiments to illustrate that the developed model can fit many data set of difference

type.

1. AD co-digestion of pig manure and food waste at different ratio [2].

2. Thai rice noodle wastewater co-digested with rice husk or rice husk by pre-treatment with potassium

permanganate (KMnO4) and difference type of manure [16].

3. Palm Oil Mill Effluent (POME) and Rubber Latex Effluent (LTE) co-digested with sludge from palm oil mill [17].

Experimental set-up/design

All digesters in the experiment have a total working-volume of 400 ml. The experiments were conducted at room

temperature (28-30 ºC) until batch completion. The BMP test was conducted using the method of [5]. All

experiments were duplicated/triplicated. Initial pH for all digesters was adjusted to 6.8-7.2 by addition of NaOH.

The digester was flushed with nitrogen gas before sealing. It was sealed with rubber plug and covered with aluminium cap. The biogas production was measured daily by water displacement method as used by other authors

[3], [8] and [13]. The methane content was measured using gas chromatograph (GC-8A Shimadzu). The

experimental setup is shown in Fig. 4.

FIGURE 4. Schematic view of the experimental set-up

25

In the first experiment set, pig manure (M) were co-digested with food waste (W) at different mixing ratios as

shown in Table 1, column 2. In column 3, Thai rice noodle wastewater (TRW 200 mL) and different kinds of

manures (C: cow manure, Ch: chicken manure and Q: quil manure) (10 g of manure added) were co-digested with

rice husk which was pretreated with potassium permanganate (P) or without any pretreatment (NP). And in the third

experiment, we co-digest POME with different ratios of rubber latex effluents (LTE). The mass ratio was shown in Table 1, column 4. In these experiments 160 mL of active AD sludge from a palm-oil-mill AD digester was used.

The digesters used in these experiments were operated in batch mode.

TABLE 1. Experimental design of this study

Digester M: W Ratio Manure: Rice husk ratio POME: LTE

1 0:100 C:P 100:0 2 25:75 Q:P 80:20

3 50:50 Ch:P 60:40

4 75:25 C:NP 40:60

5 100:0 Q:NP 20:80

6 - Ch:NP 0:100

All analytical procedures were performed in accordance with standard methods for the examination of water and

wastewater APHA [14]. The biochemical methane potential was calculated by maximum cumulative methane

divided by gVSadded and maximum cumulative methane divided by gCODremoved [4].

RESULTS AND DISCUSSION

TABLE 2. Characteristic of M were co-digest with W [2].

Digester pH Alkalinity (mg/L

asCaCO3)

VFA (mg/L

asCH3COOH) VFA/ALK

COD

(mg/L)

TKN

(mg/L) before after

1 6.8 6.2 1,345 126 0.094 9,570 543

2 7.2 6.0 1,500 194 0.130 14,940 888

3 7.5 5.9 2,500 182 0.073 20,310 1,233

4 7.7 5.9 2,245 271 0.120 25,680 1,578

5 7.8 5.8 3,129 395 0.130 31,050 1,923

TABLE 3. Characteristics of TRW and different kinds of manures were co-digested with P or NP [16].

Digester pH

Alkalinity

(mg/L

asCaCO3)

VFA

(mg/L

asCH3COOH)

VFA/ALK COD

(mg/L)

VS

(mg/L)

before after before after before after before after

1 7.05±0.07 4.40±0.88 680 1,635 720 4,358 1.06 2.67 39,394 24,933

2 7.05±0.07 5.71±0.10 1,490 2,470 2,595 4,778 1.74 1.93 19,432 12,299

3 7.20±0.28 5.15±0.21 1,670 1,870 1,530 5,010 0.92 2.68 16,138 10,214

4 7.00±0.00 6.16±0.48 1,050 910 1,335 5,663 1.27 6.22 21,032 13,312

5 7.10±0.14 5.49±0.50 780 875 1,020 5,850 1.31 6.69 34,170 21,627

6 7.05±0.07 5.73±0.12 1,600 1,885 2,340 7,083 1.46 3.76 45,856 29,023

TABLE 4. Characteristics of AD co-digestion of sludge from palm oil mill with POME and LTE [17].

Digester Alkalinity (mg/L

asCaCO3) VFA (mg/L asCH3COOH) VFA/ALK

COD

(mg/L) VS (mg/L)

1 5,700 5,400 1.06 19,200 40,200

2 4,300 4,060 1.06 11,600 32,060

3 4,840 4,600 1.05 10,800 35,480

4 6,100 6,000 1.02 14,800 34,420

5 4,500 4,200 1.07 6,400 30,060 6 3,100 3,000 1.03 4,800 22,240

ABE curves for AD co-digestion of M and W

Fig 5 show the biogas accumulation with time for modified Gompertz kinetic models. The results showed that the model fitted the experimental data well. However, the experiments were setup the find the best digestion ratio

26

between pig manure (M) and food waste (W). We chose one set of experimental data from [2]. The experimental conditions and the best-estimated parameters for GTS models for all three variances are tabulated in Table 5. We chose this set of data because it clearly represented four ABE type. For these experimental conditions, less than 1000 mL of methane was obtained for all experiments except for 75M:25W where the final value was about 2600 mL after 25 days of digestion. By visual judgment, ABE curves in the set of experiments fell into type I or II (50M:50W, 100M:0W), type III (0M:100W, 25M:75W) and type IV (75M:25W).

The results fitted very well for 50M:50W and 100M:0W as we can expect for ABE curves of type I and II. It

should be noted that the distinction between type I and II are not clear-cut particularly when the batch performed

poorly (like 100M:0W batch). Type II ABE curves are characterized by only one plateau (or no plateau at all), but

the conventional Gompertz equation cannot fit the data well for the whole regions. If we try to fit the initial and final

regions of the ABE curves, the middle region will be off and vice versa. Referred to Fig. 5(a), we could not tell

whether this batch was of type I or II but these two curves fitted the modified Gompertz well as shown in Table. 5.

For Type-III ABE data (0M:100W, 25M:75W), all variances of GTS models fitted the data very well (the

correlation coefficients, R2 > 0.96). The goodness of fit of GTS model II and III for these data were excellent (R2 ≥

0.98). GTS model II showed no superiority over model III so that λs was not needed or it was redundant with

parameters κ and τs. However, a pair of κ and τs is more flexible and superior over λs while retaining interpretative

simplicity as τs. So, it is preferable for representing the co-digestion data.

(a) (b)

(c) (d)

FIGURE 5. Biogas accumulation for (a) Modified Gompertz model, (b) GTS model I, (c) GTS model II and (d) GTS model III

TABLE 5. Summarized description of the Gompertz models, parameters and the best-fit parameter (R2)

Model Parameter Different pig manure to food waste ratio

0:100 25:75 50:50 75:25 100:0

Modified

Gompertz

equation

P∞ 1,157.35 923.15 817.29 2,637.02 517.93

Rm 133.50 83.03 479.31 311.00 35.91

λ 2.5755 -1.7200 1.0383 4.5693 1.4932

R2 0.9809 0.8915 0.9954 0.9726 0.9896

Gompertz two

substrate

(Model I: g(t)=1)

P∞ 1,221.45 1,022.31 824.80 2,531.27 531.23

Pe∞ 430.55 575.32 720.13 350.23 81.22

Rms 50.51 28.73 17.93 432.27 29.55

Rme 1,831.51 500.79 556.14 265.23 105.97

λe 1.4745 1.0621 1.1384 0.9590 3.2807

λs 5.0611 4.9495 1.3337 6.7385 2.7411

R2 0.9966 0.9911 0.9988 0.9852 0.9952

Gompertz two

substrate

(Model II: λs ≠ 0)

P∞ 1,222.65 1,034.70 846.03 2864.83 508.67

Pe∞ 178.44 549.74 605.11 386.92 230.84

Rms 61.38 404.05 193.49 131.89 41.42

27

Rme 50.46 883.51 579.27 172.18 54.17

κ 5.3161 0.2198 0.1780 3.1444 0.3005

λe 0.2008 1.4123 1.0807 0.7594 2.9331

λs -2.7296 3.5857 1.8236 -2.0171 0.6637

tr 6.14 12.86 2.00 8.2724 12.9250

R2 0.9967 0.9962 0.9956 0.9947 0.9861

Gompertz two

substrate

(Model III: λs = 0)

P∞ 1,205.12 1,055.25 830.25 2,753.60 565.25

Pe∞ 241.98 513.80 571.98 425.07 76.62

Rms 70.19 112.89 114.69 158.51 26.91

Rme 76.84 1,109.97 703.80 128.12 126.37

κ 5.3161 0.1804 0.9307 2.7103 0.3333

λe 1.0050 1.5768 1.3481 0.4850 3.3069

tr 6.19 12.46 2.82 8.54 4.77

R2 0.9966 0.9933 0.9992 0.9950 0.9962

ABE curves for TRW and different kinds of manures were co-digested with P or NP

For the experiment of AD co-digestion of TRW with rice husk and difference type of manure, GTS model III

goodness-of-fit for these data were excellent and the correlation coefficients (R2) were greater than 0.97 similar to

that of as co-digestion of M and W.

FIGURE 6. Biogas accumulation for GTS model

TABLE 6. Summarized description of the Gompertz models, parameters and the best-fit parameter (R2)

Digester Parameter

P∞ Pe∞ Rms Rme κ λe λs s R2

1 842.36 255.4 18.8 188.4 0.15 0.5015 0 21.8 0.9794

2 1144 800 289.8 135.53 0.5 0.4726 0 15.6 0.9853

3 335.95 144 775.7 111.25 0.29 0.0399 0 19.2 0.9695

4 1508.2 1,150 120 101.47 1 1.1011 0 24.5 0.9789

5 861.5 700 154.5 52.09 4.4 -0.2967 0 23.2 0.9935

6 783.5 640.8 120 1041.22 1.2 1.7249 0 14.5 0.9922

28

ABE curves for AD co-digestion of sludge from palm oil mill with POME and LTE

FIGURE 7. Biogas accumulation for GTS model

For the experiment of AD co-digestion of POME and LTE with sludge, when fitted GTS model III to these data,

the goodness-of-fit were excellent similar to the first and second experiments and the correlation coefficients (R2)

were greater than 0.98.

The effect of difference parameter of the GTS model

In Fig. 8 shown that the κ parameter affects the shape of ABE curves. Higher means more distinct preference.

Here Rms is biomethane/biogas production rate for slowly degradable substrate (SD). And s is time-lag for SD

substrate consumption.

(a)

(b)

29

(c)

FIGURE 8. The effects of parameters , Rms and s on shape of model prediction: (a) (b) Rms and (c) s

Figure 9. Flowchart guideline for fit the GTS model

Fitting GTS model to co-digestion data is quite tricky in order to obtain 9 parameters correctly and meaningfully.

So, we should start by keeping some easily-determined parameters (normally the fraction of slowly degradable

substrate, P∞) constant, then allowing least-square fitting software to search for the rest of the best-fit parameters.

Otherwise, although the software could find the best-fit parameters with very high the correlation coefficients (R2)

but the parameters would not be meaningful as expected. In the article, we used qtiplot program for fitting the gompertz two substrate model, of which the flow chart

shows some guideline for the users. When you know some parameter, you can fix these parameters such as P∞ and

Pe∞, generally for LTE and POME, Pe∞ was about 15 – 20% of Ps∞ for this example data.

TABLE 7. Summarized description of the Gompertz two-substrate model, parameters and the best-fit parameter (R2)

POME: LTE Parameter

P∞ Pe∞ Rms Rme κ λe λs tr R2

100:0 968.3 774.64 6.99 79.57 0.2049 -0.2831 0 14.34 0.9876

80:20 926.9 741.52 7.95 113.09 0.5461 -0.1740 0 11.02 0.9918

60:40 854.1 683.28 7.48 128.56 0.3269 -0.1482 0 7.76 0.9907

40:60 793.3 634.64 6.97 136.12 0.2793 -0.1646 0 7.59 0.9872

20:80 781.4 625.12 8.54 178.31 0.3218 -0.1073 0 7.77 0.9869

0:100 582.7 466.16 8.52 159.90 0.9307 0.0138 0 6.98 0.9943

CONCLUSION

Traditional Gompertz equation is not optimal for general use in representing the co-digestion data because of the

presence of substrates having different difficulty in degradation for micro-organisms involved. GTS models

developed in the work provides a better solution for co-digestion problems. The proposed models have three variants

based on switching or preference function. It was found that GTS model III provided very good representation of co-

digestion. It replaces time-lag (λs) in modified Gompertz equation for slowly degradable substrate by

switching/preference function (κ and τs). The model is very flexible and could represent co-digestion ABE curves

very well. All parameters can be interpreted in a simple and direct way. However, all Gompertz-type models

(including GTS models) do not directly provide design parameters and thus they are not very good for extending

their applications to continuous systems and to include other physico-chemical factors such as temperature, pH,

inhibitors etc. If ones need the design parameters and a more flexible approach, Monod-type models are

30

recommended. However, GTS models are excellent for compressing ABE data into the form of simple equation and

small number of parameters. It will be very useful for determination of BMP and SMA as well as a book keeping of

biogas data collection for future uses.

ACKNOWLEDGMENTS

The authors would like to thanks Walailak University for funding this research.

REFERENCES

[1] M. Das, “Comparison of kinetic Models for Biogas Production Rate from Saw Dust,” International Journal of Research in Engineering and

Technology, vol. 03, pp. 248–254, 2014.

[2] S. Jijai and C. Siripatana, “Kinetic Model of Biogas Production from Co-digestion of Thai Rice Noodle Wastewater (Khanomjeen) with

Chicken Manure,” Energy Procedia, vol. 138, pp. 386–392, 2017.

[3] S. Jijai, G. Srisuwan, S. O-Thong, N. Ismail, and C. Siripatana, “Effect of Substrates and Granules/Inocula Sizes on Biochemical Methane

Potential (BMP) and Methane Kinetics,” Iranica Journal of Energy and Environment, vol. 7, pp. 94–101, 2016.

[4] N. Kyurkchiev and A. Iliev, Extension of Gompertz-type Equation in Modern Science: 240 Anniversary of the birth of B. Gompertz. 2018.

[5] W. F. Owen, D. C. Stuckey, J. B. Healy, L. Y. Young, and P. L. McCarty, “Bioassay for monitoring biochemical methane potential and

anaerobic toxicity,” Water Research, vol. 13, no. 6, pp. 485–492, 1979.

[6] N. Paepatung, A. Nopharatana, and W. Songkasiri, “Bio-methane potential of biological solid materials and agricultural wastes,” Asian

Journal on Energy and Environment, vol. 10, pp. 19–27, 2009.

[7] T. Rachadaporn, H. Thongpan, N. Rakmak, and C. Siripatana, “Modeling of anaerobic co-digestion of pig manure and domestic organic

waste,” Jurnal Teknologi, vol. 78, 2016.

[8] A. Raj, S. Vinaykumar, H. Manjunath, A. Srinidhi, and J. Patil, “Biomethanation of Water Hyacinth, Poultry Litter, Cow Manure and Primary Sludge: A Comparative Analysis,” Research Journal of Chemical Sciences, Vol. 1, Issue 7, 2011, pp 22-26, ISSN 2231-606X (IF:

0.3725)., vol. 1, pp. 2231–606, 2011.

[9] J. Schnute, “A Versatile Growth Model with Statistically Stable Parameters,” Canadian Journal of Fisheries and Aquatic Sciences, vol. 38,

pp. 1128–1140, 2011.

[10] U. Sidik, F. Razali, S. Alwi, and F. Maigari, “Biogas Production Through Co-Digestion Of Palm Oil Mill Effluent With Cow Manure,”

Nigerian Journal of Basic and Applied Sciences, vol. 21, 2013.

[11] C. Siripatana, S. Jijai, and P. Kongjan, “Analysis and extension of Gompertz-type and Monod-type equations for estimation of design

parameters from batch anaerobic digestion experiments,” in AIP Conference Proceedings, 2016, vol. 1775, p. 030079.

[12] C. Siripatana, S. Jijai, S. O-Thong, and N. Ismail, “Modeling of Biomethane Production from Agro-Industrial Wastewaters with Constant

Biomass: Analysis of Gompertz Equation,” 2015.

[13] K. M. C. Tjørve and E. Tjørve, “The use of Gompertz models in growth analyses, and new Gompertz-model approach: An addition to the

Unified-Richards family,” PLOS ONE, vol. 12, no. 6, p. e0178691, Jun. 2017.

[14] E. W Rice and A. Public Health Association, Standard methods for the examination of water and wastewater. 2012.

[15] M. Zwietering, I. Jongenburger, F. M. Rombouts, and T. Van, “Modeling of Bacterial Growth Curve,” Applied and environmental

microbiology, vol. 56, pp. 1875–81, 1990.

[16] T. Boonkamnerd, C. Siripatana, L. Noynoo and S. Jijai, “Effect of Rice Husk by Pretreatment with Potassium Permanganate (KMnO4) for

the Potential of Biogas Production,” submitted for publication.

[17] N.Yingthavorn, L. Noynoo, T. Boonkamnerd and C. Siripatana, “Biochemical Methane Potential of Palm Oil Mill Effluent (POME) Co-

Digested with Rubber Latex Effluent (LTE): Effect of POME/LTE Ratio and Temperature,” submitted for publication.

31

The Effects of Hydrophilicity and Aggregation of

Nanoparticles in Alpha Olefin Sulfonate on Foam Stability

Bee Chea Tan1, a), Ahmad Kamal Idris1, 2, b), Nur Asyraf Md Akhir1, 2, c),

Ismail Mohd Saaid1, 2, d) and Nabilla Afzan Abdul Aziz1, 2, e)

1Department of Petroleum Engineering Universiti Teknologi PETRONAS, 32610 Seri Iskandar, Perak, Malaysia.

2Institute of Hydrocarbon Recovery

Universiti Teknologi PETRONAS, 32610 Seri Iskandar, Perak, Malaysia.

a)Corresponding author: [email protected]

b)[email protected] c)[email protected] d)[email protected] e)[email protected]

Abstract. Gas could bypass trapped oil and causing viscous fingering in high permeability zone. Foam has been proposed to minimize gas mobility in enhanced oil recovery technique. Since the surfactant-only foam was reported unstable in challenging reservoir conditions, nanoparticle acts as the additive in surfactant solution to strengthen foam lamellae and enhance foam stability. Hence, the objective of this work is to investigate the effects of hydrophilicity and aggregation of nanoparticles in alpha olefin sulfonate on bulk foam stability. Hydrophilic silicon dioxide, aluminium oxide and titanium dioxide nanoparticles were dispersed in sodium chloride brine and alpha olefin sulfonate surfactant, respectively. The hydrophilicity of nanoparticles in the solutions was determined through contact angle of brine on the

coated glasses. The aggregation of nanoparticles in the solutions was also measured by zeta potential. Foam stability in static condition was determined based on its half-life and justified through microscopic observation on the foam structure and the thickness of foam lamellae inside the glass column. Experimental results revealed that titanium dioxide nanoparticle in alpha olefin sulfonate is the most hydrophilic and aluminium oxide nanoparticle in alpha olefin sulfonate is the most aggregated. Surprisingly, titanium dioxide nanoparticle in alpha olefin sulfonate surfactant yielded the most stable carbon dioxide foam at ambient condition. These findings contributed to the understanding of foam stabilizing mechanisms using ultrafine solid particles by the virtue of nanoparticle’s hydrophilicity and aggregation in alpha olefin sulfonate, an anionic surfactant.

Keywords—hydrophilic nanoparticle; aggregation; foam stability; microscopic; contact angle; zeta potential; quartz

INTRODUCTION

Gas injection is one of the principal enhanced oil recovery (EOR) methods. For a mature oil field, immiscible

gas injection is desirable due to decreased reservoir pressure. However, the viscosity difference between gas and residual oil is remarkable and causing viscous fingering in porous medium. Moreover, accelerated gas mobility at

elevated temperature would bypass trapped oil and breakthrough early at the wellbore, especially in high

permeability zone. To overcome this issue, several approaches were proposed to minimize gas mobility in porous

medium and foam flooding is one of them.

Foam is a two-phase medium consists of gas pockets trapped in liquid lamellae whereby the lamella acts as a

barrier to reduce gas fingering and improve sweep efficiency. Even so, foam stability exhibits adverse effect towards

elevated temperature and salt ions in the brine. The cost implied is a concern when excessive surfactant is required

to generate more stable foam in unfavourable reservoir condition. As a result, foam additives were recommended to

enhance the stability of surfactant-based foam. According to Lai and Dixit [1], an effective foam additive need to

fulfil at least one of the following mechanisms: (1) increase elasticity of foam film, (2) retard liquid drainage in

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

32

foam lamellae, (3) decrease gas diffusion across foam lamellae, (4) increase thickness of electrical double layer, and

(5) increase surface and bulk viscosity of foam film.

Recent research on foam stabilization using ultrafine solid particles has evolved into nano-sized particles that

ranged from 1 nm to 100 nm. The stabilizing effect of solid particles is believed to arise from their aggregation

whereby the particles are closely packed at foam film surfaces to prevent coalescence of gas bubbles [2]. The gas

bubble coalescence process is closely associated to two mechanisms listed above, which are retarding liquid

drainage in foam lamellae and decreasing gas diffusion across foam lamellae. For instance, Tang et al. [3] concluded

that smaller particle sizes (20 – 50 nm) yielded greater stabilizing effect on aqueous foam. They discovered that for hydrophobic particles, the gas diffusion process gets dominant when particle size decreases and at the same time, the

liquid drainage process gets insignificant. This positive result draws the interest of stakeholders to investigate further

on the synergy effect of nanoparticles-surfactant generated foam.

Since the synergy effect between nanoparticles and surfactant needs to be considered, nanoparticles adsorption

on surfactant molecules is a decisive factor in bulk foam stabilization. In the context of petroleum engineering,

adsorption is the attraction and holding of a layer of a chemical on the wall of a formation which usually held by

ionic charge or wetting preference [4]. Active nanoparticles are adsorbed at the foam lamellae between the gas-

liquid interface and stabilize foam due to reduction on liquid-liquid interfacial area. In contrast to polymer

molecules, nanoparticles possess greater adsorption energy at the interface and promote more stable foam [5]. There

are three possible mechanisms of liquid film stabilization by solid particles: (1) monolayer of bridging particles, (2) bilayer of close-packed particles, and (3) network of particle aggregates in the lamella [6].

Many published literature stated that the presence of nanoparticles in surfactant enhances foam stability,

however, not every nanoparticle yields extraordinary outcome [5, 7-9]. The characteristics of nanoparticles to be

considered as an effective foam stabilizing agent is still not conclusive. Many previous studies found that

nanoparticles with intermediate hydrophilic surface (θw ≈ 90°) are the best stabilizing agent, yet a dispersing agent is

required to disperse this type of nanoparticles in water. For this reason, it is incomparable to hydrophilic

nanoparticles of the same kind due to the presence of an auxiliary agent in the nanofluids. Furthermore, some recent

studies also claim that the least aggregated nanoparticle in solution is good for foam stabilization [5, 8, 10]. This,

however, contradicts with the fundamental works by Everett [2] who stated that particle aggregation is one of the

foam lamellae stabilizing mechanisms by solid particles. Therefore, the objective of this study is to investigate the

effects of nanoparticle’s hydrophilicity and aggregation in alpha olefin sulfonate surfactant on bulk foam stability.

METHODOLOGY

Materials

Several grams of sodium chloride (NaCl) from Merck is dissolved in distilled water to make brine with 3 wt%

NaCl concentration. It served as the base fluid for all prepared samples. Alpha olefin sulfonate (AOS, Bio-Terge®

AS-40), an anionic (negatively-charged) surfactant with 39% active C14-16 from Stepan Company is used as the

foaming agent. Three different types of hydrophilic nanoparticles namely silicon dioxide (SiO2), aluminium oxide

(Al2O3) and titanium dioxide (TiO2) were used as foam stabilizer. The specifications of these nanoparticles and their

pH values in different solutions are provided in Table 1. Precleaned microscope glass slides were purchased from

HmbG were used as the contact surface. Baronia crude obtained from PETRONAS was used to treat the glasses to

oil-wet. Carbon dioxide gas (CO2) is used to generate foam in the glass column.

TABLE 1. Specification of Hydrophilic Nanoparticles

Description Silicon Dioxide (SiO2) Aluminium Oxide (Al2O3) Titanium Dioxide (TiO2)

Company US Research

Nanomaterials Sigma-Aldrich Sigma-Aldrich

Average particle sizea (nm) 20 – 30 13 21

Densitya (g/cm³) 2.4 4.0 4.26

Specific surface areaa (m2/g) 180 – 600 85 – 115 35 – 65

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

33

Particle pHa Not available 4.5 – 5.5 3.5 – 4.5

Average pH in brine 5.05 7.58 5.97

Average pH in AOS 6.82 8.02 5.87 a. Information provided by the Company.

Preparation of Foaming Solution and Nanofluids

Above its critical micelle concentration, AOS was diluted to 5000 ppm with brine as foaming solution.

Nanofluids are made by adding certain mass of nanoparticles in brine to make 0.5 wt% particle concentration (i.e.

SiO2, Al2O3 and TiO2). The same mass of nanoparticles was also added in the foaming solution (i.e. SiO2–AOS,

Al2O3–AOS and TiO2–AOS). Then, the AOS foaming solution and the six nanofluids were sonicated in an

ultrasonic bath (Telsonic) for 30 minutes to achieve fluid’s stability.

Coating of Microscope Glass

Microscope glasses were cut to 2 cm × 2 cm dimension. To make an oil-wet surface, these glasses were soaked

in Baronia crude for several days and then dried in an oven (Thermo Scientific™) at 70°C. After that, these crude-

coated glasses were socked in nanofluids for about 10 minutes without agitation, accordingly, then dried at 70°C to

make a nanofluid-coated layer for contact angle experiment.

Experimental Procedures

Zeta Potential and Contact Angle

The aggregation of nanoparticles in fluids was determined via zeta potential (ζ). An analyzer with electroacoustic

sensor (DT 1202, Dispersion Technology) was operated at ambient condition to determine the electrokinetic

potential in the colloidal dispersions. Particles aggregation occurs due to the attractive or repulsive electrostatic

force between the particles in a dispersion. The aggregation of particles is detected when the zeta potential value is

close to zero or at zero.

Hydrophilicity is the tendency of a particle or a molecule to be wetted or surrounded by water. The

hydrophilicity of nanoparticles is determined via contact angle of brine (θw) using interfacial tension meter with

integrated camera (IFT-700, Vinci Technologies) which operated at ambient condition. Sessile drop method (i.e.

brine surrounded by air) is preferred to measure the contact angle on the coated glasses.

Bulk Foam Study

This study was carried out using a foam analyser, FoamScan® (Teclis). It is equipped with a 35-mm inner

diameter glass tube with four electrodes attached along the tube at equal interval. 50 ml of AOS solution was first

injected into the tube and then the foam is generated by sparging CO2 gas at 100 ml/min, flowing through a 160-

micron pore size glass frit. The foam generation stopped when it reached 150 ml. Then, the foam started to decay

with time. Foamability is the time taken for the foaming solution to reach target foam volume. Meanwhile, bulk

foam stability is commonly determined by its half-life due to measurement reliability [11]. It is defined as the time

taken for the foam to reach half of its original foam volume. Microscopic images of foam were also captured by an

integrated camera throughout the experiment. Then, the foam is discharged when the experiment ends. The

procedures were repeated by replacing the AOS solution with SiO2–AOS, Al2O3–AOS and TiO2–AOS nanofluids,

respectively.

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

34

RESULTS AND DISCUSSION

Aggregation of Nanoparticles in NaCl Brine with and without AOS Surfactant

Fig. 1 presents the aggregation of SiO2, Al2O3 and TiO2 nanoparticles in NaCl brine with and without AOS

surfactant. Based on the results, all nanofluids are dominantly positively-charged because the zeta potential values are

all positive. The SiO2 is the most stably dispersed nanoparticles in both brine and AOS because it yields the highest

zeta potential value, followed by TiO2 and Al2O3 nanoparticles. The average zeta potential values for SiO2, Al2O3 and

TiO2 nanoparticles in brine are 61.02 mV, 44.66 mV and 46.66 mV, respectively. Meanwhile, the average zeta

potential values for SiO2, Al2O3 and TiO2 nanoparticles in AOS are 36.35 mV, 17.42 mV and 28.93 mV, accordingly.

Since the less hydrophilic Al2O3 nanoparticle is not likely to be surrounded by brine, or in other words, to be

dispersed in water (referring to Fig. 2). Consequently, Al2O3 nanoparticle tends to clump together and form bigger

aggregates in the solutions which led to lower zeta potential values.

(a) (b)

FIGURE 1. The zeta potential (ζ) and pH values of SiO2, Al2O3 and TiO2 nanoparticles in (a) NaCl brine and (b) AOS surfactant at ambient condition.

When the concentration of surfactant is above its critical micelle concentration (CMC), the hydrophobic tail of

surfactant molecules sequestrates in the centre while its hydrophilic head contacts with the surrounding water and

form micelles. And, when the positively-charged SiO2, Al2O3 or TiO2 nanoparticles added in the AOS surfactant, the hydrophilic and negatively-charged surfactant head will adsorb on to the nanoparticle’s surface thus create bigger

aggregates in brine. This finding is consistent with previous studies who have examined that the presence of

surfactant in nanofluid yielded lower zeta potential value due to floc formation in surfactant solution [12] and the

adsorption of surfactant on nanoparticle’s surface [13]. On top of that, it is also discovered that the zeta potential

decreased with increasing pH of nanofluids as shown in Fig. 1. High pH condition may favour particle aggregation in

the solutions as the charge density and pH of solutions could affect zeta potential value [14, 15].

Hydrophilicity of Different Nanoparticles in NaCl Brine with and without AOS Surfactant

Table 2 provides the contact angle of brine measured on the SiO2, Al2O3 and TiO2 coated glasses with and

without AOS surfactant. The initial contact angle of brine on the oil-wet glass is 109° at ambient condition. It is

found that TiO2 nanoparticle is the most hydrophilic, followed by SiO2 nanoparticle, and Al2O3 nanoparticle is the

least hydrophilic in both brine and AOS surfactant. The contact angle measured on the SiO2 nanoparticle is 41°,

Al2O3 nanoparticle yields 48°, and TiO2 nanoparticle yields 30°. On the other hand, the contact angle measured on

the SiO2–AOS coated glass is 31°, Al2O3–AOS coated glass yields 45°, and TiO2–AOS coated glass yields 27°. Fig.

2 illustrates the water drop images captured on oil-wet and nanofluid-coated glasses.

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

35

TABLE 2. The contact angle of brine (θw) on SiO2, Al2O3 and TiO2 nanoparticles coated glasses at ambient condition.

Coating on Oil-Wet Glass θw Coating on Oil-Wet Glass θw

SiO2 and NaCl 41° SiO2 and AOS 31° Al2O3 and NaCl 48° Al2O3 and AOS 45°

TiO2 and NaCl 30° TiO2 and AOS 27°

(a) (b)

FIGURE 2. Representative images of water drop on (a) the oil-wet glass and (b) the TiO2–AOS coated glass at ambient condition after 30 minutes.

In the case of nanofluid without AOS, the water contact angle on top of the nanoparticles layer simply signifies

the particle’s hydrophilicity. The positively-charged SiO2 and Al2O3 nanoparticles settling on top of the negatively-

charged crude layer that has already coated on microscope glass. On a different note, the presence of AOS in

respective nanofluids further reduces the water contact angle which indicates a more hydrophilic surface.

Since the coated surface is more hydrophilic when AOS is present in the nanofluids, then charge neutralization is

explanatory. As expected, nanoparticles and AOS were both coated on the crude layer on the glass but competitive

adsorption is speculated in this case. Herein, a possible explanation might be the positively-charged SiO2, Al2O3 or

TiO2 nanoparticles were first to adsorb on the crude layer, then only the AOS surfactant micelles with negatively-

charged heads adsorbed on the nanoparticle’s surface and formed a hydrophilic surface at the most top of the glass. This suggests that the combination of anionic AOS surfactant and positively-charged SiO2, Al2O3 and TiO2

nanoparticles holds stronger hydrophilic property than nanoparticles alone. In accordance with the present result,

Kuang et al. [16] have noted that nanofluids added in anionic surfactant have made the contact surface slightly more

hydrophilic.

Bulk Foam Study

Effect of Different Nanoparticles in Foamability

As presented in Fig. 3, adding SiO2, Al2O3 or TiO2 nanoparticles in AOS surfactant shows insignificance on the

foamability of CO2 foam. The foamability of AOS, SiO2–AOS, Al2O3–AOS and TiO2–AOS foam was 112 seconds,

114 seconds, 112 seconds and 101 seconds, respectively. This finding confirms that surfactant is the dominant

component in the foamability of CO2 foam regardless of the presence of nanoparticles in the AOS surfactant. This

further supports the idea of earlier studies that the presence of nanoparticles in AOS solution would not promote the

foamability of CO2 foam at ambient condition [10, 17]. It may even deteriorate the foamability of CO2 foam due to surfactant adsorption on nanoparticles [18].

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

36

Effect of Different Nanoparticles in Foam Stability

The presence of SiO2, Al2O3 and TiO2 nanoparticles in AOS surfactant enhances bulk foam stability as shown in

Fig. 3. Contrary to expectations, the TiO2 nanoparticle yields the greatest foam stabilizing effect than prevalent SiO2 nanoparticle. The foam half-life for of AOS, SiO2–AOS, Al2O3–AOS and TiO2–AOS was 321 seconds, 361 seconds,

451 seconds and 488 seconds, respectively. Unfortunately, the results show inconclusive finding in terms of

nanoparticle’s aggregation and hydrophilicity. It is noted from the Fig. 3 that the most stable foam has the shortest

foamability time. This finding does not support the previous study by Manan et al. [8] who found that Al2O3

nanoparticle in AOS surfactant could stabilize CO2 foam better than SiO2 nanoparticle, followed by TiO2

nanoparticle, and lastly copper (II) oxide nanoparticle due to their zeta potential values.

FIGURE 3. The foamability and half-life of AOS surfactant, SiO2–AOS, Al2O3–AOS and TiO2–AOS at ambient condition.

Microscopic Observation on Foams

The presence of SiO2 and Al2O3 nanoparticles in foam does not change the shape of bubbles during foam generation and at its half-life but a slightly different case for TiO2 nanoparticle. As provided in Fig. 4, the presence

of SiO2 and Al2O3 nanoparticles in AOS surfactant generated smaller and rounded bubbles. In contrast, the presence

of TiO2 nanoparticle in AOS surfactant created angular bubble shape at time t = 110 s. In fact, rounded and smaller

bubbles were indeed created at the beginning of foamability in the presence of TiO2 nanoparticle. Therefore, it is

suspected that TiO2 nanoparticle drains quickly in the lamellae once the foamability stop due to its particle density.

It can be seen from Fig. 4 that the presence of SiO2 and Al2O3 nanoparticles in AOS surfactant slows down the

liquid drainage rate in CO2 foam by observing the thickness of lamellae throughout the experiment. For instance, at

time t = 150 s, the shape of AOS foam is more angular which indicates a higher rate of liquid drainage. Meanwhile

at the same timestep, SiO2–AOS foam is somewhat in between angular and rounded in shape which means moderate

liquid drainage rate. Next, Al2O3–AOS foam is more rounded and thicker which suggests slower liquid drainage at time t = 150 s. On the contrary, TiO2–AOS foam yielded angular and thinner bubble which speculates that the TiO2

nanoparticle is not effective in delaying liquid drainage rate.

By comparing their respective bubble shapes at time t = 110 s, the bubble shape of Al2O3–AOS foam is fairly

retained which implies that it has the slowest rate of liquid drainage among all and lead to more stable foam. Except

for TiO2–AOS, this microscopic result is in accordance with most of the previous research which demonstrated that

the presence of nanoparticles in foam could slow down liquid drainage rate and bubble coalescence, thus enhancing

CO2 foam stability by increased foam half-life at dry foam regime [5, 7, 9, 10, 18, 19]. However, the result for

TiO2–AOS has not previously been described. Since TiO2 nanoparticles might have drained out of the lamellae, it is

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

37

assumed that the foam (without nanoparticles) is in its steady state when the internal pressure between the bubbles

are equal whereby the gas diffusion and liquid drainage processes are minimal at this stage and led to better foam

stability at dry foam regime.

(a) (b) (c) (d)

FIGURE 4. From the left column to the right are the microscopic foam images of (a) AOS, (b) SiO2–AOS, (c) Al2O3–AOS and (d) TiO2–AOS, respectively. The rate of liquid drainage is determined through the thickness of foam lamellae observed at

ambient condition.

CONCLUSION

The effects of nanoparticle’s hydrophilicity and aggregation in alpha olefin sulfonate surfactant on bulk foam stability were examined in this study. The SiO2 nanoparticle is the most stably dispersed in NaCl brine and AOS

surfactant, followed by TiO2 and Al2O3 nanoparticles. The increase in nanofluid’s pH also favours particle

aggregation in solutions. On the other hand, TiO2 nanoparticle is the most hydrophilic, followed by the SiO2 and

Al2O3 nanoparticles. The presence of SiO2, Al2O3 and TiO2 nanoparticles in AOS surfactant does not enhance

foamability of CO2 foam. It was surprisingly found that the TiO2 nanoparticle yielded the most stable foam,

followed by Al2O3 and SiO2 nanoparticles. The bubble shape of Al2O3–AOS foam was fairly retained which

indicates an effective control on the delay of liquid drainage rate in foam lamellae. Unfortunately, the findings on

the nanoparticle as foam stabilizer are still not conclusive. Therefore, the underlying reasons need to be further

investigated to characterize an effective foam stabilizer by using nanoparticles.

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

38

ACKNOWLEDGEMENTS

This work is funded by Fundamental Research Grant Scheme (FRGS) from the Ministry of Education (MOE)

Malaysia, and Shell–TUD–UTP Collaboration. The authors also thank for the technical support provided by Centre

of Enhanced Oil Recovery (COREOR) in Universiti Teknologi PETRONAS (UTP).

REFERENCES

[1] K. Y. Lai and N. Dixit, “Additives for foams,” in Foams: Theory, Measurement, and Applications, 1st ed. vol.

57, R. K. Prud'homme and S. A. Khan, Ed. New York: Marcel Dekker, 1996, pp. 315–338.

[2] D. H. Everett, Basic Principles of Colloid Science. London: The Royal Society of Chemistry, 1988, p. 260.

[3] F. Q. Tang, Z. Xiao, J. A. Tang, and L. Jiang, “The effect of SiO2 particles upon stabilization of foam,” J.

Colloid Interface Sci., vol. 131, no. 2, pp. 498–502, 1989.

[4] Petrowiki.org, “Glossary: Adsorption,” 2013. [Online]. Available: https://petrowiki.org/Glossary:Adsorption.

[Accessed: 17–Aug–2018].

[5] A. S. Emrani and H. A. Nasr-El-Din, “An experimental study of nanoparticle-polymer-stabilized CO2 foam,”

Colloids Surf. A, vol. 524, pp. 17–27, 2017.

[6] T. S. Horozov, “Foams and foam films stabilised by solid particles,” Curr. Opin. Colloid Interface Sci., vol. 13,

no. 3, pp. 134–140, 2008.

[7] A. S. Emrani and H. A. Nasr-El-Din, “Stabilizing CO2 foam by use of nanoparticles,” SPE Journal, vol. 22, no.

2, pp. 494–504, 2017.

[8] M. A. Manan, S. Farad, A. Piroozian, and M. J. A. Esmail, “Effects of nanoparticle types on carbon dioxide

foam flooding in enhanced oil recovery,” Petroleum Science and Technology, vol. 33, no. 12, pp. 1286–1294,

2015.

[9] F. Guo and S. Aryana, “An experimental investigation of nanoparticle-stabilized CO2 foam used in enhanced oil

recovery,” Fuel, vol. 186, pp. 430–442, 2016.

[10] F. Guo, J. He, P. A. Johnson, and S. A. Aryana, “Stabilization of CO2 foam using by-product fly ash and recyclable iron oxide nanoparticles to improve carbon utilization in EOR processes,” Sustainable Energy Fuels,

vol. 1, no. 4, pp. 814–822, 2017.

[11] E. Iglesias, J. Anderez, A. Forgiarini, and J.-L. Salager, “A new method to estimate the stability of short-life

foams,” Colloids Surf. A, vol. 98, pp. 167–174, 1995.

[12] Z. A. AlYousef, M. A. Almobarky, and D. S. Schechter, “The effect of nanoparticle aggregation on surfactant

foam stability,” J. Colloid Interface Sci., vol. 511, pp. 365–373, 2018.

[13] Y. Wu, S. Fang, K. Zhang, M. Zhao, B. Jiao, and C. Dai, “Stability mechanism of nitrogen foam in porous

media with silica nanoparticles modified by cationic surfactants,” Langmuir, vol. 34, no. 27, pp. 8015–8023,

2018.

[14] P. McElfresh, M. Wood, and D. Ector, “Stabilizing nano particle dispersions in high salinity, high temperature

downhole environments,” presented at the SPE International Oilfield Nanotechnology Conference, Noordwijk,

The Netherlands, Jun. 12–14, 2012, Paper SPE-154758-MS.

[15] N. A. Ogolo, “The trapping capacity of nanofluids on migrating fines in sand,” presented at the SPE Annual

Technical Conference and Exhibition, New Orleans, Louisiana, USA, Sep. 30–Oct. 2, 2013, Paper SPE-

167632-STU.

[16] W. Kuang, S. Saraji, and M. Piri, “A systematic experimental investigation on the synergistic effects of aqueous

nanofluids on interfacial properties and their implications for enhanced oil recovery,” Fuel, vol. 220, pp. 849–

870, 2018.

[17] B. C. Tan, N. A. M. Akhir, and A. K. Idris, “Investigation on the effect of types of nanoparticles and

temperature on nanoparticles-foam stability,” in Proceedings of the International Conference on Industrial

Engineering and Operations Management, vol. 2018-March, 2018, pp. 365–372.

Proceedings of Applied Mathematics and Applied Science in Engineering International Conference 2018, pp. 31-39,

July 2019

39

[18] N. Yekeen, A. K. Idris, M. A. Manan, A. M. Samin, A. R. Risal, and T. X. Kun, “Bulk and bubble-scale

experimental studies of influence of nanoparticles on foam stability,” Chinese Journal of Chemical

Engineering, vol. 25, no. 3, pp. 347–357, 2017.

[19] N. Yekeen, M. A. Manan, A. K. Idris, A. M. Samin, and A. R. Risal, “Experimental investigation of

minimization in surfactant adsorption and improvement in surfactant-foam stability in presence of silicon

dioxide and aluminum oxide nanoparticles,” Journal of Petroleum Science and Engineering, vol. 159, pp. 115–

134, 2017.

40

Sequence Alignment on Hadoop Computer Cluster

Ade Jamal1, a), Endang Ripmiatin1, b) and Yunus Effendi2, c)

1Informatics Department, Faculty of Science and Technology, University Al-Azhar Indonesia, Jl.Sisingamangaraja,

Jakarta 12110, Indonesia 2Biology Department, Faculty of Science and Technology, University Al-Azhar Indonesia, Jl.Sisingamangaraja,

Jakarta 12110, Indonesia

a)Corresponding author: [email protected]

b)[email protected] c)[email protected]

Abstract. A center of research for bioinformatics has been initiated at our university where faculty members from informatics department and biology department can collaborate doing research. To empower the center, we have done some work in development biological sequence database on top of a distributed file system using Hadoop framework. Due to the next generation sequencing assembly machine, publicly available DNA sequence database such as GenBank governed by the National Center for Biotechnology Information is huge and still grows exponentially. Hence the need for processing computer that can handle big data to perform computation in the bioinformatics field become more and more critical for our bioinformatics research center. The goal is to asses a scalable computer cluster that is affordable in investment cost but still can be expanded as necessary. For this purpose, the Hadoop framework is chosen which is

known as the first publicly available big data platform. This paper will present the similarity searching in the DNA sequence database using sequence alignment in parallel model based on MapReduce computation model from the Hadoop framework. Using only limited computer power resources, some speed up by increasing the number of computers node is proven.

Keywords—bioinformatics; DNA Sequence Alignment; Hadoop Framework; Distributed File System; Map-Reduce

INTRODUCTION

In 2011, we took part in a consortium for research in the vaccine of hepatitis B lead by state-owned medicine

company. This joint research which accompanied by a number of researchers from the prominent research institution in bioscience and bio-molecular in Indonesia is aimed to develop the second generation hepatitis vaccine.

Lesson learned from the joint research is that Indonesia has no central database for molecular biology which collects

and disseminates biological information from bioscience researcher in the country. This fact has been already

National interest because Indonesia as a big country and big nation potentially has huge data of molecular biology

since we have one of the largest biodiversity natural laboratories in our tropical forest and under the sea, in the

biggest palm plantation, the variety of tropical diseases and still more to name it.

After more than seven years, Indonesian bio-molecular researchers still have no nationwide molecular biology

database for collecting and disseminating their work. Undeniable, the researchers will publish their result of

bioscience research in the international database such as provided by provided by National Center for Biotechnology

Information (NCBI) at the National Institutes of Health (NIH) in the USA. One of the reasons is an excessive

computational time required to process the very large data, in this case the huge DNA sequences data on an affordable computer system.

Biological Sequence Database

To meet the necessity of managing and to distribute a rapidly increasing body of knowledge in bio-molecular,

Congers of USA established the National Center for Biotechnology Information (NCBI) [1]. GenBank (the Genetic

Sequence Data Bank) project at NCBI, provides the scientific community with a computer database of DNA and

RNA sequences, is funded from National Institutes of Health(NIH), Department of Defense (DOD), Department of

Energy (DOE) and other US institutions through a contract with the DOE acting on behalf of Los Alamos National Laboratory (LANL) [2].

41

The nowadays primary biological sequences databases (i.e. protein, DNA and RNA sequences) are Genbank

provided by a collaboration of three organizations under International Nucleotide Sequence Database Collaboration

(INSDC). These three institutions are the NCBI at the National Institutes of Health (NIH, USA); the EMBL

Nucleotide Sequence Database provided by the European Bioinformatics Institute (EBI) and the DNA Databank of

Japan (DDBJ) in which these three institutions share and exchange sequence information [3][4]. The collaboration was started in 1982, just two years after the EMBL Data Library was established in Germany. Four years after the

collaboration was established, EMBL disseminated the last printed distribution consisted of 8,823 entries

representing about 8.5 million bases. Ten years after the last printed distribution, in September 1996, EMBL

released reports of 931,582 sequence entries comprising about 609 million nucleotides bases, which still can be

disseminated using CD-ROM [5]. This is about 72 times growth in the number of nucleotide bases. Ten years later

the NCBI released in 2006 more than 50 billion bases, which is more than 80 times growth within ten years.

The size of GenBank data had been doubling about every 1.5 years, and this trend of growth rate remains due to

the enormous growth in data from expressed sequence tags (ESTs) [6]. This exponential growth continues as shown

in Fig. 1 [7]-[19].

FIGURE 1. Exponential Growth of Nucleotide Bases in Genbank

The publicly available Genbank data are stored into multiple files; for release 221 in 2018, there are 2932 files

requiring 841 GB of uncompressed disk storage [19]. Although in term size of files of Genbank is not as big as

others big data, the combination of computation work in bioinformatics fields and the amount of data and its

potentially exponentially growth, application for processing biological sequences database requires advanced

computational technology.

Time efficiency of sequence searching

From its inception of Genbank, sequence database searching has always been an important issue. The scoring

search function of sequence database is sequence similarity which is well known as sequence alignment scoring.

There are three types of sequence alignment algorithms [20], namely:

1. Exhaustive search algorithm which enumerates all possible solution to find the best solution. Thus, the

exhaustive search algorithms are the most effective in term find the best solution but may be very slow in

term of time efficiency. Many traditional sequence alignment methods that using dynamic programming is

based on this exhaustive search method such as the global sequence alignment algorithm of Needleman-Wunsch, and the local sequence alignment algorithm of Smith-Waterman.

2. Heuristic search algorithm uses a heuristic function to reduce the search space. The heuristic algorithm

works well in many cases, although there is no guarantee of the resulting quality. In practice, the heuristic

algorithm gives acceptable results very quickly and as such many practical application for database

searching based on this, i.e., FastA and BLAST.

3. Filter based algorithm which applies a filter to select candidate position in the database where the query

sequence possibly occurs with a high-level similarity.

42

The sequence search program, known as BLAST, was developed at NCBI and is enhanced continually to enable

researchers to compare a new, unknown sequence with more than 100,000 sequences in seconds and to have the

results returned instantly over the network [1]. Study on exhaustive search was also done by many such as back to

1986 [21] who optimized a search of a 500-residue protein sequence against the entire PIR database Ver. 1.0 (1)

(500.000 residues) for vector processing on a Hitachi S810-20 supercomputer. It was carried out in a CPU time of 45 sec. It required 4 min for an exhaustive search of a 1500-base nucleotide sequence against all mammalian

sequences (1.2M bases) in Genbank Ver. 29.0. The superiority of heuristic search algorithm over exhaustive search

in term of time efficiency is apparently such that the FastA and BLAST is de facto biology sequence search engine.

COMPUTER CLUSTER FOR BIOLOGICAL SEQUENCE DATA

The database work mentioned above was done on a Supercomputer [21] or a network of high-end Server such as

reported for Genbank in 1992 using a network of Sun Microsystem 4/690 server [2]. Although many application

software is free or even open-source software, the required hardware is often costly. In this traditional high-performance computing (HPC) applications, it is a common practice to have "high-end processing nodes" with a

large amount of shared memory and "storages nodes" attached by a high-capacity interconnection device. This

scaling "up" approach is not cost effective, since the cost of such machine does not scale linearly. It becomes worst,

since the hardware technology obsoletes very quickly that in four or five years one has to replace with a new more

expensive hardware.

The issue of high-cost investment in computer hardware causes an important obstacle for low budget Indonesian

laboratory. Some state-owned company can afford this investment but very often it has been reported as fail in

return on investment. Hence, the affordable alternative than invest in highly priced computers is preferable.

A study on this has been initiated by uploading the DNA sequence data on Hadoop Distributed File System on a

low-end cluster. A so-called MapReduce computation model is invoked for keyword searching algorithm in

conjunction with Hadoop Distributed File System as both technologies are the main component of Hadoop framework [22]. Recently, we extended the searching algorithm by considering global sequence alignment.

Hadoop Framework

The separation of computing node and storage node creates a bottleneck in the network. As an alternative to

moving around the data, it is more efficient to move processing around to the data; hence the processors and the

storage are co-located. In this type of computing model, one can take the benefit of data locality by running code on

the computing processor directly attached to the block of they need. The architecture of computer cluster which

complies to this model is called a distributed file system. Over this type of clusters a so-called MapReduce programming model, as proposed by Dean and Ghemawat

from Google in [23] is usually invoked. Hadoop itself was developed by Doug Cutting for his Nutch project[24]; i.e.

full-featured text indexing and searching library. Using the Hadoop framework Nutch became more scalable than

any web crawler engine at that time [25].

Hadoop framework consists of two main components, namely Hadoop Distributed File System (HDFS) and

MapReduce distributed computation programming framework. Using these two Hadoop components, we can

decompose a very large data set and do computations in parallel across many commodity servers [26][27]. HDFS is a file system that is designed for run a MapReduce application on a cluster of commodity computers. A

big data set will be divided into smaller (say 64Mbyte) blocks/chunks that are spread among computers nodes in the

cluster via HDFS. These chunks of data input will be read in parallel which provide a much higher throughput. For

data-intensive processing, the number of chunks will be too large to be moved around between nodes in the cluster. Instead of moving the data, Hadoop lets program codes move around. This move-code-to-data concept is more

efficient with respect to communication load because the program codes are orders of magnitude smaller than the

data chunk.

MapReduce is a distributed programming model which is inspired from functional programming model.

MapReduce proceeds large datasets normally in two stages, i.e. map and reduce stage. In the first stage, the map

function is applied over all input records in the large datasets and can be performed in parallel since each functional

application happens independently. The intermediate result of the first stage if required could be read in aggregation

way by the reduce (folding) function. Application programmer just needs to define these two main functions and the

MapReduce framework will execute the actual processing which decomposes the job into a set of map tasks,

shuffle-sort and a set of reduce tasks.

43

In contrary to the more common relational tables, MapReduce uses its basic data structure in form of key-value

pairs (k, v). This form of data structure provides the flexibility to tackle semi-structured or even unstructured data

sets. The mapper is operated on every input key-value pair (k1, v1) spread over a number of files (or blocks) to

produce a list of intermediate key-value pairs, namely a list of [(k2, v2)].

map: (k1,v1) → list[(k2,v2)] ()

The reducer is applied to all values corresponding to the same intermediate key (k2, [v2]) to generate a list of output key-values [(k3,v3)].

reduce: (k2, list[v2]) → list[(k3,v3)] ()

Restructure DNA Sequence Genbank Database for Hadoop

NCBI provides GenBank database available for public via ftp in two formats, namely the GenBank Flat File

format available at NCBI’s anonymous FTP server ftp://ftp.ncbi.nih.gov/genbank and ASN.1 format

available at ftp://ftp.ncbi.nih.gov/ncbi-asn1.

GenBank database groups sequence records into various divisions based either on the source taxonomy or the sequencing strategy on which the data is obtained. Some of taxonomic divisions are presented in Table 1. The

number of sequence data files shown increases for each new release.

TABLE 1. GenBank Taxonomic division

Division code Description

Bct Bacteria Inv Invertebrate

Mam Other mammals Pln Plant (inc. Fungi and algae) Pri Primate Phg Phage Rod Rodent Vrl Viruses

Vrt Other vertebrate

44

.FIGURE 2. Sample of Genbank flat i.e. file gbbct1.seq from bacterial division

The name of flat files of Genbank describes itself the content. The first two letters “gb” indicate database name,

i.e. Genbank, the second three letter is the division name given in Table. 1, and the number indicate file number as

counter of amount files in related division. Extension “.seq” means the file stores sequence data. All Genbank flat

file has the same format and consists of two main parts; i.e. header information and sequence entries. Header

information includes the name file, release number, released datum, division description, and number of sequence

entries. The second portion contains sequence entries where separator token “//” put between two successive entries.

Within sequence entry, each line consists of two part, i.e. the first ten (10) columns in line may contain:

• Keyword; if it begins in column 1. Example: REFERENCE, LOCUS, ORIGIN etc.

• Sub-keyword; if the first two columns in line are blank. Example AUTHORS as sub-keyword of

REFERENCE.

• Blank character indicating that this line a continuation of information under keyword or sub-keyword above

it.

• Number ending in column 9 of the line designates the numbering of the actual nucleotide sequence position.

• Two slashes (//) in column 1 and 2 indicating the end of entry.

The second part in position 13 to 80 contains the information associated to its keyword or sub-keyword. These

first and second parts of sequence entry form a key-value pairs (k,v) used in MapReduce data structure. For

instance, the DNA sequence data is obtained after some string processing as the value v of ORIGIN acts as the key.

Sequence Alignment in Hadoop Cluster

This article presents result of exhaustive search based on similarity obtained from sequence alignment. In this

work, the traditional dynamic programming exhaustive technique, namely Needleman-Wunsch method for global

sequence alignment is utilized. The global alignment is chosen above local alignment because the subsequent work,

namely building phylogenetic tree will need the global alignment. Global sequence alignment is usually performed

between two sequences. Hence it is known as pairwise sequence alignment process. In this case, one input sample of

sequence will be aligned against a whole sequence records in the same division database, for instance in the bacterial

division as shown in Fig. 3.

GBSMP.SEQ Genetic Sequence Data Bank

October 15 1992

GenBank Flat File Release 74.0

Structural RNA Sequences

2 loci, 236 bases, from 2 reported sequences

LOCUS AAURRA 118 bp ss-rRNA RNA 16-JUN-1986

DEFINITION A.auricula-judae (mushroom) 5S ribosomal RNA.

ACCESSION K03160

VERSION K03160.1 GI:173593

KEYWORDS 5S ribosomal RNA; ribosomal RNA.

SOURCE A.auricula-judae (mushroom) ribosomal RNA.

ORGANISM Auricularia auricula-judae

Eukaryota; Fungi; Eumycota; Basidiomycotina;

Heterobasidiomycetidae; Auriculariales; Auriculariaceae.

REFERENCE 1 (bases 1 to 118)

AUTHORS Huysmans,E., Dams,E., Vandenberghe,A. and De Wachter,R.

... deleted ....

BASE COUNT 27 a 34 c 34 g 23 t

ORIGIN 5' end of mature rRNA.

1 atccacggcc ataggactct gaaagcactg ca...deleted...

61 gtaccgccca gttagtacca cggtggggga cc...deleted...

//

LOCUS ABCRRAA 118 bp ss-rRNA RNA 15-SEP-1990

DEFINITION Acetobacter sp. (strain MB 58) 5S ribosomal

ACCESSION M34766

VERSION M34766.1 GI:173603

KEYWORDS 5S ribosomal RNA.

SOURCE Acetobacter sp. (strain MB 58) rRNA.

ORGANISM Acetobacter sp.

Prokaryotae; Graci

... deleted ....

45

FIGURE 3. Multiple pairwise sequence alignment in parallel

We apply multiple pairwise Needleman-Wunch global alignment for similarity searching in the bacterial

division. For this purpose the same database used in the previous work for keyword searching [22] is utilized,

namely DNA data from Genbank. The test was performed using commodity computer with the following specifications: AMD FX-8350 8 Core processor, 16 GB RAM for master server and AMD Phenom II X4 quad-core

processor, 16 GB RAM for five slave node computers. Because of limited computer power resources, for model

testing, only fraction of the data from bacterial division was included by taking only 196 bacterial files. After

reading input files and mapping phase, there 1,307,686 records are found from these 196 files.

The DNA sequence as input sample Escherichia coli DNA for mannosyl transferase, phosphoribosyl-ATP

pyrophosphohydrolase with AB000176 as accession number and Locus. This a short sequence with 241 residues.

Fig. 4 shows the input sample sequence in Fasta format. Before multiple alignment process is started, sequence

records read from Hadoop distributed file systems are filtered against the residue length of the sequence. Only

sequences with residue length no more than 100 residue length differs from input sample, in this test case, only

sequence which has length 141 up to 341 are taken into account in the alignment process.

FIGURE 4. Input sample sequence

Similarity searching of this input sample returned two results with 100% similarity, namely the two identic

sequence data stored in two records with different Locus, namely AB000176 and AB00180 as shown in Fig. 5.

Search test is repeated with a slightly different input sample by mutating the first ten residues, which yields the same

two records with the highest similarity, i.e. 97.52%

46

FIGURE 5. Output similarity searching using sequence alignments

The elapsed computing time for this searching on modest power Hadoop computer cluster is 975 seconds using

computer 5 nodes. Reducing number of computer nodes needed longer elapsed time, namely 1226 and 1676 second,

using 4 nodes and 3 nodes respectively.

47

CONCLUDING REMARKS AND RECOMMENDATION

The result has shown that using Hadoop Framework technology which is inspired by Google search engine

technology it is very promising to handle big data in this case similarity searching of DNA and also protein

sequences using affordable computer system. However, due to limitation of available hardware in term of

specification and number of computer nodes in Hadoop cluster, the wall-clock time consumed for exhaustive

similarity search algorithm is still can be improved a lot.

Further research is recommended in objective to speed up exhaustive search using Hadoop cluster by extend the

node number and optimize the alignment process by invoking fine parallel computation using General Purpose

Graphic Processing Unit (GP-GPU).

ACKNOWLEDGMENTS

The authors would like to thank Direktorat Jendral Pendidikan Tinggi, Kemendikbud, for funding the presented

work through research grant under PTUPT program and LP2M UAI for International Seminar Grant. We are also

grateful to our partner Solusi247 which allow us to run this work on their computer facilities. Special gratitude we

would address to Mr. Solechul Arifin for his support in using HGrid247 library for this work, and to Ms Rusnah

who programmed the multiple sequence alignment module used in this present work.

REFERENCES

[1] Rose Marie Woodsmall and D. A. Benson, ”Information resources at the National Center for Biotechnology

Information” in Proc. Annual Conference of the Special Libraries Association, San Francisco, California, 1992.

[2] C. Burks, M. J.Cinkosky, W. M. Fischer, P. Gilna*, J. E.-D.Hayden, G. M.Keen, et al., “GenBank,” Nucleic

Acids Research, 1992, vol. 20, Supplement 2065-2069

[3] E.W.Sayers and T. Barrett, ,”Database Resources of The National Center for Biotechnology Information,”

Nucleic Acids Research, January: vol.37, Database issue, pp. D5-D15, 2009.

[4] E.W.Sayers (NCBI Resource Coordinators), “Database resources of the National Center for Biotechnology

Information,” Nucleic Acids Research, January: vol. 42, Database issue D7–D17, 2014.

[5] Guenter Stoesser*, Peter Sterk, Mary Ann Tuli, Peter J. Stoehr and Graham N. Cameron “The EMBL

Nucleotide Sequence Database,” Nucleic Acids Research, vol. 25, pp. 7–13, 1997.

[6] D. A. Benson*, M. S. Boguski, D. J. Lipman, J. Ostell and B. F. Francis Ouellette, “GenBank,”, Nucleic Acids

Research, 1998, vol. 26, 1–7

[7] D. A. Benson, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, D. L. Wheeler, ”GenBank”, Nucleic Acids Research,

vol. 34, Database issue, D16-D20, 2006.

[8] D. A. Benson, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, D. L. Wheeler, ”GenBank”, Nucleic Acids Research,

vol. 35, Database issue, D21-D25, 2007.

[9] D. A. Benson, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, D. L. Wheeler, ”GenBank”, Nucleic Acids Research,

vol. 36, Database issue, D25-D30, 2008.

[10] D. A. Benson, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids Research,

vol. 37, Database issue, D26-D31, 2009.

[11] D. A. Benson, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids Research,

vol. 38, Database issue, D46-D51, 2010.

[12] D. A. Benson, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids Research,

vol. 39, Database issue, D46-D51, 2011

[13] D. A. Benson, I. Karsch-Mizrachi, K. Clark, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids

Research, vol. 40, Database issue, D48-D53, 2012.

[14] D. A. Benson, M. Cavanaugh, K. Clark, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”,

Nucleic Acids Research, vol. 41, Database issue, D36-D42, 2013.

[15] D. A. Benson, K. Clark, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids

Research, vol. 42, Database issue, D32-D37, 2014.

48

[16] D. A. Benson, K. Clark, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids

Research, vol. 43, Database issue, D30-D35, 2015.

[17] K. Clark, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers, ”GenBank”, Nucleic Acids Research, vol.

44, Database issue, D67-D72, 2016.

[18] D. A. Benson, M. Cavanaugh, K. Clark, I. Karsch-Mizrachi, D. J. Lipman, J. Ostell, E. W. Sayers ”GenBank”,

Nucleic Acids Research, vol. 45, Database issue, D37-D42, 2017.

[19] D. A. Benson, M. Cavanaugh, K. Clark, I. Karsch-Mizrachi, J. Ostell, K. D. Prultt, E. W. Sayers ”GenBank”,

Nucleic Acids Research, vol. 46, Database issue, D41-D47, 2018.

[20] W. K. Sung, 2010, Algorithms in Bioinformatics: a practical Introduction Chapman & Hall/CRC mathematical

and computational biology series, London UK.

[21] O. Gotoh and Y. Tagashira, “Sequence search on a supercomputer,” Nucleic Acids Research, vol. 14, no. 1,

1986.

[22] A. Jamal, , W. Pradani, N. Hasanati, A. Supriyanto and R. Pujianto, “Scalability of DNA sequence database on

low-end cluster using Hadoop,” in Proc. International Conference on Information Technology Systems and

Innovation (ICITSI), 2014, paper A1-10, pp. 50-55.

[23] J. Dean and S. Ghemawat, "MapReduce: Simplified Data Processing on Large Cluster", in Proceeding of the

6th Symposium on Operating System Design and Implementation (OSDI 2004), pp.137-150, California, 2004.

[24] M. Cafarella, D. Cutting, “Building Nutch: Open Source Search,” ACM Queue, April 2004, vol.2 (2), pp. 54-61,

2004.

[25] R. Khare, D. Cutting, K. Sitaker and A. Rifkin, “Nutch: A Flexible and Scalable Open-Source Web Search

Engine,” Technical Report, CN-TR-04-04, CommerceNet Labs, 2004.

[26] K. Shvachko, H. Kuang, S. Radia and R. Chansler, “The Hadoop Distributed File System,” in Proc. of the 26th

IEEE Symposium on Massive Storage Systems and Technologies (MSST 2010), May, 2010.

[27] T. White, Hadoop: The Definitive Guide, 3rd Edition, USA, O’Reilly, 2012.

49

Effects of Hydrophilic Silicon Dioxide Nanoparticles and

Brine Concentration on Alpha Olefin Sulfonate-Foam Static

Stability

Nurnajla Adnan,1, a) Dzeti Farhah Mohshim2, b) and Ahmad Kamal Idris3, c)

1, 2, 3Department of Petroleum Engineering, Institute of Hydrocarbon Recovery, Universiti Teknologi PETRONAS

Bandar Seri Iskandar, Perak, Malaysia

a)Corresponding author: [email protected]

b)[email protected] c)[email protected]

Abstract. Surfactant works as foam-stabilizer and reduces foam coalescence rate; however, the generated surfactant-stabilized foam is short-lived due to surfactant degradation in high salinity and temperature condition as well as surfactant loss due to adsorption on rock solid surfaces. Addition of commonly used nanoparticles such as silicon dioxide (SiO₂) as foam-stabilizers has the potential in improving anionic foam static stability. This study explores the effects of varying SiO₂ nanoparticles concentration and sodium chloride presence in influencing the anionic surfactant, Alpha Olefin Sulfonate (AOS) foamability and foam static stability. The initial stage of the study involves the investigation of AOS adsorption on silica mineral to ensure its capability as potential foaming agent by having negligible surfactant adsorption on rock minerals. AOS adsorption on silica surface was determined by using ultraviolet-visible (UV-Vis)

spectrophotometry. AOS foam static stability was analyzed by measuring the foam height and comparing the results with SiO₂ nanoparticles presence. The results have shown minor AOS adsorption on silica mineral, which strengthens the AOS potential as foaming agent in this study. SiO₂ addition has no significant effect on foamability but enhances AOS-foam static stability at certain SiO₂ concentration. At this specific SiO₂ concentration, SiO₂ particles accumulate at the foam lamellae and enhances AOS-foam stability. Increasing the sodium chloride concentration up to 2 wt% has shown significant improvement on the AOS-SiO₂ foam static stability. There is synergistic effect between sodium chloride ions available in the solution to slow down the liquid foam drainage rate and accumulation of SiO₂ nanoparticles at foam gas-liquid interface in creating stronger foam lamellae, subsequently enhances the foam static stability. The findings from this

research delivered an initial insight on how nanoparticles concentrations and sodium chloride concentrations may influence the AOS-foam stability.

Keywords— Alpha Olefin Sulfonate, silicon dioxide nanoparticles, surfactant adsorption, foam static stability, enhanced oil recovery

INTRODUCTION

In immiscible gas flooding, foams are introduced to limit carbon dioxide (CO₂) gas mobility issues by offering

an acceptable mobility ratio between the displacing fluid (gas) and the displaced fluid (oil or water), reducing

viscous fingering and gas segregation problems during CO₂ flooding. Besides, foam diverts the fluid flow from

high-permeability region to low-permeability region to enhance the macroscopic sweep efficiency [1]–[3].

Surfactants especially anionic surfactants are conventionally used as foaming agent in CO₂-foam flooding

application due to their ability to improve foam stability by adsorbing onto foam gas-liquid interfaces (lamellae)

consequently contributing to reduction in foam coalescence rate [4]–[7]. However, foams generated by surfactant

are short-lived and eventually collapsed as the foam lamellae becomes thinner due to reversible adsorption of

surfactant molecules from the foam gas-liquid interfaces and foam degradation due to harsh reservoir condition such

as high salinity and high temperature [8]–[11].

50

In recent years, nanoparticles are used as foam-stabilizer in foam experimental studies to enhance the foam

stability generated by surfactants. Foams are stabilized by nanoparticles through particle arrangement at foam

lamellae by forming monolayer bridging particles, a bilayer of close-packed particles or a network of particle

aggregates within foam film [12]–[14]. Besides, nanoparticles adhesion energy at gas-liquid interface is much higher

than surfactant, hence generating longer-lasting foams [15].

Based on previous nanoparticles-stabilized foam stability studies, silicon dioxide (SiO₂) nanoparticles at certain

concentration has notable effect on the foam static stability. Previous study conducted for varying nanoparticles

concentration (0.05 wt% - 5 wt%) claimed that optimum nanoparticles concentration is 1 wt% in presence of 0.3

wt% anionic surfactant, sodium dodecyl sulfonate (SDS). The same study claimed that beyond 1 wt% nanoparticle

concentration, foam stability decreases due to excessive nanoparticle agglomeration [16]. TABLE 1 summarizes

other similar studies and the results of foam static stability (foam half-life) at optimum nanoparticles concentration.

By referring to TABLE 1, most studies centered on silicon dioxide (SiO₂) nanoparticles due to its relatively low

cost compared to other nanoparticles types. Nanoparticles dispersion in surfactant solution enhances foam stability

better compared to surfactant alone. Despite numerous experimental studies conducted, the optimum SiO₂

concentration (0.05 wt% - 1 wt%) to stabilize foams are not yet conclusive, which might be due to effect of surfactant type and different brine concentrations used by different experiments.

The main objective of this study is to investigate whether the selected anionic surfactant properties is a suitable

foaming agent and analyze the influence of SiO₂ nanoparticles concentration and sodium chloride concentration on

the AOS-foam static stability. The results of foam stability by addition of SiO₂ nanoparticles as foam-stabilizer will

be compared to the foam stability results of AOS-only foams. The result of static foam stability act as screening

process to gain early insight on nanoparticles-stabilized foam stability under influence of nanoparticles

concentrations and different sodium chloride concentrations.

TABLE 1. Summary of Silicon Dioxide Nanoparticles-Stabilized Foam Based on Previous Studies

Nanoparticles Surfactant /

Surface -

Modification

Parameters Tested and Condition Results

Source Nanoparticles

Concentration

(wt%)

Salinity

(wt%)

Optimum

Nanoparticles

Concentration

(wt%)

Foam

Half-Life

(hours)

SiO₂ (Nyacol

Chemicals)

EO Alcohol 0.25, 0.5, 0.75,

1 0

0.75 *30 [14]

SiO₂ (US Research Nanomaterials)

SiO₂ (Walker

Chemicals)

SDS, Dichloro-

dimethylsilane

0.05, 0.1, 0.5, 1, 1.5, 2, 5

0.5 wt% NaCl

1

*1.7 [17]

SiO₂ (Wacker-

Chemie)

AOS, SDS,

Betaine

0.05 1 wt% NaCl 0.05 *1.4 [18]

SiO₂ AOS 0.1, 0.3, 0.5, 1 2 wt% NaCl 0.1 *0.8 [19]

SiO₂ (Nyacol DP

9711)

PEG, AOS 0.1, 0.3, 0.5 1 wt% NaCl 0.5 *100 [20]

*Foam half-life (foam static stability) at optimum silicon dioxide nanoparticles concentration

METHODOLOGY

Materials

An anionic surfactant, Alpha Olefin Sulfonate (AOS) was used as the foaming agent and acquired from Stepan

Company with 39% purity. AOS was used at fixed concentration of 0.5 wt% for all foam stability experiments

which is above its critical micelle concentration (CMC) [15], [21].

Hydrophilic silica nanoparticles (SiO₂) were used as foam-stabilizing agents which were obtained from

PlasmaChem GmbH with 99.8% purity. The particle size is in the range of 7-14 nm with specific surface area of 200

51

m²/g. Different SiO₂ concentrations (0.005-1.2 wt%) were used in the study to determine the effects of different SiO₂

concentration on nanoparticles-stabilized foam static stability.

Sodium chloride (NaCl) was acquired from Merck & Co., Germany and used at varying concentrations (1, 2, 4, 8

and 10 wt%) to examine the effects of different sodium chloride concentration on nanoparticles-stabilized foam static stability. The stock solutions of 5 wt% AOS, 5 wt% SiO₂ and 12 wt% NaCl were prepared with DI water. All

properties measurements were carried out at 25⁰C.

Silica mineral was retrieved from Bukit Batu Putih, Gopeng, Perak and used to represent sandstone reservoir

rock without significant clay content. By referring to TABLE 2, the silica mineral is made up of 90% silicon

elements obtained via x-ray fluorescence (XRF) technique with specific surface area of 2.73 m²/g based on multi-

point BET result.

TABLE 2. Silica mineral chemical composition based on x-ray fluorescence (XRF) results

Rock Mineral

(Adsorbent) Specific

Surface Area

Mineral Composition

Component Amount

(wt%)

Silica 2.73

Silicon 90.00

Phosphorus 3.31 Calcium 2.59

Aluminum 1.95

Potassium 1.33

Iron 0.64

Titanium 0.13

Copper 0.03

Zinc 0.03

Rubidium 0.02

Experimental Methods

Alpha Olefin Sulfonate Adsorption on Silica Minerals

The surfactant adsorption on silica minerals was determined by contacting a fixed volume of AOS solution with

5 grams of silica. The surfactant solution was mixed with the minerals and rotated in magnetic stirrer (3000 rpm) for

24 hours at 25⁰C. The mixture was filtered using filter paper and centrifuged at 3000 rpm for 30 minutes to separate

the adsorbent from AOS solution. The filtrate was used to analyze its equilibrium concentration by using ultraviolet-

visible (UV-Vis) spectrophotometry. Surfactant adsorption was measured based on the difference between anionic

surfactant concentration in the bulk before and after in contact with the rock sample. Equation (1) was used for

computing the surfactant adsorption density.

Surfactant adsorption density = [(Co-Cе)V]/g (1)

By referring to (1), Co is the initial concentration of surfactant in ppm, Cе is the equilibrium concentration of

surfactant in ppm, V is the volume of solution in contact with the silica mineral in liter and g is the mass of the silica mineral in gram.

Foamability and Foam Static Stability

Fixed AOS concentration (0.5 wt%) was prepared with different brine concentrations (1 wt%, 2 wt%, 4 wt%, 8

wt% and 10 wt%) to conduct foamability and foam static stability experiments without the presence of SiO₂. The

SiO₂-AOS dispersion was prepared by dispersing a certain mass of SiO₂ at different concentrations (0.005 wt%, 0.01

wt%, 0.05 wt%, 0.1 wt%, 0.2 wt%, 0.3 wt%, 0.4 wt%, 0.5 wt%, 0.6 wt%, 0.7 wt%, 0.8 wt%, 0.9 wt%, 1.0 wt%, 1.1

wt%, 1.2 wt%) and AOS (0.5 wt%) into different NaCl concentrations (1 wt% and 2 wt%). 10 ml total solution of

SiO₂-AOS dispersion was prepared in 30 ml glass vials and stirred for 24 hours and followed by one hour of

sonication to obtain well-dispersed nanoparticles. The foam was generated when SiO₂-AOS dispersion came into

52

contact with air by supplying mechanical energy when the glass vial was shaken for 10 minutes [15], [20]. The

foamability was analyzed based on the foam height whereas the foam static stability was based on the foam half-life

(t½). Foam half-life is the time taken to reach half of the foam original height after foam generation which is

complemented by results of normalized foam height as stated in (2).

Normalized foam height = h₂/h₁ ()

By referring to (2), h₂ is the foam height at specific time and h₁ is the foam height at initial time. The foam height was measured based on the distance between the liquid surface and top of the foam. This procedure was repeated

with different SiO₂ concentrations and sodium chloride concentration to examine their effects on foamability and

nanoparticles-stabilized foam static stability.

RESULTS AND DISCUSSION

Anionic surfactant, Alpha Olefin Sulfonate (AOS) is one of potential anionic surfactant to be used for CO₂-foam

flooding application. Experimental activities have been conducted on AOS surfactant to measure and analyze its adsorption on silica mineral, foamability and foam static stability with and without SiO₂ nanoparticles presence. The

experimental results are analyzed and discussed below.

Alpha Olefin Sulfonate Adsorption on Silica Minerals

To analyze the AOS adsorption on silica mineral by using ultraviolet-visible (UV-Vis) spectrophotometry, a

standard calibration curve was plotted according to Beer’s law to analyze the relationship between different

surfactant initial concentration with its corresponding absorption value. The calibration curve and AOS adsorption

result are shown in Fig. 1 and Fig.2 respectively.

FIGURE 1. Alpha Olefin Sulfonate (AOS) standard calibration curve

FIGURE 2. Alpha Olefin Sulfonate (AOS) adsorption on silica minerals in 2 wt% sodium chloride (NaCl) solution

Based on Fig. 1, AOS absorption increases as AOS initial concentration. The calibration curve R² value is

0.9991. By obtaining absorption of different solution from UV-Vis, the AOS equilibrium concentration can be

calculated. The AOS adsorption on silica is shown in Fig. 2. Region I and II are easily identified due to drastic AOS

53

adsorption on silica as the AOS concentration increases. In region I, the lower AOS concentration is adsorbed into

the silica surface because of the electrostatic attraction between surfactant head group with rock surface charges

whereas there is drastic increase in surfactant adsorption density in region II due to interaction between hydrocarbon

chain and surface monomers which resulted in surface colloids, hemi-micelles and admicelles formation. Region III

is identified as the AOS adsorption increases but with lower gradient. Region IV is the plateau region to indicate the maximum limit of AOS adsorption on silica in 2 wt% NaCl. Hence, the maximum adsorption of AOS on silica

mineral in 2 wt% NaCl is 1.57 mg/g. The results are in line with previous surfactant adsorption studies [10], [22].

These AOS adsorption results indicates that AOS has minor surfactant adsorption, subsequently avoiding surfactant

loss due to adsorption on solid surface, hence promising an effective CO₂-foam flooding application.

Foamability and Foam Static Stability of Alpha Olefin Sulfonate without Silicon Dioxide

Nanoparticles

Alpha Olefin Sulfonate (AOS) foamability and foam static stability results without SiO₂ presence as foam-

stabilizers are shown in Fig. 3 and 4. Based on results in Fig. 3, AOS foamability, which is the ability to generate

foam decreases as sodium chloride concentration increases. By referring to Fig. 4, as sodium chloride concentration

increases, foam takes longer time to reduce to half of its initial height. The stability of the foam is improved as NaCl

concentration is increased from 1 wt% to 10 wt%. This result is similar to result reported by [16], [23].

Although AOS foamability is better in 1 wt% NaCl, however the foam half-life is the shortest (16 hours),

whereas the foam half-life is the longest in 10 wt% NaCl (18 hours). The foam has the highest static stability at 10 wt% NaCl, hence at this specific value the concentration is called transition salt concentration [24]. Salts like

sodium chloride are believed to stabilize bubble coalescence by changing the hydrodynamic boundary condition

from mobile to immobile at the transition salt concentration. Due to the change in hydrodynamic boundary

condition, liquid drainage in foam film becomes slower hence the foam stability improves. Another possible

mechanism of foam stability improvement due to sodium chloride presence is Gibbs-Marangoni effect [25], [26].

This effect occurs when salt ions are non-uniformly distributed at the interface between two bubbles, which creates a

tangential stress to prevent liquid film drainage and creates an immobile interface. Hence, the immobile interface

will inhibit the bubble coalescence and enhance foam stability.

FIGURE 3. Foamability of Alpha Olefin Sulfonate without presence of silicon dioxide nanoparticles in different sodium chloride (NaCl) concentrations: (a) 1 wt% NaCl (b) 2 wt% NaCl (c) 4 wt% NaCl, (d) 8 wt% NaCl and (e) 10 wt% NaCl

FIGURE 4. Relative foam height (foam static stability) of Alpha Olefin Sulfonate without presence of silicon dioxide nanoparticles in different sodium chloride (NaCl) concentrations

(a) (b) (d) (c) (e)

54

Foamability and Foam Static Stability of Alpha Olefin Sulfonate with Addition of Silicon

Dioxide Nanoparticles

Effects of Silicon Dioxide Nanoparticles Concentration

Foamability and foam half-life tests for AOS solution with SiO₂ presence were conducted and examined by

using fixed AOS concentration with 0.005, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1 and 1.2 wt%

SiO₂ in 1 wt% NaCl and the results are shown in Fig. 5, 6 and 7. By observing AOS foamability results in Fig. 5 and

analyzing foam relative height in Fig. 7, the addition of SiO₂ nanoparticles as foam-stabilizer has no significant effect on foamability but a notable effect on foam static stability. The longest foam half-life is 20 hours at 0.7 wt%

SiO₂, which is the optimum SiO₂ concentration for generating stable foam in 1 wt% NaCl. It is suggested that

addition of SiO₂ as foam-stabilizer at certain concentration, which is at moderate concentration may create stable

foam by slowing down the drainage of thin aqueous film due to SiO₂ particle adsorption and accumulation at foam

gas-liquid interface and creates thicker foam lamellae [12], [13], [20], [27]. However, at relatively low SiO₂

concentration (0.005 to 0.3 wt%), foam static stability is not improved by SiO₂ addition as the foam half-lives are

shorter compared to AOS-only foam. But beyond 0.3 wt% SiO₂, the foam static stability has significant increase up

to 0.7 wt% SiO₂. The shortest foam half-life is 6 hours at 1.2 wt% SiO₂, where the foam is less stable compared to

AOS-foam only. The decrease in foam static stability as SiO₂ concentration increases beyond 0.7 wt% might be due

to particle excessive nanoparticles agglomeration at the foam interface. At high SiO₂ concentration, SiO₂ clumped

together to form bigger sized nanoparticles which will cause gas-liquid foam coarsening or better known as Ostwald

ripening. This process occurs when larger foam bubbles consume adjacent smaller foam bubbles due to pressure difference caused by Young-Laplace effect [28], [29]. Hence, it is not recommended to use high concentrations of

SiO₂ to enhance foam static stability.

FIGURE 5. Foamability of Alpha Olefin Sulfonate at different silicon dioxide nanoparticles concentrations in 1 wt% sodium chloride

FIGURE 6. Foamability of Alpha Olefin Sulfonate at different silicon dioxide nanoparticles concentrations in 2 wt% sodium chloride

55

FIGURE 7. Relative foam height (foam static stability) of Alpha Olefin Sulfonate with presence of silicon dioxide nanoparticles in 1 wt% sodium chloride

Effects of Sodium Chloride Concentration

Similar experiment was conducted to examine the effects of different sodium chloride concentration on AOS-

SiO₂ foamability and foam half-life. The result of AOS foamability is illustrated in Fig. 6 and the foam relative

height in 2 wt% NaCl is depicted in Fig. 8. The result has shown that the AOS-SiO₂ foam in 2 wt% NaCl is the most

stable at 0.6 wt% SiO₂ with longest foam half-life for 26 hours. By comparing 0.2 wt% SiO₂ in 2 wt% NaCl with 0.2

wt% SiO₂ in 1 wt% NaCl, the AOS-SiO₂ foam has longer half-life in higher NaCl concentration. This shows that

increase in NaCl concentration can enhance foam static stability even at low concentration of SiO₂ nanoparticles. By

increasing SiO₂ concentration, foam static stability is enhanced up to certain SiO₂ concentration before the foam

stability decreases. AOS-SiO₂ foam half-lives in 2 wt% NaCl are still longer compared to AOS-SiO₂ foam half-lives in 1 wt% NaCl. This is because as sodium chloride concentration increases, more sodium chloride ions are available

in the solution to distribute non-uniformly at the interface between two bubbles to slow the liquid foam drainage rate

by creating an immobile interface [25], [26]. The immobile interface creates slower foam bubbles coalescence;

hence the foam becomes more stable in higher sodium chloride concentration.

FIGURE 8. Relative foam height (foam static stability) of Alpha Olefin Sulfonate with presence of silicon dioxide nanoparticles in 2 wt% sodium chloride

CONCLUSIONS

This study founded that the addition of SiO₂ nanoparticles as foam-stabilizer up to certain concentration can

improve the AOS- SiO₂ foam static stability. The study has shown that increase in sodium chloride concentration up

to 2 wt% helps in enhancing AOS-foam static stability even at lower SiO₂ concentration. This is because at higher

sodium chloride concentration, the AOS-SiO₂ foam is more stable compared to AOS-only foam due to synergistic

process between salt ions in slowing down the liquid foam drainage rate and accumulation of SiO₂ nanoparticles at foam gas-liquid interface to create thicker and stronger foam lamellae.

56

ACKNOWLEDGMENTS

The authors would like to thank Institute of Hydrocarbon Recovery, Universiti Teknologi PETRONAS for

providing the necessary laboratory equipment and supporting this research through Yayasan UTP Fundamental

Research Grant (YUTP-FRG).

REFERENCES

[1] S. M. Hosseini-Nasab and P. L. J. Zitha, “Investigation of chemical-foam design as a novel approach toward

immiscible foam flooding for enhanced oil recovery,” Energy and Fuels, vol. 31, no. 10, pp. 10525–10534, 2017.

[2] M. Blunt, F. J. Fayers, and F. M. Orr Jr., “Carbon dioxide in enhanced oil recovery,” Energy Convers. Manag.,

vol. 34, no. 9, pp. 1197–1204, 1993.

[3] A. Srivastava, W. Qiao, Y. Wu, X. Li, L. Bao, and C. Liu, “Effects of silica nanoparticles and polymers on

foam stability with sodium dodecylbenzene sulfonate in water–liquid paraffin oil emulsions at high

temperatures,” J. Mol. Liq., vol. 241, pp. 1069–1078, 2017.

[4] D. Green and P. G. Willhite, Enhanced Oil Recovery. 1998.

[5] S. H. Talebian, R. Masoudi, I. M. Tan, and P. L. J. Zitha, “Foam assisted CO2-EOR: A review of concept,

challenges and future prospects,” J. Pet. Sci. Eng., vol. 120, pp. 202–215, 2014.

[6] R. F. Li, W. Yan, S. Liu, G. Hirasaki, and C. A. Miller, “Foam mobility control for surfactant enhanced oil

recovery,” SPE J., vol. 15, no. 04, pp. 928–942, 2010.

[7] Farajzadeh, R., R. Krastev, and P. L. J. Zitha. "Foam films stabilized with alpha olefin sulfonate (AOS)," Colloids and Surfaces A: Physicochemical and Engineering Aspects 324.1-3 (2008): 35-40.

[8] Lake, Enhanced Oil Recovery. 1989.

[9] Hanssen, J. E., Torleif Holt, and L. M. Surguchev, “Foam processes: an assessment of their potential in North

Sea reservoirs based on a critical evaluation of current field experience.” SPE/DOE Improved Oil Recovery

Symposium. Society of Petroleum Engineers, 1994.

[10] M. R. Azam, I. M. Tan, L. Ismail, M. Mushtaq, M. Nadeem, and M. Sagir, “Static adsorption of anionic

surfactant onto crushed Berea sandstone,” J. Pet. Explor. Prod. Technol., vol. 3, no. 3, pp. 195–201, 2013.

[11] M. F. Hamza, C. M. Sinnathambi, Z. M. A. Merican, H. Soleimani, and S. K. D, “An overview of the present

stability and performance of EOR-foam,” Sains Malaysiana, vol. 46, no. 9, pp. 1641–1650, 2017.

[12] T. S. Horozov, “Foams and foam films stabilised by solid particles,” Curr. Opin. Colloid Interface Sci., vol.

13, no. 3, pp. 134–140, 2008. [13] H. Farhadi, S. Riahi, S. Ayatollahi, and H. Ahmadi, “Experimental study of nanoparticle-surfactant-stabilized

CO2foam: Stability and mobility control,” Chem. Eng. Res. Des., vol. 111, pp. 449–460, 2016.

[14] Z. A. AlYousef, M. A. Almobarky, and D. S. Schechter, “The effect of nanoparticle aggregation on surfactant

foam stability,” J. Colloid Interface Sci., vol. 511, pp. 365–373, 2018.

[15] A. S. Emrani and H. A. Nasr-El-Din, “Stabilizing CO2-foam by use of nanoparticles,” SPE J., vol. 22, no. 02,

pp. 494–504, 2017.

[16] N. Yekeen, M. A. Manan, A. K. Idris, A. M. Samin, and A. R. Risal, “Experimental investigation of

minimization in surfactant adsorption and improvement in surfactant-foam stability in presence of silicon

dioxide and aluminum oxide nanoparticles,” J. Pet. Sci. Eng., vol. 159, no. September, pp. 115–134, 2017.

[17] N. Yekeen, A. K. Idris, M. A. Manan, A. M. Samin, A. R. Risal, and T. X. Kun, “Bulk and bubble-scale

experimental studies of influence of nanoparticles on foam stability,” Chinese J. Chem. Eng., vol. 25, no. 3, pp. 347–357, 2017.

[18] F. Guo and S. Aryana, “An experimental investigation of nanoparticle-stabilized CO2-foam used in enhanced

oil recovery,” Fuel, vol. 186, pp. 430–442, 2016.

[19] M. A. Manan, S. Farad, A. Piroozian, and M. J. A. Esmail, “Effects of nanoparticle types on carbon dioxide

foam flooding in enhanced oil recovery,” Pet. Sci. Technol., vol. 33, no. 12, pp. 1286–1294, 2015.

[20] Singh, Robin, and Kishore K. Mohanty. "Synergistic stabilization of foams by a mixture of nanoparticles and

surfactants," SPE improved oil recovery symposium. Society of Petroleum Engineers, 2014.

[21] R. Singh and K. K. Mohanty, “Foam flow in a layered, heterogeneous porous medium: A visualization study,”

Fuel, vol. 197, pp. 58–69, 2017.

57

[22] H. Shamsijazeyi, G. J. Hirasaki, and R. Verduzco, “Sacrificial agent for reducing adsorption of anionic

surfactants,” SPE International Symposium on Oilfield Chemistry. Society of Petroleum Engineers, 2013. pp.

1–16, 2013.

[23] J. San, S. Wang, J. Yu, N. Liu, and R. Lee, “Nanoparticle-stabilized carbon dioxide foam used in enhanced oil

recovery: effect of different ions and temperatures,” SPE J., vol. 22, no. 05, pp. 1416–1423, 2017. [24] M. Firouzi, T. Howes, and A. V. Nguyen, “A quantitative review of the transition salt concentration for

inhibiting bubble coalescence,” Adv. Colloid Interface Sci., vol. 222, pp. 305–318, 2015.

[25] G. Marrucci and L. Nicodemo, “Coalescence of gas bubbles in aqueous solutions of inorganic electrolytes,”

Chem. Eng. Sci., vol. 22, no. 9, pp. 1257–1265, 1967.

[26] V. V. Yaminsky, S. Ohnishi, E. A. Vogler, and R. G. Horn, “Stability of aqueous films between bubbles. Part

1. the effect of speed on bubble coalescence in purified water and simple electrolyte solutions,” Langmuir, vol.

26, no. 11, pp. 8061–8074, 2010.

[27] P. Nguyen, H. Fadaei, and D. Sinton, “Pore-Scale assessment of nanoparticle-atabilized CO2 foam for

enhanced oil recovery,” Energy & Fuels, vol. 28, no. 10, pp. 6221–6227, 2014.

[28] P. Stevenson, “Current opinion in colloid & interface science inter-bubble gas diffusion in liquid foam,” Curr.

Opin. Colloid Interface Sci., vol. 15, no. 5, pp. 374–381, 2010.

[29] Gandolfo, François G., and Henri L. Rosano. "Interbubble gas diffusion and the stability of foams," Journal of colloid and interface science, pp. 31-36, 1997.

58

Optimization of the Enzymatic Saccharification

Process Condition of the Enzymatic Pretreated

Sawdust

Shah Samiur Rashid1, a)

, Amani M. Hashabra1, Essam A. Makky

1, Dr. Aizi

Nor Mazila Binti Ramli1, Jalal K C A

2, Shaheen M. Sarkar

3

1Faculty of Industrial Sciences & Technology, University Malaysia Pahang, Gambang, Pahang, Malaysia 2Kulliyyah of Science, International Islamic University Malaysia, Kuantan, Malaysia

3Bernal Institute, Department of Chemical Sciences, University of Limerick, Castletroy, Limerick, Ireland

a)Corresponding author: [email protected]

Abstract. The saccharification of laccase-pretreated sawdust was optimized using one-factor-at-a-time (OFAT) and response surface methodology (RSM). OFAT was used to investigate four (4) important parameters and it showed that the enzyme pretreated sawdust was best saccharified at the following conditions: cellulase concentration of 30 IU/g of sawdust, substrate concentration of 5.0% w/v, 50 °C, saccharification time of 36 h, and pH 5 where higher yield of sugar was obtained. Based on the OFAT result and previous studies three parameters such as saccharification period, pH and temperature were further optimized statistically using FCCCD (face centered central composite design) of the RSM. The ANOVA (Analysis of Variance) results from the RSM study explained significant probability of interaction of the studied parameters on the saccharification conditions (p<0.05). The model F-value (160.56) and p-value (<0.0001) implies significance of the studied RSM model. The

developed model from RSM study was further validated. Therefore, the optimal saccharification condition was obtained using 5% of the laccase-pretreated sawdust, cellulase enzyme concentration of 30 IU/g, pH 5 at 50°C which yielded a maximum reducing sugar of 4.5 mg/ml after 36 h of saccharification.

Keywords— RSM, saccharification, FCCCD, enzyme, sugar

INTRODUCTION

Saccharification is simply defined as a way of getting smaller units of sugar like glucose from higher

complex molecules such as cellulose. Saccharification is also called hydrolysis due to the involvement of water.

Cellulose is one of the well-known polysaccharides on earth that can be broken down into its primary

building units using chemicals or biological catalysts, enzyme (Sun & Cheng, 2002). However, cellulose-

containing biomass must first be partially degraded to allow hydrolytic enzymes or chemicals to attack the cellulosic polysaccharides and convert them into monosaccharides which can be easily fermented by organisms

into bio-ethanol (Mood et al., 2013).

59

Saccharification process can be affected by several internal and/or external factors that may interfere with the

rate of hydrolysis and enzyme action. These factors are: type of substrate, the temperature of the saccharification process, pH of the medium, substrate concentration, enzyme concentration and addition of surfactant (Jagatee et al.,

2015; Pandiyan et al., 2014).

Delignification of sawdust was done using laccase enzyme as a pretreatment agent, in this study, instead of

conventional chemical, physical or biological pretreatment. Fermentable sugar was produced through enzymatic saccharification followed by the enzymatic delignification of the biomass. The optimization of the saccharification

process of the enzyme-pretreated sawdust was studied using OFAT and FCCCD of the RSM. Different process

parameters such as pH, cellulose enzyme concentration, reaction time, and temperature were studied. The laccase pretreated sawdust was subjected to the saccharification process using cellulase enzyme as a hydrolyzing agent. The

optimization study of the saccharification process was first carried out using OFAT studies to examine the effect of

individual parameter on the rate of saccharification.

EXPERIMENTAL

Sawdust was collected from the residues of different wood mills in Pahang, Malaysia. The sample was washed

and dried until a constant weight was reached, after which the sample was milled to 1 mm size. The sawdust was

pretreated with laccase enzyme 51003 from Myceliophthora thermophilia, supplied by Novozymes, Bagsværd,

Denmark. The pretreatment of the sawdust was carried out by the process condition optimized by Amani et al

(2016).

Analytical Methods

Lignin, cellulose, hemicellulose and ash content of sawdust were determined by the sequential fractionation of

sawdust according to the method reported by Datta (1981). Reducing sugar content was assayed using the

Dinitrosalicylic Acid (DNS) method of (Wood et al., 2012). A powder form of cellulose enzyme was dissolved in

de-ionized water (1 IU/µL) to estimate cellulose enzyme activity. Enzyme activity was estimated by using filter

paper assay. One filter paper unit (FPU) was defined as the amount of enzyme that will liberate one micromole of

reducing sugar per minute and the unit was expressed in international units (IU).

Optimization of the Saccharification Process

Saccharification process of the optimized pretreated sawdust with cellulase enzyme was conducted by using

OFAT followed by RSM. Cellulase enzyme concentration, medium pH, reaction duration and temperature were

studied using OFAT method. Based on the result from OFAT the enzymatic saccharification process was optimized

using RSM. Design-Expert Version 6.0.8 was used to design the experimental conditions using FCCCD under RSM.

Two factors (pH and temperature) were studied because of their effect on the process during the OFAT studies. A sawdust concentration of 5% (w/v), agitation rate of 150 rpm and cellulase enzyme concentration of 10 IU/g were

used in the reaction system of 10 mL. The responses were the percentage yield of reducing sugar by sawdust content

and the percentage weight loss and the design of the parameters is shown in the Table 1.

TABLE 1. Design of the parameters for enzymatic saccharification of sawdust in RSM

Factor Name Units Low

Actual

High Actual Low

Coded

High

Coded

A Time hr 12.00 60.00 -1.00 1.00 B pH - 3.00 7.00 -1.00 1.00

C Temperature °C 30.00 60.00 -1.00 1.00

60

RESULTS AND DISCUSSION

Sawdust Characterization

The sawdust constituents were characterized before and after pretreatment with laccase enzyme at optimized

conditions using the method of (Datta, 1981), and the results were shown in Table 2. Lignocellulosic biomass

consists mainly of cellulose and hemicellulose, with an appreciable amount of lignin that forms rigidity to biomass cell along with the sugar bases. Cellulose is higher than that found in Meranti wood sawdust by Rafiqul & Sakinah

(2012) of percentage 41.06. Where water soluble content 5.9% is less than 7.15 % that found by Huang et al. (2015).

Similar results were reported by Ishmael et al. (2016) and Shah et al. (2016) for the pretreated and non-pretreated

OPEFB (Oil Palm Empty Fruit Bunches) as well.

Table 2. Constituents of Characterized sawdust

Sample Water soluble part

(%)

H-cellulose (%)

Cellulose (%)

Lignin (%) Ash content

(%)

Non- pretreated 5.9 30.9 41 26 2.1 Pretreated - 33.0 50 17 -

OFAT study for Enzymatic Saccharification of Pretreated Sawdust

Effect of cellulase enzyme concentration, medium pH, reaction duration and temperature on saccharification of

the optimized pretreated sawdust was studied using OFAT optimization process. Reducing sugar yield and weight

loss percentages were the responses of process and they were expressed as percentage (%) of reducing sugar yield

per mg of sawdust and percentage of weight loss. The saccharification yield of reducing sugars were calculated

using methods described by (Chen & Dixon, 2007; De Farias Silva et al., 2015).

Enzyme Concentration Effect

Aliquots of cellulase enzyme (5, 10, 20, 30 and 40 IU/g) were used to study the effects on the saccharification

process. The results showed saccharification rate increases with the increasing concentration of the enzyme (Figure 1). The outcome of this study was found to be consistent with previous findings by Phuengjayaem et al (2014) and

Shah et al (2016).

pH Effect

From the results, it was evident that pH is the most important factor that determines enzymatic saccharification

rate of the pretreated sawdust. It was found from the study (Figure 2) that the optimum pH for the saccharification

process was pH 5 and any changes in pH caused a reduction in the responses as proteins are denaturation by gaining

or losing electrons. It is well-known that all enzymes work within a range of pH and most hydrolytic enzymes work

at a pH between 3 and 6 (Phuengjayaem et al., 2014; De la Torre et al.; 2017).

Effect of Reaction time

The saccharification rate increases with time when the enzymes are effectively attached with the biomass but

with the less active enzymes, longer time could have little or no effects on the sugar yield. The results presented in

Figure 3 showed that after 48 h, the rate of sugar production reduced even with the increase in time. Reducing sugar

yield and weight loss were 0.14% (w/w) and 5.8% (w/w) respectively after half day of saccharification but when

prolonged to 24 and 36 h, there was a little increase in the percentages of reducing sugar and weight loss. Previous

61

studies were also reported the similar trends by many researchers (Kalhorinia et al., 2013; Phuengjayaem et al., 2014).

FIGURE 1. Effect of cellulase enzyme concentration on the saccharification of laccase pretreated sawdust

FIGURE 2. Effect of medium pH on the saccharification of laccase pretreated sawdust using process

FIGURE 3. Effect of reaction duration on the saccharification

of laccase pretreated sawdust using

FIGURE 4. Effect of temperature on the saccharification of

laccase pretreated sawdust

Effect of Temperature

From the results shown in Figure 4, the best performance of enzymatic saccharification of the pretreated sawdust was observed at 50°C. Saccharification at temperatures below or above 50 oC showed decreased saccharification

responses. Many previous studies have reported the range of temperature is 40-60 °C for better saccharification of

different agro-industrial natural solid wastes (Sun and Cheng, 2002;Pandiyan et al., 2014; Sakimoto et al., 2017).

Optimization of Enzymatic Saccharification of Pretreated sawdust by RSM

Optimization of the enzymatic saccharification process condition of the laccase pretreated sawdust was studied

statistically using FCCCD of the RSM. Temperature and pH were the varied parameters during the RSM studies

while the responses were the reducing sugar expressed in mg/mL and weight loss expressed in percentage after

saccharification. Due to page limitation all the detailed data of the RSM investigation were not given.

The ANOVA results suggested that the model is significant as the model F-value was 160.56 while the p-value was < 0.0001. The model accuracy was assessed by plotting the experimental and predicted values of weight loss

62

(%) response and the reducing sugar mg/ml response. The results calculated as a mean of the triplicates showed a

coefficient of determination (R^2) of 0.9966 for the weight loss and 0.9875 for reducing sugar. Based on these

values, the model was considered reliable as the R^2 values were close to 1.0 or > 0.75 (Ferreira et al., 2009).

After studying the 3D plots, RSM statistical design and quadratic polynomial equation (data not shown) the

model was validated. A total of 5 sets of data were chosen and tested from the suggested solutions (data not shown)

to validate the predicted model. In summary, it can be concluded that the optimal saccharification conditions obtained from this statistical study was 5% of the laccase-pretreated sawdust, cellulase enzyme concentration of 30

IU/g, pH 5, and a temperature of 50°C which yielded a maximum reducing sugar yield of 4.5 mg/ml after 36 h of

saccharification. Other studies also reported the optimal saccharification condition which is close to the current

findings (Cara et al., 2008; Jeya et al., 2010).

Comparative Investigation of the Optimized Saccharification Condition of Sawdust

Saccharification of the sawdust was investigated under different condition of sawdust such as without enzymatic

pretreatment, subsequent pretreatment & saccharification and separate pretreatment & saccharification of sawdust. It

was found from the study that separate ptretreatment & saccharification of sawdust yielded the best sugar (about 4.5

mg/ml).

ACKNOWLEDGMENTS

The authors are indebted to the Universiti Malaysia Pahang for the financial contribution to this project under

Grant No. RDU1703195.

REFERENCES

[1] Sun, Y., and Cheng, J. (2002). "Hydrolysis of lignocellulosic materials for ethanol production: A review,"

Bioresource Technology 83(1), 1-11. DOI: 10.1016/S0960- 8524(01)00212-7

[2] Mood, S. H., Golfeshan, A. H., Tabatabaei, M., Jouzani, G. S., Najafi, G. H., Gholami, M., & Ardjmand, M.

(2013). Lignocellulosic biomass to bioethanol, a comprehensive review with a focus on pretreatment.

Renewable and Sustainable Energy Reviews, 27, 77-93.

[3] Jagatee, S., Pradhan, C., Dash, P. K., Sahoo, S., & Mohanty, R. C. (2015). Optimization for saccharification of

sweet potato (Ipomoea batata) flour for enhanced ethanol production. International Journal of Science,

Technology and Management, 4(01), 67-76.

[4] Pandiyan, K., Tiwari, R., Singh, S., Nain, P. K., Rana, S., Arora, A., Singh, S. B., & Nain, L. (2014). Optimization of enzymatic saccharification of alkali pretreated Parthenium sp. using response surface

methodology. Enzyme research, 2014.

[5] Amani M. H., Essam A. M., Shah S. R. (2017). Laccase as bio-pretreatment step of sawdust for ethanol

production: optimization and statistical modeling. FGIC 1st Conference on Governance & Integrity, 2017, 3 – 4

April 2017, Yayasan Pahang, Kuantan, Malaysia

[6] Datta, R. (1981). Energy requirements for lignocellulose pretreatment processes. Process Biochemistry, 16, 16.

[7] Wood, I. P., Elliston, A., Ryden, P., Bancroft, I., Roberts, I. N., & Waldron, K. W. (2012). Rapid quantification

of reducing sugars in biomass hydrolysates: improving the speed and precision of the dinitrosalicylic acid assay. Biomass and Bioenergy, 44, 117-121.

[8] Rafiqul, I., & Sakinah, A. M. (2012). Kinetic studies on acid hydrolysis of Meranti wood sawdust for xylose

production. Chemical Engineering Science, 71, 431-437.

[9] Huang, Y., Wei, X., Zhou, S., Liu, M., Tu, Y., Li, A., Chen, P., Wang, Y., Zhang, X., & Tai, H. (2015). Steam

explosion distinctively enhances biomass enzymatic saccharification of cotton stalks by largely reducing cellulose polymerization degree in G. barbadense and G. hirsutum. Bioresource technology, 181, 224-230.

63

[10] Ishmael et al. (2016). “EFB saccharification,” BioResources 11(2), 5013-5032. Statistical Modeling and Optimization of Enzymatic Pretreatment of Empty Fruit Bunches with Laccase Enzyme. Ukaegbu Chinonso

Ishmael,* a Samiur Rashid Shah,a Jaya Vejayan Palliah,a Mohammed Fazli Farida Asras,a Sharifah Suhaiza

Binti Nik Wan Ahmad,a and Victor Bamidele Ayodele

[11] Shah et al. (2016). “EFB saccharification,” BioResources 11(2), 5138-5154. Optimization of the Enzymatic

Saccharification Process of Empty Fruit Bunch Pretreated with Laccase Enzyme. Samiur Rashid Shah, Ukaegbu

Chinonso Ishmael*, Jaya Vejayan Palliah, Mohammed Fazli Farida Asras, and Sharifah Suhaiza Binti Nik Wan Ahmad

[12] Chen, F., & Dixon, R. A. (2007). Lignin modification improves fermentable sugar yields for biofuel production.

Nature biotechnology, 25(7), 759.

[13] De Farias Silva, C. E., Gois, G. N. S. B., Da Silva, L. M. O., Almeida, R. M. R. G., & De Souza Abud, A. K.

(2015). Citric waste saccharification under different chemical treatments. Acta Scientiarum. Technology, 37(4),

387.

[14] De la Torre, I., Ravelo, M., Segarra, S., Tortajada, M., Santos, V. E., & Ladero, M. (2017). Study on the effects

of several operational variables on the enzymatic batch saccharification of orange solid waste. Bioresource technology, 245, 906-915.

[15] Phuengjayaem, S., Poonsrisawat, A., Petsom, A., & Teeradakorn, S. (2014). Optimization of saccharification

conditions of acid-pretreated sweet sorghum straw using response surface methodology. Journal of Agricultural

Science, 6(9), 120.

[16] Kalhorinia, S., Naseeruddin, S., Yadav, K., Goli, J., & Rao, L. (2013). Optimization of acid and enzymatic saccharification of lignocellulosic substrate Water Hyacinth (Eichhornia crassipes). Indian Streams Research

Journal, 3(9), 1-10.

[17] Sakimoto, K., Kanna, M., & Matsumura, Y. (2017). Kinetic model of cellulose degradation using simultaneous

saccharification and fermentation. Biomass and Bioenergy, 99, 116-121.

[18] Ferreira, S., Duarte, A. P., Ribeiro, M. H., Queiroz, J. A., & Domingues, F. C. (2009). Response surface

optimization of enzymatic hydrolysis of Cistus ladanifer and Cytisus striatus for bioethanol production.

Biochemical Engineering Journal, 45(3), 192-200.

[19] Cara, C., Ruiz, E., Oliva, J. M., Sáez, F., & Castro, E. (2008). Conversion of olive tree biomass into fermentable

sugars by dilute acid pretreatment and enzymatic saccharification. Bioresource technology, 99(6), 1869-1876.

[20] Jeya, M., Moon, H.-J., Kim, S.-H., & Lee, J.-K. (2010). Conversion of woody biomass into fermentable sugars

by cellulase from Agaricus arvensis. Bioresource technology, 101(22), 8742-8749.

64

Implementation Of A Deep Learning Neural Network Model For

Partial Optimization Of Cellulase Enzyme Production Process

Saiful Azada, Shah Samiur Rashid

b*, Jalal K C A

c

aFaculty of Computer Systems and Software Engineering, Universiti Malaysia Pahang,

Gambang, Kuantan 26300, Pahang Darul Makmur, Malaysia bFaculty of Industrial Sciences & Technology, Universiti Malaysia Pahang, Gambang,

Kuantan 26300, Pahang Darul Makmur, Malaysia cKulliyyah of Science, International Islamic University Malaysia, Kuantan, Malaysia

*Corresponding Author: [email protected]

ABSTRACT

Cellulase enzyme is one of the important industrial enzymes. Achieving higher productivity

using either synthetic or organic medium is a great challenge. Response surface methodology

(RSM) is widely being used for optimizing production process parameters to achieve higher

productivity and it is considered as one of the shortest/efficient routes/methods in this regard.

However it requires a number of set of sequential investigations. Following the successes of

artificial neural network in different areas, a deep learning model based on Artificial Neural

Network was employed for the partial optimization of the production of cellulase enzyme

using palm oil mill effluent (POME) as a basal medium. In this study, eleven (11) parameters namely TSS (total sedimented solid) of POME, cassava powder, wheat powder, table sugar,

cellulose powder, peptone, (NH4)2SO4, KH2PO4, Tween 80, MnSO4.H2O, MgSO4.7H2O.

After investigation by the deep learning method and comparing the simulated data with the

previous published report, it could be concluded that the deep learning neural network can be

suitably employed for the optimization of the bioprocess of cellulase enzyme production.

Keywords: Cellulase Enzyme, Optimization, Artificial Neural Network, Deep Learning, OFAT

1. INTRODUCTION Cellulase is one of the most important industrial enzymes and it also can play key role to earn

benefits for biomass utilization. Cellulase enzymes such as exocellobiohydrolase and

endoglucanase have wide spread applications in the feed, textiles, pulp and paper industries

(Wen et al, 2005). In the biomass management this enzymes are applied through

bioconversion/biodegradation of the most abundant cellulosic natural material into simple

fermentable sugars such as glucose, fructose, cellobiose and different oligosaccharides

(Heikinheimo et al, 2000). The most well-known industrial applications of this enzyme is in

the textile and garment industries as a biostoning agent. It was also reported that cellulase

enzyme, EGII, from T. reesei found to be the most effective in removing color from denim,

therefore produces an attractive stonewashing effect (Buchert and Heikinheimo, 1998).

Endoglucanases are widely used in the feed industry to degrade β-glucans of feed; therefore,

feed quality is improved through lowering the viscosity of the intestinal contents (Bedford,

1995). Among the industrial enzymes cellulase enzyme is the second highest contributor and

global cellulase enzyme business value is more than $ 400 million in 2017 and each year

enzyme business is expanded by around 20%. Therefore, cellulase enzyme market is

considered as one of the most promising area of white biotechnology (BCC research, 2017).

Malaysia is one of the leaders in palm oil production and according to the report of MPOB

(Malaysian Palm Oil Board) in 2017 it was produced around 20 million tons of crude palm

oil. About 0.5-0.7 tons of POME is produced as an effluent for processing of each ton of

OPFFB (oil palm fresh fruit bunches) in which more than 40 million tons of POME is

discharged from the palm oil industry in Malaysia every year (MPOB Report, 2018).

65

Considering abundance and high nutrient values, POME can be used to produce different

sustainable bioconversion products such as organic acid, enzyme etc. (Rashid et al, 2009).

Synthetic culture medium requires various nutrient and minerals in the medium composition during enzyme production (Macris et al, 1989; Krishna, 1999). The most important part of economic bioconversion/biodegradation process is to identify the suitable medium components for higher productivity of enzyme. Cellulase enzyme production may requires several medium constituents such as glucose, yeast extract, peptone, urea, KH2PO4,

(NH4)2SO4, MgSO4, FeSO4, MnSO4, CoCl2, CaCl2, but not all the before said components are

required in the same medium (Niranjane et al, 2007; Martins et al., 2008). Therefore selection process for major medium components is a very important step of bioprocess. Researchers apply different techniques such as OFAT (One-factor-at-a-time) and statistical modelling to optimize the process condition (Ishmael et al, 2016).

OFAT study involves one varied variable and kept other variables constant in the same

investigation. This technique becomes laborious as it requires a number of set of experiments

and moreover it doesn’t explain complete or broader effects of the involved variables of the

process. In addition OFAT even can’t outline combined interactions of the process parameters

(Shah et al, 2016). Therefore, RSM together with OFAT gives a comprehensive result to

identify optimized process parameters. Recently, it was found that artificial intelligence and

evolutionary computing become a dependable option to explain different biological

challenges. Hence, ANN has become a popular choice for solving problems in different

biotechnological applications ranges from chromatographic pattern recognition and

expression profiles to the sequence analysis in proteomics and genomics (Eriola et al, 2015).

In many instances scientists reported about better performance of ANN-based models over

RSM in the predictions of different biological processes. Therefore, in this present study, a

deep learning neural network based model was employed to optimize medium constituents for

cellulase enzyme production.

2. Experimental

In this study, based on the literature review and outcome reported by Rashid et al (2009)

several lab-scale experiments were carried out for deep learning interaction investigation of

the 11 fermentation/biodegradation medium constituents on the productivity of cellulase

enzyme (CMCase, U/ml). The experimental results used for the neuron training of the deep

learning network were presented in the Table 1. The medium constituents of the

fermentation/biodegradation process are TSS (total sedimented solid) of POME, cassava powder, wheat powder, table sugar, cellulose powder, peptone, (NH4)2SO4, KH2PO4, Tween 80, MnSO4.H2O, MgSO4.7H2O.

2.1. Analytical Methods

After 5 days of fermentation/biodegradation with Trichoderma reesei, samples (fermented

broth) were filtered using Whatman no. 1 filter paper and the filtrate was assayed for

endoglucanase activity (Ghose, 1987). The endoglucanase activity was measured using

carboxymethyl cellulose (CMC) as a substrate by carboxymethyl cellulase assay (CMCase)

and the derived unit for CMC assay is CMCase cellulase (U/ml). One CMCase unit is the

concentration of enzyme that can release 0.5 mg of glucose from 0.5 ml of the substrate CMC

in 30 min. TSS was determined according to the standard method of APHA (APHA, 1989).

66

Table 1. Experimental results with different medium compositions

Run

A B C D E F G H I J K CMCase

%(w/v) U/ml

1 1.25 1.00 1.00 1.00 1.00 0.25 0.25 0.13 0.10 0.03 0.01 8.39

2 1.25 1.00 1.00 1.00 1.00 0.25 0.25 0.13 0.10 0.03 0.01 8.85

3 1.25 1.00 1.00 1.00 1.00 0.25 0.25 0.13 0.10 0.03 0.01 8.60

4 0.50 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.21

5 0.50 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.25

6 0.50 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.30

7 2.00 0.00 2.00 0.00 0.00 0.00 0.50 0.25 0.20 0.00 0.02 0.06

8 2.00 0.00 2.00 0.00 0.00 0.00 0.50 0.25 0.20 0.00 0.02 0.07

9 2.00 0.00 2.00 0.00 0.00 0.00 0.50 0.25 0.20 0.00 0.02 0.06

10 2.00 0.00 2.00 2.00 0.00 0.50 0.00 0.00 0.00 0.05 0.02 0.00

11 2.00 0.00 2.00 2.00 0.00 0.50 0.00 0.00 0.00 0.05 0.02 0.01

12 2.00 0.00 2.00 2.00 0.00 0.50 0.00 0.00 0.00 0.05 0.02 0.00

13 2.00 0.00 0.00 0.00 2.00 0.50 0.50 0.00 0.20 0.05 0.00 0.07

14 2.00 0.00 0.00 0.00 2.00 0.50 0.50 0.00 0.20 0.05 0.00 0.07

15 2.00 0.00 0.00 0.00 2.00 0.50 0.50 0.00 0.20 0.05 0.00 0.05

16 2.00 2.00 0.00 2.00 2.00 0.00 0.50 0.00 0.00 0.00 0.02 0.01

17 2.00 2.00 0.00 2.00 2.00 0.00 0.50 0.00 0.00 0.00 0.02 0.02

18 2.00 2.00 0.00 2.00 2.00 0.00 0.50 0.00 0.00 0.00 0.02 0.00

19 0.50 0.00 2.00 2.00 2.00 0.00 0.50 0.25 0.00 0.05 0.00 3.72

20 0.50 0.00 2.00 2.00 2.00 0.00 0.50 0.25 0.00 0.05 0.00 3.80

21 0.50 0.00 2.00 2.00 2.00 0.00 0.50 0.25 0.00 0.05 0.00 3.60

22 0.50 2.00 0.00 0.00 0.00 0.50 0.50 0.25 0.00 0.05 0.02 0.90

23 0.50 2.00 0.00 0.00 0.00 0.50 0.50 0.25 0.00 0.05 0.02 1.10

24 0.50 2.00 0.00 0.00 0.00 0.50 0.50 0.25 0.00 0.05 0.02 0.85

25 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.25 0.20 0.05 0.00 0.08

26 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.25 0.20 0.05 0.00 0.09

27 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.25 0.20 0.05 0.00 0.08

28 0.50 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.20 0.05 0.02 3.95

29 0.50 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.20 0.05 0.02 4.02

30 0.50 2.00 2.00 0.00 2.00 0.00 0.00 0.00 0.20 0.05 0.02 4.20

31 0.50 2.00 2.00 2.00 0.00 0.50 0.50 0.00 0.20 0.00 0.00 1.26

32 0.50 2.00 2.00 2.00 0.00 0.50 0.50 0.00 0.20 0.00 0.00 1.15

33 0.50 2.00 2.00 2.00 0.00 0.50 0.50 0.00 0.20 0.00 0.00 1.30

34 2.00 2.00 2.00 0.00 2.00 0.50 0.00 0.25 0.00 0.00 0.00 0.08

35 2.00 2.00 2.00 0.00 2.00 0.50 0.00 0.25 0.00 0.00 0.00 0.09

36 2.00 2.00 2.00 0.00 2.00 0.50 0.00 0.25 0.00 0.00 0.00 0.10

37 0.50 0.00 0.00 2.00 2.00 0.50 0.00 0.25 0.20 0.00 0.02 7.50

38 0.50 0.00 0.00 2.00 2.00 0.50 0.00 0.25 0.20 0.00 0.02 7.50

39 0.50 0.00 0.00 2.00 2.00 0.50 0.00 0.25 0.20 0.00 0.02 8.08

A= TSS of POME; B= Cassava powder; C= Wheat flower; D= Sugar; E= Cellulose; F= Peptone; G=

(NH4)2SO4; H= KH2PO4; I= Tween 80; J= MnSO4.H2O; K= MgSO4.7H2O

67

2.2. Simple Deep Learning Model

Deep learning models enhanced the capacity of the standard artificial neural network to be

used to analyse different data sets (biological or commercial) with high precision of prediction. Dataset presented in Table 1 was used as a training and test data. After this, a set

of OFAT data was prepared for each of the parameters in a certain range (Table 2) to predict

possible optimized value or range by the designed deep learning model. TensorFlow software

was used as a computational framework.

Table 2. Parameters’ ranges and step of increment

Parameters

A B C D E F G H I J K

Range 0.5-5.0 0-5 0-5 0-5 0-5 0-3 0-3 0-2 0-1 0-1 0-1

Step 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.1 0.05 0.05 0.05

A= TSS of POME; B= Cassava powder; C= Wheat flower; D= Sugar; E= Cellulose; F= Peptone; G=

(NH4)2SO4; H= KH2PO4; I= Tween 80; J= MnSO4.H2O; K= MgSO4.7H2O

3. Results and discussion Deep learning neural network (DLNN) was employed in this study to exclude the non-

interacted or negatively interacted parameters for optimization of medium constituents to

produce cellulase enzyme using POME as a basal medium. Usually, process optimization is

conducted in three stages such parameter screening, single factor study i.e. OFAT and

statistical optimization i.e. RSM (Shah et al, 2016). This study was started with the laboratory

experiment of cellulase enzyme production using a fermentation medium containing 11

interaction parameters (Table 1). The neurons were trained using the obtained data where

enzyme activity CMC (U/ml) was used as output. Based on this experiment (Table 1) 11 set

of data were prepared as if OFAT data set to be investigated using DLNW to exclude non-

interacting or negatively interacting parameters for further examination achieving optimized

medium condition. There are few reports can be found of using neural network (Das et al,

2015; Eiola et al, 2015) in bioprocess study and hardly it was found any report of using deep

learning in this kind of study.

Simulated results of five (5) parameters were presented by Figure 1. The parameter A was

studied from 0.5 to 5.0 and the incremental value from the starting point was 0.25. Whereas

the all other four (4) parameters were studied from 0.0 to 5.0 and the incremental value was

same. It was found from the investigation that at lower TSS, higher enzyme activity was

predicted. As it is impossible to discard the TSS completely and higher TSS is sustainable,

therefore, it was decided to fix the TSS value at 2% (w/v).

68

En

zym

e A

ctiv

ity C

MC

(U

/ml)

10.00

9.00

8.00

7.00

6.00

5.00

4.00

3.00

2.00

1.00

0.00

A

B

C

D

E

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00 3.25 3.50 3.75 4.00 4.25 4.50 4.75

Medium Constituents Value

A= TSS of POME; B= Cassava powder; C= Wheat flower; D= Sugar; E= Cellulose

Figure 1. OFAT study results of five (5) parameters predicted by DLNN

Deep learning simulated trend suggested that at even without cassava powder, the bioprocess

system can yielded a considerable amount of enzyme. Hence, cassava powder is considered as

a negatively interacted parameter that can be excluded from the medium composition. The

third parameter, wheat flower, found to be investigated further in a range of 0 % - 2.0 % (w/v)

based on the nature of interaction simulated by deep learning algorithm. Sugar is one of the

very important and easily digestible nutrients (Kusum et al, 2018) and it was also evident in

this study to be interacted from 0.0 to 3.0 % (w/v). Cellulose also found to be interacted

between 0.0 to 3.0 % (w/v). In many bioprocess systems, cellulose shows influence on the

production of many bio-products (Furkan and Becer, 2015). It was predicted maximum

production (8.63 CMC U/ml) at 1% (w/v). Therefore, based on the nature of the simulated

investigation it can be fixed at 0.75 % (w/v) or investigated further.

Results of the remaining six (6) parameters were presented below in Figure 2. Three parameters namely peptone, (NH4)2SO4 and KH2PO4 found to be influential around 0.7 %

(w/v), but the last three (3) parameters such as Tween 80, MnSO4.H2O and MgSO4.7H2O could be excluded from the enzyme production medium as they didn’t show any influence on

the predicted study. Based on the simulated results, as peptone, (NH4)2SO4 and KH2PO4 can

be studied further either in-silico or through laboratory experiments within a certain range based on the current findings.

69

En

zym

e A

ctiv

ity C

MC

(U

/ml)

10.00

9.00

8.00

7.00

6.00

5.00

4.00

3.00 F G H I

2.00 J K

1.00

0.00

Medium Constituents Value

F= Peptone; G= (NH4)2SO4; H= KH2PO4; I= Tween 80; J= MnSO4.H2O; K= MgSO4.7H2O

Figure 1. OFAT study results of six (6) parameters predicted by DLNN

It has already been mentioned that the current DLNN study was designed on the basis of the

obtained experimental results following the medium composition employed in the Plackett-

Burman (PB) design reported by Rashid et al. (2009). The most interesting aspect of this in-

silico study is that the decision from the simulated model support mostly with the

experimental results of the PB design and OFAT study reported by Rashid et al (2009).

Therefore, it could be concluded that deep learning neural network can be used in the

optimization of cellulase enzyme production process.

Acknowledgments The authors are grateful to the Universiti Malaysia Pahang for funding this project under

Grant No. RDU1703195.

References [1] Wen et al., 2005 Z. Wen, W. Liao and S. Chen, Production of cellulase by Trichoderma

reesei from dairy manure, Bioresour. Technol. 96 (2005), pp. 491–499

[2] Heikinheimo L, Buchert J, Miettinen-Oinonen A, Suominen P (2000) Treating denim

fabrics with Trichoderma reesei cellulases. Textile Res J 70:969–973

[3] Buchert J, Heikinheimo L (1998) New cellulase processes for the textile industry.

Carbohyd Europe 22:32–34

[4] BCC Research (2018) https://www.bccresearch.com/pressroom/egy/global-biofuel-

enzyme-market-value-expected-reach-nearly-$915-million-2017) [5] MPOB report (2018) http://bepi.mpob.gov.my/stat/web_report1.php?val=201703

[6] Macris BJ, Kekos D, Evrangelidou X (1989) A simple and inexpensive method for

cellulase and b-glusidase production by Aspergillus niger. Appl Microbiol Technol

31:150–151

[7] Krishna C (1999) Production of bacterial cellulases by solid state bioprocessing banana

wastes. Bioresour Technol 69:231–239

[8] Niranjane AP, Madhou P, Stevenson TW (2007). The effect of carbohydrate carbon

sources on the production of cellulase by Phlebia gigantean. Enzyme & Microbial

Technol 40: 1464-1468

70

[9] Martins LF, Kolling D, Camassola M, Dillon AJ, Ramos LP (2008). Comparison of

Penicillium echinulatum and Trichoderma reesei cellulases in relation to their activity

against various cellulosic substrates. Bioresour technol 99(5):1417-24.

[10] Rashid S S, Alam M Z, Karim M I A, Salleh, M H (2009). Optimization of the Nutrient

Supplients for Cellulase Production with the Basal Medium Palm Oil Mill Effluent World Academy of Science, Engineering and Technology 60 2009.

[11] Ishmael U. C., Samiur R. S., Jaya V. P., Asras M.F.F., Sharifah S.B.N.W.A., Victor

B.A. (2016) Statistical Modeling and Optimization of Enzymatic Pretreatment of

Empty Fruit Bunches with Laccase Enzyme. BioResources 11(2), 5013-5032

[12] Shah S.R., Ukaegbu C.I., Jaya V.P., Asras M.F.F, Sharifah S.B.N.W.A. (2016).

Optimization of the Enzymatic Saccharification Process of Empty Fruit Bunch

Pretreated with Laccase Enzyme. BioResources 11(2), 5138-5154 [13] Eriola B, Abiola E.T. (2015) Modeling and optimization of bioethanol production from

breadfruit starch hydrolyzate vis-a-vis response surface methodology and artificial

neural network. Renewable Energy 74 (2015) 87-94 [14] Kusum L. Manisha S.Satya N. P. Rajender S. S. Sudhir P. S. (2018). An integrated bio-

process for production of functional biomolecules utilizing raw and by-products from

dairy and sugarcane industries. Bioprocess and Biosystems Engineering (2018)

41:1121–1131

[15] Furkan H. I. and Becer C. R. (2015). Lignocellulosic biomass: a sustainable platform

for the production of bio-based chemicals and polymers. Polym. Chem., 2015, 6, 4497-

4559

[16] Hashim, F. S., et. al., “Enzymatic Hydrolysis of Pretreated Fibre Pressed Oil Palm

Frond by using Sacchariseb C6,” IOP Conf. Ser.: Mater. Sci. Eng., 2017.

[17] Marcos, M., García-Cubero, M. T., González-Benito, G., Coca, M., Bolado, S., Lucas,

S., “Improvement of Enzymatic Hydrolysis of Steam-exploded Wheat Straw by

Simultaneous Glucose and Xylose Liberation, Chem. Biochem. Eng. Q., vol. 27, no. 4,

2013, pp. 499-509. [18] Ghose TK (1987) Measurement of cellulase activities. Pure Appl Chem 59:257–268.

[19] APHA (1989) Standard methods for the examination of water and wastewater, 17th

edn.America PublicHealth Association,Washington Arhan Y, Oztuk I, Ciftci T (1996)

Settling and dewatering characteristics of sludge from Baker’s yeast production

wastewater treatment. Water Sci Technol 34:459–467.

[20] Das S., Bhattachary A., Haldar S., Ganguly A., Sai G, Ting Y.P., Chatterjee P.K.

(2015) Optimization of enzymatic saccharification of water hyacinth biomass for bio- ethanol: Comparison between artificial neural network and response surface

methodology. Sustainable Materials and Technologies 3 (2015) 17–28

71

Particle Grain Analysis of Sand Samples from North Malay

Basin

Angga Pratama Herman1, a), Muhammad Luqman Hasan1, b), Noor Ilyana Ismail1, c),

Nasiman Sapari2, d), Mohd Azuwan Maoinser1, e), and Jirapha Skulsangjuntr3, f)

1Department of Petroleum Engineering, Universiti Teknologi PETRONAS, Malaysia. 2Department of Civil and Environmental Engineering, Universiti Teknologi PETRONAS, Malaysia.

3Malaysia-Thailand Joint Authority, Malaysia.

a)[email protected]

b)Corresponding author: [email protected]

c)[email protected]

d)[email protected]

e)[email protected]

f)[email protected]

Abstract. Sand production is an undesirable phenomenon as it can cause accumulation of sand in the wellbore, surface

equipment and eventually lead to production lost. It has been proven that the reservoirs in North Malay Basin are poorly

compacted, have minor cementation and are referred to as being poorly consolidated reservoirs that may lead to sand

production. Sand production issue can be managed using various sand control techniques and the particle grain analysis

data usually will provide important information needed in order to select the suitable completion solution for the

respective reservoir. Thus, the aim of this study is to analyze the sand particle from the North Malay Basin in terms of

particle size distribution, uniformity and sorting coefficient, percentage of fines, grain classification, roundness and

sphericity classification. It was found out that the sand particle size distribution for the sand sample was single peak with

the maximum size of 158.49 µm and frequency of 9.93 volume %. The D10, D40, D50, D90 and D95 were determined by

analysing the S-Curve graph and it was found out that the values for LPSA were slightly smaller compared to the dry

sieve analysis. In contrast to that, the percentage of fines in LPSA was higher compared to dry sieve result with 13.34 %

and 5.24 % values respectively. In overall, the sand samples can be classified as highly uniform and well sorted sand. In

terms of roundness and sphericity, the sand sample can be classified as sub rounded with very low sphericity sand with

the values of 0.37 and 0.38 respectively. Finally, the sand sample grain size can be classified as fine sand to very fine

sand respectively due to the calculated Krumbein phi scale, Ø was ranging from 2.15 to 3.22 respectively.

Keywords—particle size distribution, sand production, north malay basin

INTRODUCTION

Nowadays, sand production is one of the major problems in oil and gas industry. Before drilling process, in-situ

stresses and pore pressure of a formation are in a static equilibrium. Completion and drilling activities including

perforations and oil production will reposition pore pressure and stresses over the production cavity. Under special

circumstances, sand particles in the formation will move from the reservoir into the well along with the hydrocarbon

flow when hydrocarbons are being produced from a reservoir [1].

This process is called as “sand production” and it can cause severe damages on surface production facilities such

as sand accumulations in the separators and erosion of pipelines and valves. Every sand sample is unique due to its

origin and depositional factors. According to Wu & Tan (2005), the problems regarding sand production costs oil

companies’ tens of billions of dollars yearly [2]. If the company manage to predict the occurrence of sand

production earlier and equip their wells with sand control equipment, they can save cost up to millions of dollars

yearly as the cost of sand control equipment and operation is around 2 to 11 millions of dollars yearly.

North Malay Basin consists of multiple gas bearing zones at depths of 3,500 – 10,000 feet and 180 feet of a

water depth which located approximately 186 miles (300 km) offshore of the Terengganu Gas Terminal in the Gulf

72

of Thailand. It is a natural gas asset with an approximated recoverable resource of more than 20 million barrels of

condensate and more than 1.5 trillion cubic feet of natural gas. The fields consist of Kesumba, Anggerik, Zetung,

Kamelia, Melati, Gajah, Teratai, Bunga Dahlia and Bergading [3]. According to Carney et al. (2010), the area of

North Malay Basin consists of a massive interval of more than 2,100 m of shales and interbedded sands with

abundance shales and coals [4].

The reservoirs in North Malay Basin are thought to be poorly compacted and have minor cementation. Thus, it is

referred to as being poorly consolidated reservoir and this may lead to sand production [4]. According to Adeyanju

& Olafuyi (2011), reservoir of an unconsolidated and weak sandstone formation may produce some amount of sand

content along with the production of the hydrocarbons [5]. Unconsolidated sands relatively consist of fine sand with

slightly greater porosity and more angular in shape than consolidated sands. The fine sand is more angular because it

is produced by chipping of the corners of a large pebble [6]. Lastly, in terms of particle size distribution (PSD),

according to Ibrahim (2013), well sorted sediment indicates that the grain size distribution is fairly uniform and

hence has increased porosity [7].

Thus, the sand production issue can be managed using various sand control options. The particle grain analysis

data usually will provide important information needed in order to select the suitable completion solution for the

respective reservoir. Hence, the purpose of this paper is to analyze the sand particle from the North Malay Basin in

terms of particle size distribution, uniformity and sorting coefficient, percentage of fines, grain classification,

roundness and sphericity classification.

MATERIAL AND METHODS

Field Emission Scanning Electron Microscope (FESEM)

FESEM test was conducted to characterize the grain morphology of the sand samples such as sphericity and

roundness of the grains. In order to conduct this test, the sand samples were first coated with electrically conductive

material of gold-palladium by putting the samples into the machine of Denton Desk II for 1 minute. Then, the coated

sand samples were placed into the FESEM, SUPRA 35VP machine until the spectra vacuum reaches 105 Mbar.

Finally, the sphericity and roundness of the sand samples can be observed by using a computer with SmartSEM

software.

Particle Characterization

The particle characterization consists of particle size distribution analysis using dry sieve and laser particle size

analyzer (LPSA) which can contribute to determine the sand sample uniformity coefficient, sorting coefficient,

percentage of fines, grain classification, sphericity and roundness rating of the sample.

Dry Sieve Analysis

The dry sieve analysis is a mechanical separation of particles and was carried out to determine particle size

distribution (PSD) of the sand samples. In this study, Endecotts EFL 2000 sieve machine was used and a sequence

of sieve opening was stacked from 600 µm, 425 µm, 300 µm, 212 µm, 150 µm, 63 µm, and 45 µm respectively.

Then, a sample of 100 g was prepared, weighed and the mass were recorded. Next, the split sample was poured onto

the top sieve and the stack of sieves, pan and cover were placed in the testing sieve shaker to be agitated for 10

minutes. Lastly, the sieves stack, pan and cover were removed from the testing sieve shaker and each sieve and pan

which now contain retained sand sample were weighed [8].

Laser Particle Size Analyzer (LPSA)

LPSA is an important technique to measure the accurate particle size distribution using modern laser diffraction

devices (LD). Due to its simplicity and accuracy, LPSA is nowadays the primary method for examination of the size

distribution [9]. In order to conduct this test, the minimum weight of sand samples that were required was

approximately 1 to 3 gram and this test can determine the particle size over a range of 0.1 µm to 5000 µm. In

comparison with the sieve analysis, laser particle size analysis test was faster, fully automated as it can be

standardized for certain systems and more reliable [10]. First, the sand samples were placed into the Mastersizer

73

2000 by Malvern Instruments. Then, the particles in the measuring zone were lighted up by the laser source. The

detectors at the focal plane of the Fourier lens measured the scattered lights. Then, the scattering data were

transformed to electrical signals and transferred to a computer to obtain the particle size distribution data.

Uniformity Coefficient, Sorting Coefficient and Percentage of Fines

Three important indicators to determine the formation sand quality are uniformity coefficient, sorting coefficient

and percentage of fines. These three indicators can be determined using the S – Curve graph from dry sieve analysis

or LPSA. Basically, the slope of the curve in the semi log graph will determine the formation sand quality. Using the

concept, the graph with a very high slope or almost vertical slope can be considered as good quality sand due to the

majority of the sand diameter do not vary very much. In terms of fines, anything below 40 microns can be classified

as fines and it also can be determined directly from the graph. In other words, the formation sand quality indicators

can be numerically defined as equation 1, 2 and 3 while the classification for the uniformity coefficient and sorting

coefficient can be presented in Table 1.

Uniformity Coefficient, Uc = D40 / D90 (1)

Sorting Coefficient, Sc = D10 / D95 (2)

Percentage of fines = Particles less than 40 µm (3)

TABLE 1. The classification of sorting and uniformity coefficient.

Uniformity Coefficient

Classification

Uc < 3 Highly uniform sand

3 < Uc < 5 Uniform sand

5 < Uc < 10 Moderately or poorly uniform sand

Uc > 10 Highly non uniform sand

Sorting Coefficient Sc < 10 Well sorted sand

Sc > 10 Poorly sorted sand

Grain Size Classification

The Wentworth scale is used to determine the grain sizes of the sediments. Fragmental particles of sedimentary

rocks cover a wide range of grain sizes from boulder which is greater than 250 mm to clay and dust which is less

than 0.0005 mm in diameter [6]. Since handling of such wider range of grain sizes is a challenging problem, a

geometric scale has been introduced and used instead of an arithmetic one. The Udden-Wentworth system not only

provided an appropriate scale for representation of grain size but also standardised the sedimentological

nomenclature [6]. According to Krumbein (1937), the sediments grain sizes can be determined by using the equation

4 [11].

(4)

Where Ø is the Krumbein phi scale, D is the diameter of the grain in millimetres and Do is a reference diameter

which is equal to 1 millimeter. The Wentworth scale for classifying and describing sediments grain size is shown in

the Table 2.

TABLE 2. The Udden-Wentworth scale [6].

Category Krumbein phi scale, Ø Type Grain Diameter (mm)

Boulder < -8 Boulders 250 and above

Gravel -6 to -8 Cobbles 65-250

-2 to -6 Pebbles 4-65

-1 to -2 Granules 2-4

Sand 0 to -1 Very coarse sand 1-2

0 to 1 Coarse sand 0.5-1

2 to 1 Medium sand 0.25-0.5

74

3 to 2 Fine sand 0.125-0.25

4 to 3 Very fine sand 0.0625-0.125

Mud 5 to 4 Coarse silt 0.031-0.0625

6 to 5 Medium silt 0.0156-0.031

7 to 6 Fine silt 0.0078-0.0156

8 to 7 Very fine silt 0.0039-0.0078

10 to 8 Clay <0.0039

20 to 10 Dust <0.0005

Roundness and Sphericity

A chart of sphericity against roundness has been used to determine the morphology of grains. According to

American Petroleum Institute (2016), ten individual particles were randomly selected in the field of view for the

sand sample [8].

FIGURE 1. The chart for visual estimation of sphericity and roundness [9].

Then, the sphericity and roundness of each chosen particle were determined and recorded by referring to the

chart shown in Figure 1. Next, the arithmetic average of the recorded sphericity and roundness of the particles were

calculated and classified based on classification in Figure 2. and Table 3. respectively.

FIGURE 2. Comparison of grain images with standard of roundness [12].

TABLE 3. Classification scheme of the of the sphericity [13].

Classification Class Limit Geometric Mid-Point

75

Very High Sphericity 0.949 – 1.000 0.949

High Sphericity 0.775 – 0.894 0.837

Moderate Sphericity 0.632 – 0.775 0.707

Low Sphericity 0.447 – 0.632 0.548

Very Low Sphericity 0.000 – 0.447 0.316

RESULTS AND DISCUSSIONS

Dry Sieve Analysis

The dry sieve analysis for the sand sample is presented in Table 4. while the S-Curve graph for the sand sample

is illustrated in Figure 3. respectively.

TABLE 4. Dry Sieve cumulative percentage of the sand sample.

Mesh Size Particle Interval

Size, µm

Weight Retain, g Frequency of

Occurrence, %

Cumulative

Percentage, %

600 > 600 6.24 15.14 100.00

500 500 – 599 1.08 2.62 84.86

425 425 – 499 1.08 2.62 82.24

300 300 – 424 2.65 6.43 79.62

212 212 – 299 4.81 11.67 73.19

150 150 – 211 8.91 21.62 61.52

63 63 – 149 4.73 11.48 39.91

45 45 - 62 9.56 23.19 28.43

Pan 1 - 44 2.16 5.24 5.24

Total 41.22 100.00 -

These data can be used to plot the particle size distribution curve which is also known as ‘S curve’. The curve

can be plotted where the cumulative weight (%) in the y-axis constructed against the sand grain diameter (µm) in the

x-axis. The scale of the x-axis is logarithmic scale while the y-axis has a scale of percentage (0 to 100%) using

linear scale. Then, from these curves, the values of D10, D40, D50, D90, D95 and percentage of fines can be determined

in order to calculate the uniformity coefficient and sorting coefficient.

Table 4. presents the dry sieve cumulative percentage of sand sample. 41.22 g of sand sample were poured onto

the sieve pan and sieved mechanically. The highest occurrence was on particle interval size of 45 µm to 62 µm with

23.19 % followed by particle interval size of 150 µm to 211 µm with percentage of occurrence of 21.62 %

respectively. The lowest percentage of occurrence was 2.62 % at particle interval size of 425 µm to 599 µm. The

percentage of fines for sand sample was 5.24 %.

76

FIGURE 3. Dry Sieve S–Curve of the sand sample.

Laser particle size analyzer (LPSA)

The result of LPSA data for the sand sample were tabulated and plotted to obtain the overview of the size

distribution as shown in Figure 4 (a). It was revealed that the sand particle size ranging from 0.96 µm to 630.00 µm

and the trend of the particle size distribution graph for the sand sample was a single peak graph. It was observed that

the particle size concentration gradually increased from 0.96 µm and peak at 158.49 µm with 9.93 volume %. The

sand particle distribution starts to decrease to 0.71 volume % at 630.00 µm.

Figure 4 (b) represents the S–Curve of the sand sample. The curve can be plotted where the cumulative weight

(%) in the y-axis constructed against the sand grain diameter (µm) in the x-axis. The scale of the x-axis is

logarithmic scale while the y-axis has a scale of percent (0 to 100%) using linear scale. From the S-Curve graph, the

D10, D40, D50, D90 and D95 values of the sand sample can be determined. It was observed that the S-Curve for the

LPSA analysis was smoother compared to the S-Curve graph from dry sieve analysis. It was due to the LPSA

analysis can detect more data compared to dry sieve analysis, thus resulting in more accurate results representation.

(a)

(b)

FIGURE 4. (a) Particle size distribution for sand sample; (b) LPSA S–Curve of the sand sample.

77

Uniformity Coefficient, Sorting Coefficient, Percentage of Fines

Table 5. represents the criteria D for dry sieve analysis and laser particle size analyzer for sand sample. The D10,

D40, D50, D90 and D95 values for dry sieve analysis were 28.00 µm, 120.00 µm, 150.00 µm, 590.00 µm and 620.00

µm while as for the laser particle size analyzer, the D10, D40, D50, D90 and D95 values were 20.00 µm, 102.00 µm,

120.00 µm, 220.00 µm and 280.00 µm respectively. The percentages of fines for dry sieve analysis and laser particle

size analyzer were 5.24 % and 13.34 % respectively. The percentage of fines for laser particle size analyzer samples

were higher compared to the dry sieve analysis.

TABLE 5. Criteria D for the sand samples.

Criteria D

Sand Samples

Dry Sieve Analysis Laser Particle Sand Analyzer

Sand Grain Diameter (µm)

D10 28.00 20.00

D40 120.00 102.00

D50 150.00 120.00

D90 590.00 220.00

D95 620.00 280.00

Percentage of Fines, (< 45 µm), % 5.24 13.34

Table 6. shows the uniformity coefficient (Uc), sorting coefficient (Sc) and its classification for sand sample. It

was found that the Uc and Sc for sand sample values were 0.20 and 0.05 for dry sieve analysis and 0.46 and 0.07 for

laser particle size analyzer. The Uc and Sc for both analyses can be classified as highly uniform and well sorted sand.

In addition, the percentage of fines for dry sieve analysis and LPSA were 5.24 % and 13.34 % respectively. It was

mainly due to the nature of LPSA that able to measure more accurate measurement of the sand particle size. Since

all the samples were well sorted and highly uniform sand, the samples may have greater porosity [7]. The porosity is

greater due to the lack of cementation in the formation and may cause sand production to occur. Hence, according to

the sorting and uniformity coefficient, all samples have higher probability of having sand production.

TABLE 6. Uniformity coefficient, sorting coefficient and classification.

Sample Parameter Uniformity

Coefficient, Uc

Sorting

Coefficient, Sc

Classification

Sand Sample

Dry Sieve

Analysis

0.20 0.05 Highly uniform and

well sorted sand

Laser Particle

Sand Analyzer

0.46 0.07 Highly uniform and

well sorted sand

Grain Size Specification

Figure 5. shows the sand particle image of the sand sample captured using FESEM at 100 times magnification.

Five random sand particles were selected and listed in Table 7. The diameters selected were ranging from 107.00

µm to 225.00 µm. The Krumbein phi scale, Ø was ranging from 2.15 to 3.22 and in overall, the sand sample can be

classified as fine sand to very fine sand.

TABLE 7. Sand size and classification for the sand sample.

Grain Diameter of Grain

(µm)

Diameter of Grain

(mm)

Ø Classification

Particle 1 144.90 0.145 2.79

Fine sand to very

fine sand

Particle 2 107.30 0.107 3.22

Particle 3 126.10 0.126 2.99

Particle 4 224.70 0.225 2.15

Particle 5 132.40 0.132 2.92

78

FIGURE 5. Random grain size measurement for the sand sample.

Roundness and Sphericity

The geometrical natures of the sand sample were evaluated in this section. Figure 6. shows the picture of the

sand grain sample captured using FESEM at 100 times magnification. Ten random particles were selected in the

field of view, and then their roundness and sphericity rating were evaluated using the chart for visual estimation of

roundness and sphericity [8]. The average roundness and sphericity were calculated and listed in Table 8. The

average roundness and sphericity was used to classify the samples and the results listed in Table 9.

The sand sample can be classified as sub rounded and very low sphericity with average roundness and sphericity

of 0.37 and 0.38 respectively. According to Ibrahim 2013, the increase in angularity of sand grains will increase its

porosity [7]. The unconsolidated sands relatively consist of fine sand and greater porosity due to the less compaction

and little burial than consolidated sands. Thus, the higher porosity has higher probability of having sand production.

Furthermore, the irregularly-shaped sand grains have higher value of porosity than the spherically-shaped sand

grains.

FIGURE 6. FESEM view of the sand sample.

TABLE 8. Averaging method for the

sand grains.

Particle X=

Roundness

Y =

Sphericity

1 0.30 0.40

2 0.30 0.30

3 0.40 0.20

4 0.40 0.30

5 0.40 0.50

6 0.40 0.50

7 0.40 0.50

8 0.40 0.50

9 0.40 0.20

10 0.30 0.40

Average 0.40 0.38

79

TABLE 9. The roundness and sphericity classification of the sand sample.

Sample X= Roundness Y= Sphericity Classification

Sand sample 0.37 0.38 Sub rounded and very

low sphericity

CONCLUSION

The particle grain analysis of sand samples from North Malay Basin was discussed in this study. In summary,

the dry sieve analysis and LPSA were performed and the S-Curve graphs were plotted. The D10, D40, D50, D90 and

D95 values were determined and compared. It was found out the values for LPSA were slightly smaller compared to

the dry sieve analysis. In contrast to that, the percentage of fines in LPSA was higher compared to dry sieve. In

overall, the sand sample can be classified as highly uniform and well sorted sand. In terms of roundness and

sphericity, the sand sample can be classified as sub rounded with very low sphericity sand. Finally, the sand sample

grain size can be classified as fine sand to very fine sand respectively.

ACKNOWLEDGEMENT

The author would like to thank the Malaysia-Thailand Joint Authority (MTJA) Research Cess Fund for the financial

support and Universiti Teknologi PETRONAS (UTP) for the research facilities.

REFERENCES

[1] Fattahpour, V., Moosavi, M., & Mehranpour, M. (2012) An experimental investigation on the effect of rock

strength and perforation size on sand production

[2] Wu, B., & Tan, C.P., (2005) Effect of Water Cut on Sand Production – An Experimental Study.

[3] Hess Corporation. North Malay Basin. (2017) Retrieved from

http://www.hess.com/operations/offshore/north-malay-basin.

[4] Carney, S., Abd Aziz, I., Martins. W., & Kennedy. J. Hess Oil & Gas Sdn. Bhd. & CH Mutiara Petroleum

(CHMP) (2010). Reservoir Characterisation of the Mio-Pliocene Reservoirs of PM301 in the North

Malay Basin.

[5] Adeyanju, O. A., & Olafuyi, O. A. (2011) Experimental Studies of Sand Production From Unconsolidated

Sandstone Petroleum Reservoirs in Niger-Delta.

[6] Sengupta, S. (1994) Introduction to Sedimentology. A.A Balkema Publishers.

[7] Ibrahim, B. F. E. (2013) Porosity and Permeability. Retrieved from Suez University Faculty of Petroleum &

Mining Engineering, https://www.slideshare.net/belalelnagar3/porosity-and-permeability.

[8] American Petroleum Institute. (2016) Measurement of Properties of Proppants Used in Hydraulic Fracturing

and Gravel-packing Operations.

[9] Stojanović, Z., Marković, S., & Uskoković, D. (2012). Determination of particle size distributions by laser

diffraction.

[10] Pan, L., Ge, B., & Zhang, F. (2017) Indetermination of particle sizing by laser diffraction in the anomalous

size ranges.

[11] Krumbein, W. C.; Aberdeen, Esther (April 1937). "The Sediments of Barataria Bay". Journal of Sedimentary

Petrology.

[12] Powers, M. C. (1953). “A new roundness scale for sedimentary particles.” J. Sediment. Res.

[13] Riley, N.A. (1941) Projection sphericity. J. Sed. Petrol., 11, 94–97.

80

Structure and Benefits of the Multi-Agent Resource

Conversion Process Architecture

Konstantin Aksyonov a), Eugene Bykov, Olga Aksyonova,

Natalia Goncharova, and Irina Pelymskaya

Department of Information Technology

Ural Federal University

Ekaterinburg, Russia a)[email protected]

Abstract. In this paper we discuss the application of the multi-agent approach to the analysis systems. We also present

the modeling of the land development. We provide the details of the multi-agent simulation modeling platform by

integrating the artificial intelligence together with distributed computations and simulation modeling. In this paper we

present the overview of development and application of the multi-agent approach to the resource conversion processes.

The land development planning problems composed the sample application for the analysis and modeling. We cover the

five classed of processes within the scope of analysis and modeling the multi-agent resource conversion processes. They

include: industrial, technological, logistical, social-economic, and finally the business processes.

Keywords—multi-agent simulation; resource conversion process architecture

INTRODUCTION

Russian and international scientist have made a significant contribution into the development of ideas for modeling of the resource conversion processes. These include N. Buslenko [1], A. Vavilov with B. Fomin [2 3], B. Sovetov [4], A. Pritzker [5], G. Forrester [6-7], A. Sheer [8, 9] and others [10-11].

To proceed, here is what we would mean under the resource conversion process in the scope of our paper, as well as our development in general. This is the process of converting the input resources, required for the process operation, into the output products of such conversion. The discrete resource conversion process is a process, featuring state changes in only the particular moments of the timeline. We also consider the discrete equivalents of the continuous processes [5]. The transformation is achieved by the discrete sampling of the process variables over the timeline. The resource conversion process may be represented in form of a graphical structure. In the general case it would contain the following elements, presented on Figure 1. Among those we mention the inputs and outputs in the beginning and in the end, as well as the start conditions, conversion operations and conversion tools in the middle. The resource condition determines at which point of time the resource conversion process operation begins. It is based on the conversion process state, which is in turn affected by the input and output resources, mechanisms, and the other events of the external environment. Process execution time is calculate at the beginning of the conversion.

ARCHITECTURE OF THE RESOURCE CONVERSION PROCESSES

In general, the resource conversion process consumes the inputs and produces the output, which is modeled by increasing and decreasing of the output and input amounts respectively. The initialization of the launch condition means the consumption of the input resources, and the seizure of the tools, which are engaged into the transformation. When the conversion completes, the value modifications for the input and output resources are committed, and the tools are released for further use by other operations. The following areas of the problem domain of the resource conversion processes are covered:

• estimation of the resource consumption and the dynamics of the tool operation;

• estimation of the overall cost and duration of the processes;

81

• definition of the new designs of the resource conversion processes;

• prognostication of state of the resources and tools in the particular moments;

• enhancement and enrichment of existing resource conversion processes.

Startcondition

Input Output

Tools

Conversion

Figure 1. Resource conversion process

Looking at the different perspectives of the resource use [12], the resource conversion process can be classified as presented on the Figure 2.

Non-deterioratingDeteriorating IndirectDirect

(substantially consumed)

Products Waste

OutputsToolsInputs

(consumed)

Resources

Figure 2. Resource classification by their use types

The consumed resources (inputs) are the resources that are used only once in the process. Depending on their role in the resource transformation process, the consumed resources can be subdivided into direct, which directly represent the part of the end product and form part of it, and indirectly, which are only part of the end product but participate in the resource conversion process.

Tools are not consumed, but they are used during the conversion process. Their amount does not decrease during use. Depending on its potential use, it can be used multiple times in most cases. Depending on whether their potential use decreases over time, the tools can be subdivided into worsening and not worsening.

As a result of the conversion processes, outputs are generated. The outputs can be separated into products and waste.

From a physical perspective, resources can be classified by the following types:

• Material resource that can represent materials, spare parts, product units, hardware;

• information resources (information, documents);

• financial resources;

• energy resources, including all types of energy and fuels;

• work resources. We should also keep in mind that the amount of resources decreases when it comes to inputs in terms of material,

financial and energy resources. But there are two specific aspects to the information resources.

82

• There are such information conversion processes that do not lead to the decrease of the input resource (file copy, training), in this case only the resource that processes the conversion decreases (CPU time, professor time and effort).

• In order to ensure the equivalence of the information resource with the other resource types from the point of view of reducing their quantity to the input, we present the concept of a transact. In represents a task to perform certain actions with a message (or a permission to execute a job).

Within a single resource transformation process, we can transform resources of various types: a resource can be captured (as input), generated (as output), or used (as a tool). An example of the use of different resource types is the freight delivery process (Figure 3).

ResourcesInputs Outputs

Material (cargo at a warehouse)Financial (receiving funds on account)Information (invoice, other documents)Energy (fuel)

Material (delivered cargo)Financial (Salary, taxation)

Information (signatures in documents)Energy (remaining fuel)Tools (truck)

Labor (driver, loader) Figure 3. Cargo delivery process

Every industrial enterprise is an example of a resource transformation process. An industrial process is defined as a process when certain goods, including material, non-material or both, are transformed into other goods, material, non-material or both [12]. Goods are resources here. Thus, an industrial process is a process of transforming some goods into others, fully in line with the definition of a resource transformation process.

Resources warehouse

Production

Finished products

warehouseSales

Supply Invoicing

Personnel

Machinery

Wear and tear

Salary

TaxationPayments

Figure 4. Enterprise flow model

A higher-level node of the general graphical representation of a business activity can be defined as in Fig. 5. Here, the rectangles contain the company resources and the ellipses the resource converters.

83

2. Authorities

9. Enterprise

customers

10. Advertised

markets

Enterprise

1. Enterprise branches

3. Managed

enterprises

4. Federal budget and

funds

8. Suppliers of goods and

services 5. Labor

7. Suppliers, services, goods markets 6. Labor markets

Figure 5. Interaction of an enterprise with the external environment

The more detailed classification of the resource-based interaction of a company with the external environment is shown in Fig. 6 [13]. Here, arrows represent the possible direction of resource flow between the enterprise and the external environment without specifying a particular resource type.

The processes of the interaction of the company resources with the external environment define the contents of the internal processes in the enterprise. To comprehensively examine and analyze the resource transformation processes, we use the models of the internal processes obtained by decomposing the external processes. This creates the hierarchical multi-level process model. At the lowest levels, the process can be represented by the precise elemental resource conversion operations.

External environment

(convertors, agents, resources,

tools, parameters, goals)

Simulation model

Planning subsystemReactive subsystem

External environment interface

Sub-system of cooperation with other agents

Inner behaviour

(Activity diagram)

Logical output

from strategic

knowledge base

(Decision search

diagram)

Strategic knowledge

base

(frames)

Tactical knowledge

base

(productions)

Logical output from tactical

knowledge base

(Direct output)

Figure 6. Multi-agent resource conversion process model structure

THE AGENT MODEL

For the agent model, we recommend using the process model to transform resources from multiple agents. This model is successfully used to model and control technology, logistics and business processes. The potential developments of this model have been successfully used for research on social, economic and enterprise systems.

84

The multi-agent model of resource transformation processes represents the integration of simulation, expert, situational and multi-agent modeling. The dynamic model of the multi-agent resource transformation process is based on the hybrid multi-agent architecture InteRRaP.

Multi-Agent Resource Transformation Process Agent can be hybrid in nature and contain two components:

• Intelligent (production rules and / or frame-based access to expert systems)

• Reactive (the agent activity is defined in the UML action diagram). In accordance with the common concept of the InteRRaP architecture, the multi-agent RCP agent model is

presented in four levels.

CONCLUSION

We use several resource conversion process models: 1. Active and passive converter model used for production planning, 2. Hierarchical converter system model, based on the system dynamic approach and its application in the socio-economic development system of the community, 3. Strategic project method , 4. Agent-based system-dynamic approach to community modeling and decision-support system implementation based on the AnyLogic multi-approach modeling system 5. Information support model for regionally open decentralized innovative structures, 6th approach to the implementation of the monitoring System for the floating budget funds in Sverdlovsk region, based on the dynamic real-time expert system G2 (GIS, SM, Neuron-online, Tele-Windows (Distribution)), and 7. Process model for the conversion of multiple agent resources and its implementation into the BPsim software suite and the automated metallurgical production system as described above. Due to many limitations of the other systems, the multi-agent resource conversion process architecture has become one of the foundations of the software tools used in many areas of our region.

ACKNOWLEDGMENT

This work is supported by Act 211 Government of the Russian Federation, contract № 02.A03.21.0006.

REFERENCES

[1] Buslenko N. P., Modeling complex systems, Moscow, Science, 1978

[2] Simulation modeling of industrial systems, edited by A. A. Vavilov, Berlin: Techniques, 1988

[3] Avramchuk, E. F., A. A. Vavilov, and S. V. Emelianov, Technology of system simulation, M.: machine construction industry, 1988

[4] Sovetov B. Y. & Yakovlev S. A. Systems modelling, Moscow, High School, 2001

[5] Pritsker, A. A. B.. Introduction to simulation and SLAM II. System Publishing Corporation, West Lafayette, 1984

[6] Forrester J., Industrial Dynamics, Cambridge, MA: MIT Press, 1961.

[7] Forrester J., World Dynamics, Productivity Press, 2nd edition, 1979

[8] Sheer A. V., Business processes: main concepts, theory, methods, Moscow, 1999.

[9] Sheer A. V., Modeling business processes, Moscow, 2000.

[10] Hammer M. Reengineering the Corporation: A Manifesto for Business Revolutions / M.Hammer, J.Champy. HarperBusiness, 1993

[11] Newell, “Production systems: models of control structures // Visual information processing”, New York: Academic Press, 1973, pp. 463-526.

[12] Pischulov V., Introduction to the production theory, Ural Economic University, Ekaterinburg, 2003

[13] Klebanov B. I., Methodology for developing the corporate information system, Ural State Technical University, Ekaterinburg, 1999

85

Kinetics of Anaerobic Digestion of Chicken Manure Co-

Digested with Wastewater from Thai-noodle Factory: The

Effect of Dilution and Ash Supplement

Rachadaporn Thongnan 1), Chairat Siripatana 1, 2), Sunwanee Jijai 3) and Nirattisai

Rakmak 1, 2, a)

1School of Engineering and Resources, Walailak University, Nakhon Si Thammarat, Thailand. 2Biomass and Oil-Palm Excellence Center,Walailak University, Nakhon Si Thammarat Thailand.

3Faculty of Science Technology and Agriculture, Yala Rajabhat University, Yala, Thailand.

a) Corresponding author: [email protected]

Abstract. Thai rice noodle is normally produced in local small factories or household production. Although the

wastewater produced from these factories are localized and distributed it causes considerable environmental problems.

This work attempt to optimize the methane yield obtained from anaerobic digestion (AD) of Thai rice noodle wastewater

(TRW). It is well established that TRW is not suitable for single AD because of nutrient imbalance and very low pH (~3).

Thus, in this study TRW was co-digested with chicken manure (CM) and supplemented with biomass ash obtained from

biomass power plants. The results showed that the best ratio of CM, TRW and ashs (A) was 30g:200ml:4g which gave

the BMP of 309.7 ml CH4/gVSadded. This work has also found that for single AD of CM, the best dilution was 1 g CM/6

ml of water (W) which gave the BMP of 241.2 ml CH4/gVSadded. This dilution is a good agreement with the CM:TRW

ratio of 30g:200ml (equivalent to 1:6.6 dilution), thus confirming that the best yield for CM:TRW ratio of 30g:200ml

was due to minimum substrate inhibition while no nutrient deficit played the role. It has also found that Gompertz two

substrate (GTS) and Monod two substrate (MTS) models was able to describe the methane evolution (ME) data of AD

co-digestion very well although we did not attempt to interpret their parameters to get more insight in this article.

Keywords—biogas; TRW; ash supplement; AD co-digestion; GTS and MTS models; dilution

INTRODUCTION

One of global trend in renewable energy is the expansion of distributed biogas produced through AD based on

variety of organic waste/wastewater released from industries and communities [14,16]. In Thailand and other

tropical regions, biogas can be produced from a wide range of organic wastes including, municipal liquid and solid

wastes, animal waste, agricultural waste, industrial wastewater, sewage sludge and landfill waste. Organic waste

composition is complex, each waste has different nutrients, inhibitors and other peculiarities which requires specific

tests (eg. BMP test) in order to evaluate its potential for biomethane production [2]. Traditional Thai rice noodle is

very popular in Thailand and nearby countries. The noodle is normally produced in local small factories or

household production, so the wastewater produced from these factories are localized and distributed but cause

considerable environmental problems if added them up or supplementing to already large amount of municipal and

industrial wastewater. As the environmental awareness of Thai community increases as well as more strict

regulation is in place, Thai rice noodle producers are forced to act accordingly. There are at least two alternatives to

treat TRW, aerobic treatment and AD [7]. AD is preferred because it greatly reduce stringent odor, very low energy

usage and generate methane as a renewable energy. However, TRW is not suitable for single digestion because of its

low pH, imbalance nutrient containing mainly carbohydrate and protein (and ammonia) [15]. So, effective AD for

TRW require co-substrate and nutrient supplements.

Study on biogas production from TRW has been very limited. One of pioneer work on using TRW as a substrate

was due to Reference [7] who evaluated the BMP of TRW co-digested in 200-digesters with 5 levels of CM

supplement. They found that the optimal CM supplement was 30 g where the initial pH and the ratio of VFA/ALK

was 7.5 and 0.073 respectively. The methane composition in biogas was 50.66 and the BMP was 299 ml

86

CH4/gCODadded. However, COD/N ratio of the co-substrates were 16-20 which is not in the optimal range 71.4-85.7

[1,3,8] and the author also suspected that there existed the inhibitory effect due to high ammonia content in CM.

This research is a follow-up of the work of Reference [3]. Here we investigated the inhibitory effect of CM and

the enhancing effect by biomass ash supplement (AS). In the first one, we want to clarify the inhibitory effect of the

CM concentration so that and to avoid making assumption about it. The later investigation is the attempt to enhance

methane production by adding biomass ash (bottom ash from biomass power plants) to adjust the initial pH and to

the supply mineral nutrients to the involved microorganisms.

MATERIAL AND METHODS

The TRW sample and CM were collected from the community in Yala province and the local layer chicken

farm, respectively. Characteristics of TRW and CM are the same as tabulated in Reference [7]. The TRW and CM

were kept at 0-4 oC until being used in the experiments.

Experimental Set-Up

The experiments were conducted at room temperature (28-30 ºC) until batch completion. The 300-ml-volume

serum bottles were used as reactors and a working volume of 200 ml was used in all experiments. The serum bottles

were covered with the rubber stoppers and sealed with aluminum caps. The volume of biogas was measured daily by

using water displacement method [1,3,4]. The methane content was measured using Gas Chromatography (GC-8A

Shimadzu). The experiments were duplicated in all experiments. The experiment setup is shown in Fig. 1.

FIGURE 1. Schematic view of the experimental set-up

Experiment 1: BMP tests for CM at different dilutions

All experiments were operated in batch reactors with a total working volume of 200 ml. Each reactor contains

different of ratio of CM and water (W) (1:2, 1:4, 1:6, 1:8 and 1:10). After preliminary tests, only three ratios (1:2,

1:4 and 1:6) were chosen for repeating them replicate. Other ratios were dropped out because of their low BMP.

The variables designed in this study, some results from chemical analysis were shown in Table 1.

TABLE 1. Characteristic of raw material (CM and W)

CM: W

(g: ml)

pH (mg/l) COD (mg/l) VS (mg/l) TS (mg/l) P∞

ml CH4 /

g VS before after before after before after before after

1:2 (Triplicate) 7.95 7.16 15550 5200 37420 26888 84770 67300 1609 159.25

1:4 (Triplicate) 7.83 7.31 7775 2600 18710 11780 42385 17780 1022 202.38

1:6 (Triplicate) 7.82 7.25 5183 1733 12473 8962 28256 15260 812 241.2

1:8 7.80 7.20 3887 - 9355 - - - 295 116.7

1:10 7.79 7.19 3110 - 7484 - - - 117 58.07

87

Experiment 2: BMP tests for different levels of ash supplement

BMP tests were operated in batch reactors with a total working volume of 200 ml. Each reactor contains

different levels of AS (1, 2, 3, 4, and 5 g) and fix amount of CM and TRW (30g and 200 ml, respectively). All

experiments were carried out in a triplicate manner. The variables designed in this study, some results from chemical

analysis were shown in Table 2.

TABLE 2. Characteristic of raw material (CM, TRW and A)

Analytical Procedures

In all experiments, we analyzed pH, Chemical Oxygen Demand (COD), Total Solids (TS), and Volatile Solids

(VS). All analytical procedures are performed in accordance with standard methods for the examination of water and

wastewater APHA [5]. The BMP can calculate by maximum cumulative methane divided by gCODadded [7].

Kinetic Model of Biogas Production

The GTS Model

Reference [10] used Gompertz postulation and rewrote the specific growth rate as a time function of the GTS

model as follow:

( ) ( )( ) ( )11tan

2

−

− −

' ' '

e e0 e s e s0 s

πP = P + P θ + κ t τ + P P +P θ

π (1)

( )exp exp 1

− −

mee e'

e

R eθ = λ t +

P (2)

( )exp exp 1

− −

mss s'

s

R eθ = λ t +

P (3)

Where θe and θs are fractional conversions of easily and slowly digestible substrates respectively and ' ' ' ' ' '

0 e e e0 s s s0P = P+P , P = P +P , P = P +P

The MTS Model

Reference [13] used Monod kinetic model postulation and rewrote the switching or preference function g(t, Se,

Si, …) which is introduced into the model to describe how each group of microbes deals with multiple substrates

according to its preference. The ODEs as follow:

= −ss s ds s

dXX k X

dt (4)

CM: TRW: A

(g:ml:g)

pH (mg/l) COD (mg/l) VS (mg/l) TS (mg/l) P∞

ml CH4 /

g VS before after before after before after before after

30:200:1 6.424 3.969 17145.6 14652.8 11,280 9,640 57,664 49,080 400 177.30

30:200:2 6.828 4.014 17145.6 16347.6 11,280 10,755 57,664 46,920 414 183.51

30:200:3 7.479 4.285 17145.6 14166.4 11,280 9,320 57,664 39,000 481 213.20

30:200:4 7.634 4.883 24684.8 19699.2 16.240 12,960 70,960 42,120 1006 309.73

30:200:5 8.15 4.98 25110.4 17996.8 16,520 11,840 67,640 37,840 996 301.45

88

( )[ , , ]= + −ee e i i de e

dXg t S S k X

dt (5)

( )= − + +s ss SsX ds s de e

XsSs

dSX f k X k X

dt Y

(6)

= −e ee

XeSe

dSX

dt Y

(7)

( ), ,1 −

= −

e ii XsSsis s s i e

XsSs XeSi

g t S SdS Yf X X

dt Y Y (8)

( ), ,= +PSe PSie e e i i e

XeSe XeSi

Y YdPX g t S S X

dt Y Y (9)

=+

me ee

Se e

S

K S

(10)

=+

ms ss

Ss s

S

K S

(11)

=+

mi ii

Si i

S

K S

(12)

Where kde and kds are the specific death rate of dXs/dt = µsXs – kdsXs and dSs/dt = -(µs/YXsSs)Xs + fSsX(kdsXs +

kdeXe) respectively.

YXsSs, YXeSe, YXeSi, YPSe and YPSi are the corresponding yield coefficient as specified by the subscripts.

fSsX and fis are conversion factors for X → Ss and Ss → Si respectively.

This case we will only show the switching or preference function g(t, P) to fit the current data.

( ) ( )( )11tan

2

− = − +

cg t t t

(13)

( ) ( )( )11tan ( )

2

− = − +

c cg P P P

(14)

( )= −PS e ecP Y S S (15)

RESULTS AND DISCUSSION

Traditional modified Gompertz equation and single-substrate Monod model were tried to represent the data but

fittings were not very satisfactory, presumably because the substrates (both TRW and CM) contain large portion of

slowly degradable organic matter. In other words, all ME curves in this study fall into type-II according to

Reference [10,13] which can be interpreted as complex substrate containing significant amount of slowly degradable

component.

Based on two categories of organic components (easily and slowly degradable), GTS and MTS models

[9,11,15] (abbreviated as GTS and MTS models respectively) were used to represent the ME data satisfactory. Both

GTS and MTS fit ME data equally well and all parameters of the models were summarized in table 3 and 4. Both

89

models predicted that the portion of slowly degradable organic matter of the co-substrate was approximately 0.5-0.8

for CM and 0.4-0.8 for the co-substrate mixture.

TABLE 3. The experimental conditions and parameters estimated from kinetic model of different CM and W ratio

Model Parameter CM: W ratio

1:2 1:4 1:6 1:8 1:10 P ( mL ) 1609.14 1022.36 812.31 294.86 117.42

eP ( mL ) 524.57 169.82 288.80 127.68 0.00

1 ( / ) = −Ss ef P P 0.67 0.83 0.64 0.57 1.00

msR ( /mL d ) 67.41 144.28 31.79 43.60 6.34

Gompertz two

substrate meR ( /mL d ) 47.82 22.49 26.08 22.49 1.00

(no unit) 4.5524 0.1316 0.8081 0.0963 0.4806

e ( 1d − ) 13.3171 14.5341 13.1904 6.4313 3.7604

s ( 1d − ) 0 0 0 0 0

rt 1.0471 11.2822 1.8129 15.7169 15.4404 2R 0.9992 0.9990 0.9998 0.9898 0.9902

Monod two

substrate

me ( 1d − ) 0.1 0.22 0.22 0.23 0.4

sef (no unit) 1 0.8 0.8 0.5 0.5

ms ( 1d − ) 0.1 0.176 0.176 0.115 0.2

ief (no unit) 1 1 1 1 1

mi ( 1d − ) 0.1 0.22 0.22 0.23 0.4

dek ( 1d − ) 0.0429 0.0429 0.0429 0.0429 0.0429

dsk ( 1d − )

0.01 0.018 0.018 0.012 0.02

SsXf (no unit) 0.7 0.7 0.7 0.7 0.7

isf (no unit) 1 1 1 1 1

SeK ( /mg L) 1900 2000 2000 1300 1200

SiK ( /mg L ) 3000 2200 2200 2200 1250

SsK ( /mg L) 3000 2200 2700 6500 1300

Model Parameter CM: W ratio

1:2 1:4 1:6 1:8 1:10

Monod two

Substrate

(continuous)

XeSeY ( /mL ( /mg L )) 0.1523 0.1523 0.1523 0.1523 0.1523

XeSiY ( /mL ( /mg L )) 0.1523 0.1523 0.1523 0.1523 0.1523

XeSsY ( /mL ( /mg L )) 0.24 0.215 0.215 0.215 0.215

PSeY ( /mL ( /mg L )) 0.111 0.121 0.14 0.104 0.06

PSiY ( /mL ( /mg L )) 0.104 0.13 0.225 0.225 0.08

SeSsY ( /mL ( /mg L )) 1 1 1 1 1

0S ( /mg L ) 15550 7775 5183 3888 3110

Ssf (no unit) 0.4 0.4 0.4 0.4 0.4

0sS ( /mgVS L ) 6220 3110 2073 1555 1244

0eS ( /mgVS L ) 9330 4665 3110 2333 1866

0iS ( /mgVS L ) 0 0 0 0 0

0X ( /mgVS L ) 2000 400 400 170 10

Xsf (no unit) 0.5 0.5 0.5 0.5 0.5

90

0sX ( /mgVS L ) 1000 200 200 85 5

0eX ( /mgVS L ) 1000 200 200 85 5

0P ( mL ) 0 0 0 0 0

(no unit) 0.1 0.1 0.1 100 0.1

cf (no unit) 0.91 0.91 0.91 0.98 0.98

cP ( mL ) 942 514 396 238 110 2R

0.9991 0.9966 0.9983 0.9909 0.9748

TABLE 4. The experimental conditions and parameters estimated from kinetic model of different amount of CM: TRW: A

Model Parameter CM: TRW: A ratio (g: ml: g)

30:200:1 30:200:2 30:200:3 30:200:4 30:200:5 P ( mL ) 400.03 414.82 481.37 1006.12 995.97

eP ( mL ) 133.05 138.54 119.17 149.39 195.41

1 ( / ) = −Ss ef P P 0.67 0.67 0.75 0.85 0.80

msR ( /mL d ) 170.81 294.62 320.89 674.12 723.55

Gompertz two

substrate meR ( /mL d ) 6.04 3.24 6.86 509.38 83.48

(no unit) 65.3695 12.2412 4.9202 0.1973 0.1761

e ( 1d − ) 13.6808 18.1327 16.3286 5.8666 5.2774

s ( 1d − ) 0 0 0 0 0

rt 0.1000 0.8299 0.6588 2.7807 2.8043 2R 0.9985 0.9991 0.9984 0.9964 0.9939

Monod two

substrate

me ( 1d − ) 0.4 0.49 0.49 0.35 0.4

sef (no unit) 0.15 0.035 0.03 0.28 0.28

ms ( 1d − ) 0.06 0.017 0.015 0.098 0.112

ief (no unit) 0.9 0.9 1.1 1 0.9

mi ( 1d − ) 0.36 0.441 0.539 0.35 0.36

Model Parameter CM: TRW: A ratio (g: ml: g)

30:200:1 30:200:2 30:200:3 30:200:4 30:200:5

Monod two

Substrate

(continuous)

dek ( 1d − ) 0.0429 0.0429 0.0429 0.0429 0.0429

dsk ( 1d − ) 0.006 0.002 0.001 0.01 0.011

SsXf (no unit) 0.7 0.7 0.7 0.7 0.7

isf (no unit) 1 1 1 1 1

SeK ( /mg L) 3000 3000 3000 3500 3000

SiK ( /mg L) 1200 1200 1200 1200 1200

SsK ( /mg L) 1000 1000 1000 1000 1000

XeSeY ( /mL ( /mg L )) 0.1523 0.1523 0.1523 0.1523 0.1523

XeSiY ( /mL ( /mg L )) 0.1523 0.1523 0.1523 0.1523 0.1523

XeSsY ( /mL ( /mg L )) 0.215 0.215 0.215 0.215 0.215

PSeY ( /mL ( /mg L )) 0.033 0.035 0.045 0.038 0.038

PSiY ( /mL ( /mg L )) 0.033 0.035 0.045 0.038 0.038

SeSsY ( /mL ( /mg L )) 1 1 1 1 1

0S ( /mg L ) 17145.6 17145.6 17145.6 24684.8 25110.4

Ssf (no unit) 0.55 0.55 0.55 0.45 0.55

0sS ( /mgVS L ) 9430 9430 9430 11108 13811

91

0eS ( /mgVS L ) 7716 7716 7716 13577 11300

0iS ( /mgVS L ) 0 0 0 0 0

0X ( /mgVS L ) 4000 4000 4000 4000 4000

Xsf (no unit) 0.5 0.5 0.5 0.5 0.5

0sX ( /mgVS L ) 2000 2000 2000 2000 2000

0eX ( /mgVS L ) 2000 2000 2000 2000 2000

0P ( mL ) 0 0 0 0 0

(no unit) 0.1 0.1 0.1 0.1 0.1

cf (no unit) 1.7 1.7 1.47 1.12 1.14

cP ( mL ) 433 459 510 578 490 2R

0.9619 0.9629 0.9853 0.9866 0.9891

The effect of dilution on single digestion of CM

CM is rich in nutrients and (possibly) inhibitors if present in high concentration. The most obvious one is

ammonia, both as ammonium salt and as free ammonia. It is well-known that free ammonia is very toxic to

methanogens in AD process. Fig. 2 shows the ME curves of CM at different dilutions of which the BMP at different

dilutions is shown in Fig. 3. As the CM:W ratio decreased from 1:2 to 1:6, the BMP increased from 159 to 241

mlCH4 / gVSadded. However, if the ratio decreases further (1:8 and 1:10) the BMP started to decrease gradually.

Presumably, the increase of BMP as the ratio increased from 1:2 to 1:6 was due to the decrease of free ammonia (or

other inhibitors) as CM was diluted. However, whereas the decrease in BMP as the ratio increased from 1:8 to 1:10

was likely due to nutrient deficiency although it was not clear what nutrients were limited. Regardless of the

mechanism behind the effect of CM dilution. It is clearly that the optimal CM:W ratio was around the neighborhood

of 1:6.

1000

800

600

400

200

00 5 10 15 20 25 30Time (Days)

1:2

1:4

1:6

1:81:10

1200

1400

1600

Accu

mulated

Bioga

s (ml)

(a) (b)

FIGURE 2. Accumulated biogas of different CM:W ratios (a) GTS model (b) MTS model

FIGURE 3. The effect of CM:W ratio on methane yield in term of BMP

92

The effect of AS

In this set of experiments, 200 ml of TRW was mixed with 30 g of CM which approximately equivalent to the

CM dilution of 1:6. This TRW:CM ratio was also gave the best BMP in the work of Reference [7] thus is used for

all experiments in this set. The AS was in the range of 0-5 g. The resulting ME curves were shown in Fig. 4 and the

corresponding BMPs were presented in Fig. 5. It is clearly that in the range of 1-4 g AS, higher AS is associated

with higher BMP. However, at the level of 4 g adding more A did not improve BMP. Thus, the optimal value of AS

was around the neighborhood of 4-5 g AS.

1000

800

600

400

200

00 5 10 15 20 25 30

Acc

um

ula

ted

Bio

gas

(ml)

Time (Days)

30:200:4

30:200:5

30:200:3

30:200:230:200:1

(a) (b)

FIGURE 4. Accumulated biogas of different amount of CM: TRW: A ratio; (a) GTS model (b) MTS model

FIGURE 5. The effect of AS on methane yield in term of BMP

CONCLUSION

TRW is characterized by high carbohydrate and nitrogen sources in the form of protein and ammonia, and very

low pH (~3). Its nutrient deficit makes it unsuitable for single AD, thus require complementary co-substrate and

nutrient supplement, which are, in this case, CM and biomass ash respectively. Our early work Reference [7] found

that the best TRW:CM ratio was 200 ml: 30 g. This work confirms the same finding by showing that the best CM

dilution is 1g CM and 6 ml W. Which is equivalent to TRW:CM ratio of 200 ml: 30 g. Regarding biomass AS, it is

quite clear that optimal level of AS is 4 g. which corresponds to the best pH range (7.0-7.6) for methanogens. So, the

main reason for AS to improve the BMP is by establishing the initial pH in the optimal range. It should be stated

here that GTS and MTS models was able to describe the ME data of AD co-digestion very well although we did not

attempt to interpret their parameters to get more insight in this article. It is something should be explored in the

future.

ACKNOWLEDGMENTS

This work was supported by Walailak University Fund Contract number 07/2559.

93

REFERENCES

[1] M. Abdel-Hadi, “A simple apparatus for biogas quality determination,” Misr J Ag Eng, vol. 25, pp. 1055–1066,

2008.

[2] H. Bouallagui, B. Rachdi, H. Gannoun, and M. Hamdi, “Mesophilic and thermophilic anaerobic co-digestion of

abattoir wastewater and fruit and vegetable waste in anaerobic sequencing batch reactors,” Biodegradation, vol.

20, no. 3, p. 401, Nov. 2008.

[3] A. J. Cavaleiro, T. Ferreira, F. Pereira, G. Tommaso, and M. M. Alves, “Biochemical methane potential of raw

and pre-treated meat-processing wastes,” Bioresource Technology, vol. 129, pp. 519–525, 2013.

[4] S. Dechrugsa, D. Kantachote, and S. Chaiprapat, “Effects of inoculum to substrate ratio, substrate mix ratio and

inoculum source on batch co-digestion of grass and pig manure,” Bioresource Technology, vol. 146, pp. 101–

108, 2013.

[5] R. E.W. and B. R.B., Standard Methods for the Examination of Water and Wastewater, 23rd Edition. American

Public Health Association, American Water Works Association, Water Environment Federation, 2017.

[6] A. I. et al., “Defining the biomethane potential (BMP) of solid organic wastes and energy crops: a proposed

protocol for batch assays,” Water Science & Technology, vol. 59, no. 5, Mar. 2009.

[7] S. Jijai and C. Siripatana, “Kinetic Model of Biogas Production from Co-digestion of Thai Rice Noodle

Wastewater (Khanomjeen) with Chicken Manure,” Energy Procedia, vol. 138, pp. 386–392, 2017.

[8] T. Kaosol and N. Sohgrathok, “Enhancement of biogas production potential for anaerobic co-digestion of

wastewater using decanter cake,” American Journal of Agricultural and Biological Science, vol. 7, pp. 494–502,

2012.

[9] J. MONOD, “The Growth of Bacterial Cultures,” vol. 3, pp. 371–394, 2003.

[10] L. Noynoo, N. Rakmak, C. Siripatana, S. Jijai, and K. Phayunphan, “Gompertz-Type Two-Substrate Models for

Batch Anaerobic Co-Digestion,” submitted for publication.

[11] T. Rachadaporn, H. Thongpan, N. Rakmak, and C. Siripatana, “Modeling of anaerobic co-digestion of pig

manure and domestic organic waste,” Jurnal Teknologi, vol. 78, 2016.

[12] A. Raj, S. Vinaykumar, H. Manjunath, A. Srinidhi, and J. Patil, “Biomethanation of Water Hyacinth, Poultry

Litter, Cow Manure and Primary Sludge: A Comparative Analysis,” Research Journal of Chemical Sciences,

Vol. 1, Issue 7, 2011, pp 22-26, ISSN 2231-606X (IF: 0.3725)., vol. 1, pp. 2231–606, 2011.

[13] N. Rakmak, C. Siripatana, S. Jijai, and L. Noynoo, “Monod-type two-substrate models for batch anaerobic co-

digestion,” submitted for publication.

[14] J. Shen, H. Yan, R. Zhang, G. Liu, and C. Chen, “Characterization and methane production of different nut

residue wastes in anaerobic digestion,” Renewable Energy, vol. 116, pp. 835–841, 2018.

[15] C. Siripatana, S. Jijai, and P. Kongjan, “Analysis and extension of Gompertz-type and Monod-type equations

for estimation of design parameters from batch anaerobic digestion experiments,” in AIP Conference

Proceedings, 2016, vol. 1775, p. 030079.

[16] N. Scarlat, J.-F. Dallemand, and F. Fahl, “Biogas: Developments and perspectives in Europe,” Renewable

Energy, vol. 129, pp. 457–472, 2018.

94

Effect of Temperature on Biogas Production from Oil Palm

Empty Fruit Bunch and Mesocarp Fiber: Experimental and

Modeling

Pornwimon Wadchasit1, Chairat Siripattana1, 2, Jiravut Seengenyoung3,

Santi Thaweesaksakul4 and Kamchai Nuithitikul1, 2, a)

1Department of Civil and Environmental Engineering, School of Engineering and Resources, Walailak University,

Nakhon Si Thammarat, Thailand. 2Biomass and Oil-Palm Center of Excellence, Walailak University, Nakhon Si Thammarat, Thailand.

3Department of Biotechnology, Faculty of Science, Thaksin University, Patthalung, Thailand. 4Waste and Energy Management Company Limited, Suratthani, Thailand.

a)Corresponding author: [email protected]; [email protected]

Abstract. This study aims to investigate the effect of temperature on biogas production from oil palm empty fruit bunch

(EFB) and mesocarp fiber (MF) residues. EFB and MF were mixed with inoculum and anaerobically digested at either

mesophilic (40oC) or thermophilic (55oC) condition. Two kinetic models (Gompertz and Monod two-substrate models)

were tested with the experimental data. Preliminary economic analysis was also performed. The results showed that

thermophilic digestion gave higher yields of biogas than mesophilic digestion for both EFB and MF. The highest biogas

yield (439.0 ml/g VS) was obtained from the digestion of MF at 55 oC. Due to a high level of slowly degradable

substances (50-80% as estimated using Monod two-substrate model) in both EFB and MF, the biogas yield data could not

be represented well by traditional Gompertz and simple Monod models so that their two-substrate variants were needed

to describe the curves adequately.

Keywords—biogas; empty fruit bunch; mesocarp fiber; modeling.

INTRODUCTION

Recently, renewable energy business in Thailand is blooming. Palm oil mill effluent (POME) has become a

valuable asset and mostly used as a substrate for biogas production. Anaerobic digestion is employed to produce

biogas which is finally fed to gas engines to generate electricity. This process greatly reduces energy cost in

factories and the produced surplus electricity can be sold to electrical authorities or consumed in local communities.

However, the current amount of POME alone cannot meet the demand for biogas production anymore. Thus, the

quest for new (and renewable) raw materials is warrant and it is found that two most potential sources from the

residues in palm-oil mills are empty fruit bunch (EFB) and mesocarp fiber (MF). According to the statistics,

Thailand has produced more than 11 million tons of palm oil annually. It is reported that palm residues from the

processing of one ton of fresh fruit bunches contain about 23% EFB, 12% MF and 5% shells [1]. These residues

could be combusted directly in biomass power plants, or alternatively, used as raw materials to produce biogas.

Most of EFB wastes in Thailand are currently incinerated or dumped in the field. These practices create

environmental problems. EFB is reported to have the average fiber length of 0.53 mm which is shorter than other

non-wood materials [2]. This promotes more digestion under anaerobic condition. To minimize the pollution and

efficiently utilize EFB, previous studies have been carried out to evaluate the feasibility of biogas production from

EFB under either mesophilic or thermophilic condition. The yields of methane from thermophilic digestion of EFB

were 153-202 ml CH4/g VS depending on the initial organic loadings [1]. Under mesophilic condition, the yield of

methane from EFB was 358 ml CH4/g VS [3]. MF is a part of the external layer of palm fruits and is usually left after the extraction of palm oil. MF is currently

used as a fuel in boilers which generate steam for sterilizing palm fruit bunches and generating electricity, a cover of

the soil around palm trees to retain moisture, and a fertilizer because of its high nutrient content. It is a potential raw

material for the production of biogas. Biological pretreated MF was digested resulting in the biogas yield of 37 ml/g

95

VS whereas the untreated MF gave the lower yield of 16 ml/g VS [4]. Appropriate pretreatment of MF could

improve anaerobic digestion at 35 oC leading to the maximum yield of 199 ml/g substrate [5].

Anaerobic digestion occurs when microorganisms, especially bacteria, degrade organic matters in the absence of

oxygen and produce biogas which typically consists of 40-60% CH4, 60-40% CO2 and small amounts of other gases.

The process is complex involving four main steps: hydrolysis, acidogenesis, acetogenesis/dehydrogenation and

methanogenesis [6]. To achieve complete anaerobic digestion of organic matters, several factors must be considered.

One of the most important factors is digestion temperature. In anaerobic digestion, there are generally two

temperature ranges: (i) mesophilic range which is approximately 20-45°C but the conventional range was 35-40 oC,

and (ii) thermophilic range which is approximately 45-70°C but the conventional range was 55-70°C [7]. In general,

mesophilic anaerobic digestion is more stable but gives lower loading capacity and biogas yield than thermophilic

digestion [7]. The growth of methanogens bacteria is more favored under thermophilic condition than mesophilic

condition [8]. Therefore, the main objective of this research is to investigate the effect of temperature on biogas

production from EFB and MF residues. Kinetic models based on Gompertz and Monod two-substrate models were

proposed in order to explain biogas production from EFB and MF. Finally, preliminary economic analysis was

performed.

MATERIALS AND METHODS

Characterization of Substrates and Inoculum

EFB was acquired from EFB shredders in a palm-oil mill in Thailand. It had an average size of 3-5 cm. The

EFB samples were kept in a closed container avoiding the contact of sunlight and humidity. Similarly, MF was

obtained from the same mill and had the fiber length of 1-2 cm. Total solid (TS) and volatile solid (VS) of EFB and

MF were analyzed according to standard methods for the examination of water and wastewater (APHA).

The inoculum of methanogenic bacteria was obtained from an anaerobic sludge at the bottom of the most active

anaerobic pond in a palm-oil mill. The inoculum was capable to produce methane as confirmed by the Specific

Methanogenic Activity (SMA): 0.2350 g CH4-COD/g VSS/day. Prior to use in the experiment, the inoculum was

left at 40oC for 3 days in order to completely consume the residue nutrients.

Biochemical Methane Potential (BMP) Experiments

The methane production from EFB and MF was carried out under high solid anaerobic digestion (10-15 %TS) in

a batch mode at 40 oC (mesophilic) and 55 oC (thermophilic). The 500-ml serum bottles were used as digesters in all

experiments. The bottles were covered with air-tight caps. Oxygen was first removed by flushing the bottle

headspace with nitrogen gas. The temperature was maintained in an incubator. Digestion was continued for 40 days.

The volume of biogas was measured daily by water displacement method. At the end of the digestion process,

biogas samples were analyzed for the percentages of CH4, CO2 and H2S using Biogas GFM 416. The structure of

EFB residue after digestion was analyzed by Stereo Microscope (Nikon SMZ645) and compared with the original

one. The experimental apparatus is shown in Fig. 1. All experiments were carried out in triplicate manners. Five

different operating conditions were used: (i) 10 g of EFB and 160 g of inoculum digested at 40 oC; (ii) 10 g of MF

and 160 g of inoculum digested at 40 oC; (iii) 10 g of EFB and 160 g of inoculum digested at 55 oC; (iv) 10 g of MF

and 160 g of inoculum digested at 55 oC; and (v) inoculum 250 g without substrate (control experiment).

FIGURE 1. Substrates and experimental apparatus.

96

Kinetic Models of Biogas Production

Four types of accumulated biogas evolution are proposed as shown in Fig. 2 [16]. Type I is referred to a single

substrate consumed by a single group of microorganisms. Type II explains multiple substrates consumed in parallel

or sequentially by one or multiple groups of microorganisms. Type III describes multiple substrates but can be

simplified by three categories: easily degradable, slowly degradable and intermediates. Finally, type IV is referred to

multiple substrates with complex chain of consumption by several groups of microorganisms [16].

FIGURE 2. Four types of accumulated biogas curves.

Gompertz Two-Substrate Model

Traditional Gompertz model was used and modified by researchers [16] to Gompertz two-substrate model as

shown below:

( ) ( ) ( )( ) ( ) ( )s

R e R e1 π' ' -1 'me msP = P + P exp -exp λ - t +1 + tan κ t - τ + P - P + P exp -exp λ - t +1e e e se0 s0' 'π 2P Pe s

where '

P is accumulated production of biogas (ml), P∞ is biogas production potential (ml), Rm is the maximum

specific biogas production rates (ml/d), λ is lag phase period or minimum time to produce biogas (days), τ is the

switching time, and e is a constant (2.7183). The subscripts e and s represent easily and slowly degradable

substances (ED and SD), respectively.

Monod Two-Substrate Model

Monod two-substrate model has been developed [11] based on the following assumptions: (i) endogenous

metabolism is included; (ii) there are two groups of microorganisms: Xe (consumes Se and Si) and Xs (grows only on

Xe); (iii) a so-called intermediate Si obtained from SS in hydrolysis step is to be consumed by the microorganism

(Xe). Based on these assumptions, the following ODEs can be written:

dX μ Ss ms s= μ X - k X , μ =s s s sds

dt K + SsSs

( )dX μ Se me e

= [μ + g t, S , S μ - k ]X , μ =e e e ei i dedt K + SeSe

97

dS μs s= - X + f (k X + k X )s s eSsX ds de

dt YXsSs

dS μe e= - Xedt Y

XeSe

( )g t,S ,SdS 1- Y μ Se iXsSsi mi i= f μ X - μ X , μ =s s eis i idt Y Y K + SXsSs XeSi Si i

( )YYdP PSe PSi= μ X + g t,S ,S μ Xe e e ei idt Y Y

XeSe XeSi

where X, S, P are the concentrations of microorganism, the substrate and product (biogas), respectively; and

m are general and maximum specific growth rates; kd is specific death rate; KS is saturation constant; YXS and YPS

are microorganism yield and product yield factors, respectively. The subscripts s and e represent slowly degradable

(SD) and easily degradable (ED) substrates, i.e. KSs and KSe are saturation constants of SD and ED substrates; de

k

and ds

k are specific death rates of the equations: dXs/dt = sXs - kdsXs and dSs/dt = -(s/YXsSs)Xs + fSsX(kdsXs +

kdeXe), respectively. YXsSs, YXeSe, YXeSi, YPSe and YPSi are the corresponding yield coefficients as specified by the

subscripts. fSsX and fis are conversion factors for X→ SS and SS → Si, respectively. Finally, the switching or

preference function g(t, P) to fit the experimental data was obtained as following:

( )1 π-1g t = tan κ t-t +π 2c

( ) ( )( )1 π-1g P = tan κ P - (P ) +c cπ 2

and ( )P = Y S -SPS e ec

RESULTS AND DISCUSSION

Properties of Inoculum and Substrates

The inoculum, MF and EFB were analyzed for volatile solid (VS) and total solid (TS) contents as summarized in

Table 1. MF had slightly lower TS but higher VS content than EFB. As a result, the mass ratio of VS to TS of MF

was much higher than EFB. This ratio is believed to play an important role on the yield of biogas as discussed later.

The TS and VS contents of EFB in this study were within the ranges (43.6-99.7 %TS and 39.2-79.2 %VS)

previously reported by researchers [1,3,12].

TABLE 1. Properties of the prepared inoculum and substrates.

Properties Inoculum EFB MF

Total solid (TS), % 6.14 79.40 78.16

Volatile solid (VS), % 3.99 61.10 72.20

VS/TS, % 64.98 76.95 92.37

Under the microscope at 300x magnification, the initial fibers of EFB were clearly different from those after

digestion (Fig. 3). The initial fibers were tightly adhered which is similar to the fiber structure of wood materials.

After digestion, the fiber bundles were swollen and disintegrated slightly. This is because the lignocellulose

structure of EFB is very difficult to degrade under anaerobic digestion. Chemical or biological pretreatments of EFB

prior to digestion are usually required to promote biodegradability and consequently biogas production [12].

98

(a) (b)

FIGURE 3. Structures of EFB before and after digestion: (a) 100x; (b) 300x.

Effect of Temperature on Biogas Production

The yields of biogas produced from digestion of EFB and MF at two temperatures for 40 days are reported in

Table 2. At 40 o C, the total biogas yields from EFB and MF were 124.4 ml/g EFB (equivalent to 234.0 mL/g VS)

and 226.4 ml/g MF (equivalent to 340.3 ml/g VS), respectively. At 55 oC, biogas yields from EFB and MF were

190.0 ml/g EFB (341.4 mL/g VS) and 297.8 ml/g MF (439.0 ml/g VS), respectively. For the same substrate,

thermophilic digestion (55 oC) significantly gave higher yield of biogas than mesophilic digestion (40 oC), which is

in agreement with previous studies [8,13,14]. MF gave the higher yields of biogas than EFB under either mesophilic

or thermophilic condition. This is owing to the smaller size of MF compared to EFB. Moreover, MF is likely to

contain more organic content than EFB. The yields of biogas are in agreement with the initial VS/TS ratio: MF

exhibited the higher VS/TS ratio than EFB (92.37% versus 76.95%, respectively). The lignin content of EFB has

been reported to lower than that of MF [15]. It is generally accepted that lignin component is hard to degrade under

anaerobic digestion. Therefore, EFB with lower lignin content than MF should produce higher yield of biogas.

However, this is not always true as evidenced by the results of this study. It is noted that without nutrients, inoculum

produced very small amount of biogas as indicated by the result of the control experiment (Table 2).

The biogas compositions from EFB at 40 oC were 64.5 % CH4, 32.1 % CO2, and 120 ppm H2S. For biogas

produced from MF at 40 oC, there were 62.4 % CH4, 34.5 % CO2, and 145 ppm H2S. At 55 oC, the yields of CH4

slightly decreased: biogas compositions from EFB were 62.0 % CH4, 34.7 % CO2, and 155 ppm H2S; from MF were

61.3 % CH4, 35.5 % CO2, and 160 ppm H2S. The CH4 contents in biogas produced from MF were slightly lower

than those from EFB at the same temperature although the total yields of biogas were higher. This might be

attributed to the difference in compositions of the raw materials. Increasing the digestion temperature slightly

decreased the CH4 contents in the produced biogas. This is likely to be due to the diversity of methanogenic genera

at different temperatures. The dominant methanogenic community was found to change significantly under

thermophilic digestion of EFB at 55 oC compared to mesophilic digestion at 37 oC [14].

TABLE 2. Total biogas yields from EFB and MF at 40 and 55 oC.

Sample Total biogas volume Total biogas yield

(ml) (ml)* (ml/g Substrate) (ml/g VS)

EFB (40oC) 1430 1244 124.4 234.0

MF (40oC) 2450 2264 226.4 340.3

EFB (55oC) 2086 1900 190.0 341.4

MF (55oC) 3164 2978 297.8 439.0

Control 186 - - -

*Values after deducted by biogas volume of the control experiment.

A comparison of accumulated yields of biogas from EFB and MF at 40 oC and 55 oC is shown in Fig. 4. It is

obvious that 40 days are not sufficient for microbes to completely digest the substrates because both EFB and MF

contain large portion of slowly-degradable components. All the accumulated-biogas curves were consistent both in

terms of the effect of substrate type and temperature. The higher yields were associated with MF and higher

temperature. The accumulated yield of biogas from MF was about 45% higher than EFB at 40 oC and 28% higher at

After

Before

99

55 oC. The accumulated-biogas curves (both EFB and MF) were characterized by fast initial biogas generation rate

which continued until 10 days, followed by slow biogas production rate which extended to more than 40 days (the

end of the experiments). The same trend of accumulated biogas yields versus time was previously reported in

thermophilic digestion of EFB [13].

FIGURE 4. Accumulated biogas produced from EFB and MF at 40 oC and 55 oC.

When looking into the daily yields of biogas (Fig. 5), it was found that the maximum biogas yields from EFB at

40 and 55 oC were 116 ml/g EFB/day (on 3rd day) and 188 ml/g EFB/day (on 5th day), respectively. The yields were

lower than those reported by O-Thong et al. [1]; the yield from pretreated EFB was 202 ml CH4/g VS at initial

organic loading of 20 g VS/l. This is owing to the low biodegradability of lignocellulosic structure of untreated EFB

used in this study. Large composition of biofibers in EFB is difficult to be digested [1]. Moreover, the difference in

biogas yields is partly due to the use of different inocula.

For MF, the maximum yields at 40 and 55 oC were 296 ml/g MF/day (on 3rd day) and 412 ml/g MF/day (on 5th

day). The reduction of biogas production from both EFB and MF was noticed after 20 days. Without inoculum, very

small yields of biogas (16 and 37 ml/g VS) were obtained from untreated and biologically pretreated MF at ambient

temperature [4]. This suggests the importance of inoculum for anaerobic digestion of hardly digested substrates such

as MF. Another way to improve biogas yield from MF is co-fermentation technique, e.g. MF mixed with POME.

There are several factors needed to be controlled properly in order to increase the yields of biogas from EFB and

MF. Such factors include C/N ratio, pH, initial organic loading [1,15]. Therefore, a further study is required in order

to achieve the highest yield of biogas from the digestion of EFB and MF.

FIGURE 5. Biogas productivity rate of EFB and MF at 40 oC and 55 oC.

100

Kinetic Models

Initially, the traditional Gompertz model and simple Monod model were tested and could not satisfactorily fit the

experimental data (not shown here): if we forced the models to fit the initial period of the curves, large discrepancy

showed up at the following period. However, the modified Gompertz and Monod two-substrate models were able to

describe the experimental data very well as shown in Fig. 6 with the model parameters summarized in Table 3.

FIGURE 6. Biogas accumulation versus time for: (a) Gompertz two-substrate model; (b) Monod two-substrate model.

TABLE 3. Parameters for Gompertz and Monod two-substrate models and correlation coefficients (R2).

Model Parameter EFB (40oC) MF (40oC) EFB (55oC) MF (55oC) Control

Gompertz two-

substrate

P∞ (ml) 276.50 350.15 550.23 485.41 22.53

Pe∞ (ml) 105.34 171.79 237.06 234.38 14.92

fSs = 1-( Pe∞/ P∞)

0.619 0.509 0.569 0.517 -

Rms (ml/d) 3.71 5.58 2.82 6.10 0.04

Rme (ml/d) 12.49 38.01 23.70 39.64 0.61

κ (no unit) 0.34 1.60 0.23 1.46 0.40

αe (d-1) 0 1.66 0.70 1.33 0

αs (d-1) 0 0 0 0 0

tr 10.19 1.39 21.59 0.36 1.65

R2 0.9988 0.9993 0.9988 0.9988 0.8857

Monod two-

substrate

µme (d-1) 0.137 0.29 0.2 0.66 0.817

fse (no unit) 0.108 0.2 0.19 0.013 0.301

µms (d-1) 0.015 0.058 0.038 0.009 0.246

µmi (d-1)

0.137 0.29 0.2 0.66 0.817

kde (d-1) 0.0429 0.0429 0.0429 0.0429 0.0429

kds (d-1)

0.001 0.006 0.004 0.0009 0.025

fSsX (no unit) 0.99 0.9 0.9 0.9 1.1

KSe (mg/l) 8500 19000 5000 20000 2850

KSi (mg/l) 3000 19000 3000 20000 2850

Kss (mg/l) 3000 18000 3000 12000 2978

YXeSe 0.15 0.057 0.15 0.112 0.16

YXeSi 0.15 0.05 0.15 0.2 0.16

YXeSs 0.24 0.29 0.24 0.29 0.408

YPSe (ml/(mg/l)) 0.0158 0.0152 0.0092 0.033 0.005

YPSi (ml/(mg/l)) 0.0158 0.021 0.0092 0.428 0.005

S0 (mg/l) 38190 45125 38190 45125 3990

fSs (no unit) 0.702 0.644 0.5 0.799 0.695

S0s (mg VS/l)

26809 29061 19095 36055 2773

101

S0e (mg VS/l)

11381 16065 19095 9070 1217

X0 (mg/l) 4527 1950 5000 1400 539

fXs (no unit) 0.702 0.6 0.7 0.64 0.287

X0s (mg/l)

3178 1170 3500 896 155

X0e (mg/l)

1349 780 1500 504 384

κ (no unit) 0.01 0.0002 0.01 0.0001 0.002

fc (no unit) 2 0.1 1 0.1 0.8

Pc (ml) 360 24 176 30 4.868

R2

0.997 0.994 0.997 0.997 0.988

As shown in Table 3, the correlation coefficients (R2) were greater than 0.99 for both models, except for the

control experiment. However, the Gompertz model gave a better fit than the Monod model as indicated by the

higher values of R2. One interesting parameter is the fraction of the slowly degradable substrate (fSs) of which both

models estimated as in 0.5-0.8 range although both models did not agree well in term of the values. This is because

40 days is not sufficient to observe the final trend of the ultimate biomass generation (P). It was confirmed that the

accumulated biogas evolution curves were classified as type II according to Fig. 2, suggesting that multiple

substrates are consumed in parallel or sequentially by one or multiple groups of microorganisms.

Preliminary Economic Analysis

Preliminary economic analysis for the production of biogas from EFB and MF was simply performed and

summarized in Table 4. The incomes of produced biogas were calculated based on the current values that 1 Nm3 of

biogas could generate 2.2 kW.h of electricity and 1 kW.h could be sold to local authorities at the price of 3.5 Baht.

The net incomes were calculated from the deduction of raw material and heating costs from the incomes obtained

from selling of the produced gas. As shown in Table 4, anaerobic digestion at 55 oC gave higher net incomes than at

40 oC significantly. Nevertheless, the net incomes from the production of biogas from MF and EFB at either

mesophilic or thermophilic condition were not different significantly. At present, the amounts of MF residues from

palm-oil mills in Thailand are not large as EFB. Since MF contains lower moisture content than EFB, it is usually

used as a solid fuel in palm-oil mills. Moreover, MF residues are more easily contaminated, e.g. with shells, than

EFB. With these reasons, it is recommended to produce biogas from EFB under thermophilic condition when it

comes to a commercial-scale production.

TABLE 4. Preliminary economic analysis of biogas production from EFB and MF.

Material Produced biogas

(m3/ton Substrate)

Cost of material

(Baht/ton)

Income

(Baht/ton)

Heating cost*

(Baht/ton)

Net income

(Baht/ton)

EFB (40oC) 124.4 200 957.88 - 757.88

MF (40oC) 226.4 1000 1773.28 - 773.28

EFB (55oC) 190.0 200 1463.00 160 1113.00

MF (55oC) 297.8 1000 2293.06 160 1143.06

*Calculated from the heat required to raise the temperature from 40 to 55 oC.

CONCLUSIONS

The yields of biogas produced from EFB and MF under thermophilic digestion (55 oC) were greater than

those under mesophilic condition (40 oC). At the same temperature, MF gave the higher yield of biogas than EFB.

The highest biogas yield obtained from the digestion of MF at 55 oC was 439.0 ml/g VS. Both Gompertz and Monod

two-substrate models were able to describe the experimental data very well; a slightly better fit was observed with

the Gompertz two-substrate model. The economic analysis suggests that thermophilic digestion of EFB to produce

biogas is likely to be scaled up to commercial manufacture.

102

ACKNOWLEDGMENTS

This project was financially supported by Research and Researchers for Industries, The Thailand Research Fund

(RRI_PHD61I0044). Raw materials for biogas production were kindly supported by Waste and Energy Management

Company.

REFERENCES

[1] S. O-Thong, K. Boe, and I. Angelidaki, “Thermophilic anaerobic co-digestion of oil palm empty fruit bunches

with palm oil mill effluent for efficient biogas production,” Applied Energy, vol. 93, pp. 648-654, 2012.

[2] A. Ferrer, A. Vega, P. Ligero, and A. Rodríguez, “Pulping of empty fruit bunches (EFB) from the palm oil

industry by formic acid,” BioResources, vol. 6, no. 4, pp. 4282-4301, 2011.

[3] W. Suksong, P. Kongjan, P. Prasertsan, T. Imai, and S. O-Thong, “Optimization and microbial community

analysis for production of biogas from solid waste residues of palm oil mill industry by solid-state anaerobic

digestion,” Bioresource Technology, vol. 214, pp. 166-174, 2016.

[4] M. Saidu, A. Yuzir, M. R. Salim, Salmiati, S. Azman, and N. Abdullah, “Biological pre-treated oil palm

mesocarp fibre with cattle manure for biogas production by anaerobic digestion during acclimatization phase,”

International Biodeterioration & Biodegradation, vol. 95, 189-194, 2014.

[5] A. G. Costa, G. C. Pinheiro, F. G. C. Pinheiro, A. B. D. Santos, S. T. Santaella, and R. C. Leitão, “Pretreatment

strategies to improve anaerobic biodegradability and methane production potential of the palm oil mesocarp

fibre,” Chemical Engineering Journal, vol. 230, pp. 158-165, 2013.

[6] N. Aryal, T. Kvist, F. Ammam, D. Pant, L. D. M. Ottosen, “An overview of microbial biogas enrichment,”

Bioresource Technology, vol. 264, pp. 359-369, 2018.

[7] Y. Y. Choong, K. W. Chou, and I. Norli, “Strategies for improving biogas production of palm oil mill effluent

(POME) anaerobic digestion: A critical review,” Renewable and Sustainable Energy Reviews, vol. 82, pp.

2993-3006, 2018.

[8] P. Vindis, B. Mursec, W. Janzekovic, and F. Cus, “The impact of mesophilic and thermophilic anaerobic

digestion on biogas production,” J. of Achievements in Materials and Manufacturing Engineering, vol. 36, pp.

192-198, 2009.

[9] S. Jijai and C. Siripatana, “Kinetic model of biogas production from co-digestion of Thai rice noodle

wastewater (Khanomjeen) with chicken manure,” Energy Procedia, vol. 138, pp. 386-392, 2017.

[10] C. Siripatana, S. Jijai, and P. Kongjan, “Analysis and extension of Gompertz-type and Monod-type equations

for estimation of design parameters from batch anaerobic digestion experiments,” AIP Conference Proceedings,

vol. 1775, pp. 1-8, 2016.

[11] N. Rakmak, C. Siripatana, S. Jijai, and L. Noynoo, “Monod-type two-substrate models for batch anaerobic co-

digestion,” submitted for publication.

[12] D. C. Nieves, K. Karimi, I. S. Horváth, “Improvement of biogas production from oil palm empty fruit bunches

(OPEFB),” Industrial Crops and Products, vol. 34, pp. 1097-1101, 2011.

[13] W. Suksong, A. Jehlee, A. Singkhala, P. Kongjan, P. Prasertsan, T. Imai, and S. O-Thong, “Thermophilic solid-

state anaerobic digestion of solid waste residues from palm oil mill industry for biogas production,” Industrial

Crops and Products, vol. 95, pp. 502-511, 2017.

[14] A. Walter, I. K. Franke-Whittle, A. O. Wagner, and H. Insam, “Methane yields and methanogenic community

changes during co-fermentation of cattle slurry with empty fruit bunches of oil palm,” Bioresource Technology,

vol. 175, pp. 619-623, 2015.

[15] M. Y. Nurliyana, P. S. H’ng, H. Rasmina, M.S. UmiKalsom, K. L. Chin, S. H. Lee, W. C. Lum, and G. D.

Khoo, “Effect of C/N ratio in methane productivity and biodegradability during facultative co-digestion of palm

oil mill effluent and empty fruit bunch,” Industrial Crops and Products, vol. 76, pp. 409-415, 2015.

[16] L. Noynoo, N. Rakmak, C. Siripatana, S. Jijai, and K. Phayunphan, “Gompertz-Type Two-Substrate Models for

Batch Anaerobic Co-Digestion,” submitted for publication.