Selection of variables for on-line monitoring, diagnosis,

and control of anaerobic digestion processes

F. Molina, M. Castellano, C. Garcıa, E. Roca and J. M. Lema

ABSTRACT

F. Molina

C. Garcıa

E. Roca

J. M. Lema

Department of Chemical Engineering,

Universidad de Santiago de Compostela,

Spain

E-mail: carlos.garcia.dieguez@usc.es;

enrique.roca@usc.es;

juanmanuel.lema@usc.es

F. Molina

Department of Sanitary Engineering,

Universidad de Antioquia,

Colombia

E-mail: fmolina@udea.edu.co

M. Castellano

Department of Statistics and O.R,

Universidad de Santiago de Compostela,

Spain

E-mail: maria.castellano@usc.es

This work aims to systematize the study of indicators for two types of wastewaters:

carbohydrate-based and protein-based synthetic wastewaters. Characterization of steady

states and dynamic response analysis against disturbances were carried out using both

a factorial discriminant analysis (FDA) and a phenomenological analysis, respectively.

This research seeks reconciling both sets of indicators in order to optimize resources and

provide a minimal cost in instrumentation for its implementation at industrial scale. According

to the results of this research, the best indicators for the two types of wastewaters, considering

both process steady states and organic load perturbations are: Biogas flow rate or Methane

flow rate, and Hydrogen concentration in the biogas; Volatile fatty acids and Partial alkalinity

in the liquid phase.

Key words | anaerobic digestion, FDA, process state identification, selection of variables

INTRODUCTION

Anaerobic wastewater treatment is a very complex process,

which involves biochemical transformations as well as

physicochemical processes. To avoid destabilization and

to keep a stable state, the process requires the application of

both a monitoring and a control system. The complexity and

cost of this system depend on the number of variables to be

monitored on-line. The development of a monitoring

system, diagnosis, and control (MD&C) for anaerobic

digestion require an accurately selection of variables

and/or indicators, which give valuable information about

the metabolic status of the process (Castellano et al. 2007).

These variables should fulfill the following key features:

high sensitivity, fast response, and low cost of monitoring

(Switzembaum et al. 1990; Mathiot et al. 1992). Many

researchers have studied and proposed different variables

and groups of variables as indicators of process stability for

anaerobic digestion in presence and/or in absence of

disturbances. Table 1 shows a summary of some of these

proposals. However, there is not a final conclusion about

which is the variable or group of variables that make it

possible to identify the state of the process, in an

appropriate way, and which allows an early detection of

disturbances. Meanwhile, in the proposals presented in

Table 1, only a few variables were compared at the same

time, and a high diversity of wastewater was studied using

heuristic criteria rather than statistical criteria for the

selection of variables.

In order to improve the selection of process indicators,

Castellano et al. (2007) determined the minimum number of

monitored variables for process state identification, using

factorial discriminant analysis (FDA) for winery waste-

water. When only one variable for process state identifi-

cation is used, Castellano and collaborators found that

hydrogen concentration in the gas phase seems to be the

best, since it has a high discriminatory ability among

the process states studied.

One of the purposes of this work is to find out if FDA

classification performance for winery wastewater can be

doi: 10.2166/wst.2009.379

615 Q IWA Publishing 2009 Water Science & Technology—WST | 60.3 | 2009

extended to more chemically complex wastewaters or it

would be necessary to use more sophisticated non-linear

non-symmetric discriminant techniques. So using a similar

strategy, this research aims to systematize the study of

indicators for two types of wastewaters: carbohydrate-based

and protein-based synthetic wastewaters. In addition, the

process states were characterized by evaluating the dynamic

response against disturbances. Finally, this research seeks

reconciling both sets of indicators (steady states and

dynamic states) in order to optimize resources and provide

a minimal cost in instrumentation for their implementation

at industrial scale.

MATERIALS AND METHODS

Pilot plant

Experiments were conducted in a hybrid USBF (UASB þ

AF) pilot plant with an overall volume of 1.15 m3. The on-

line monitoring system consisted of feeding and recycling

flow-meters (ABB-COPAXE and Siemens-7ME2531), input

and output reactor pH (Cole Parmer) and temperature

(Pt 100); biogas (QBiogas) flow-meter (Brooks-3240);

infrared gas analyzer (Siemens-Ultramat 22P) for measure-

ment of CH4 and CO concentrations in the gas phase;

electrochemical H2 gas analyzer (Sensotrans-Sensotox

420); Total Organic and Inorganic Carbon (TOC/TIC) in

the influent and effluent were determined on-line by catalyst

combustion oxidation and non dispersive infra red (NDIR)

CO2 detection (Shimadzu-TOC 4100); reactor top pressure

(Phead) was measured using a Siemens TMF1563 sensor.

A titrimetric AnaSensew analyzer (Ruiz-Filippi et al. 2005;

Molina et al. 2009) was used for determining both volatile

fatty acids (VFA) and partial alkalinity (PA) every 15

minutes during experimentation. The signals from the

sensors were recorded every 5 seconds. Owing to the high-

stability of the signal, the large time scale of the process, and

hydraulic retention time (HRT) of 24 h, an average moving

window of 15 min was selected for filtering these on-line

signals. A more detailed description of the pilot plant and

monitoring system has been given in Garcia et al. (2007) and

Molina et al. (2007).

Synthetic wastewaters

Both hydrolyzed protein and dextrin solution have been

used as models of rich proteins wastewater and rich

Table 1 | Summary of proposed variables and indicators used for the identification of states in anaerobic digestion

Phase Variable(s) Kind of waste or synthetic wastewater Reference

Biomass (Solid) Acetotrophic methanogenesis Municipal sludge Conklin et al. (2008)

Physiological and morphologicalcharacteristics

Ethanol plus detergent andsolvent addition

Costa et al. (2009)

Liquid H2 Glucose, protein, and lipids Bjornsson et al. (2001)

Bicarbonate alkalinity Ice cream Hawkes et al. (1994)

Partial alkalinity Municipal sludge Jantsch & Mattiasson (2004)

VFA Manure Ahring et al. (1995)

Manure plus ammoniumchloride addition

Nakakubo et al. (2008)

Gas H2 Vinasses Moletta (1989)

Butyric acid Slater et al. (1990)

VFA and glucose Jones et al. (1992)

Synthetic baker’s yeast Guwy et al. (1997)

H2 and CO Sludge plus heavy metals addition Hickey & Switzenbaum(1990, 1991)

H2, CH4 and CO2 Vinasses Mathiot et al. (1992)

H2 and QBiogas PET manufacture Huang et al. (2000)

616 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009

carbohydrates wastewater, respectively. Macro-nutrients

(nitrogen and phosphorus), alkalinity, and micro-nutrients

were added to carbohydrate-based wastewater, reaching a

final COD/Bicarbonate/N/P ratio of 1000/600/7/1. Based

on the fact that protein-based wastewater had been

contributed with nitrogen and alkalinity, phosphorus and

micronutrients were added. In both wastewaters, micro-

nutrients were added according to the proportions

suggested by Angelidaki & Sanders (2004) for testing

specific methanogenic activity (SMA).

Experimental conditions

Experimental conditions for both steady states and dyna-

mical states were 37 ^ 28C, feed flow rate of 48.5 L h21,

recirculation flow rate of 200 L h21, and a 24 h hydraulic

retention time (HRT). Organic loading rate (OLR) has kept

constant at least during 5 days before each dynamical

disturbance experiment. Organic overloads (pulses or steps)

were carried out by increasing the concentration of COD in

the feeding and maintaining the feed flow rate constant.

Table 2 shows the summary of steady states for each

type of wastewater. For protein-based wastewater in some

steady states, mixtures of protein and ethanol in order to

increase the activity of methanogenic biomass were used.

Pulse-type disturbances were intended to explore the

dynamical response of the different variables to a big

organic overload in a short time (approximately 10% of

HRT). In contrast, step-type disturbances provide infor-

mation on the dynamic adjustment of the system, in

presence of a sustained increased in the OLR. Pulse-type

and Step-type disturbances were made by increasing of

COD previous value from 5 to 10 times and 1.5 to 3 times,

respectively. Table 3 shows the OLR increment used to

provoke pulse-type and step-type disturbances for both

carbohydrate-based and protein-based-wastewaters.

Multivariate analysis of steady states

The classification between different steady states was carried

out using FDA. There are two main kind of classification

tools, the supervised classification methods and the unsu-

pervised classification methods (e.g. cluster methods)

(Hand 1981). The first kind is employed when the groups

are well-known and a well-classified data set is available,

and the second group allow to work when the either

available data can not be classified or the number of groups

are not known. FDA is a supervised discrimination tool.

FDA uses a data set (represented by the matrix X), well

classified in g known groups. The belonging to a group can

be represented in different ways. The most common

approach uses a vector Y that take from 1 to g. When a

data Xj is classified in the group h, Yj ¼ h. In this work, data

were a priori classified in the different steady states using

first Principal Component Analysis (PCA), in order to

reduce the dimension of the problem, and a linear

regression on a mobile window for checking the steady

state in first and second principal components.

Table 3 | OLR (kgCODm23 d21) changes for provoked pulse-type and step-type

disturbances

Disturbance Carbohydrate-based wastewater Protein-based wastewater

Pulse 1 5–45 4–20

Pulse 2 3.5–35 3–20

Step 1 5–7.5 4–6

Step 2 3.5–6 3–6

Table 2 | Summary of steady states for each type of wastewater

Carbohydrate-based wastewater Protein-based wastewater

Steady state OLR (kgCODm23 d21) Composition (Dextrin %) Steady state OLR (kgCODm23d21) Composition (Protein/Ethanol %)

1 3.5 100 1 2 100/0

2 5 100 2 3 67/33

3 7.5 100 3 4 50/50

4 4 100/0

5 6 67/33

6 6 67/33

617 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009

The FDA is a linear tool, so it seeks linear combination

of variables, from a geometry approach, hyper-planes (Pena

2002), where the projection of the original data set appears,

as much as possible, separated in the same way as original

groups do. FDA provides a classification of the data using

this geometrical separation between groups. The selection

of the hyper-planes or factors is made in order to minimize

the probability of miss-classification. The analytical

framework used for factors determination involves the

study of some variance and covariance matrix, in order to

minimize the Mahalanobis distances into each group and

maximize the same distance among different groups.

Geometrically the projections into hyper-planes try to

separate the data of different groups and at the same time

try to make the data of the same group be as closer as possible.

FDA provides no only the separation factor but also a

probability function of belonging to each group, so it can

be used to classified new data, not a priori classified. So it

provides a decision rule for assigning a new case to one

class, and its accuracy can be estimated using the original

data using cross-validation.

In this research several FDA analysis were carried out

in order to minimize the number of variables needed to

differentiate steady states. FDA procedure was applied on

whole data belonging to stationary states using the complete

set of monitored variables and also process state classifi-

cation capability of each variable independently and all the

variables combinations were evaluated by FDA.

Disturbances analysis

A phenomenological analysis of dynamic responses facing

organic overloads (pulses or steps) was done taking into

account the normalized slope of response (Equation 1), the

excitation level (Equation 2), and the time to maximum

value. An example of a pulse disturbance can be seen in

Figure 1, this figure shows the parameters used to evaluate

the dynamical system’s response.

Sn ¼ S=Vi ð1Þ

Le ¼ ðVf 2 ViÞ=Vi ð2Þ

Where

Sn: Normalized slope of response (h21)

Le: Excitation level

S: Slope of response

Vi: Initial value of response

Vf: Final value of response

The Excitation level expresses the sensitivity of process

when occurs an increase of OLR. The Normalized slope as a

result of slope of response divided by the initial value of the

variable includes both sensitivity and speed of response; for

this reason, it is used as a leading indicator of the system’s

response to disturbances.

RESULTS AND DISCUSSION

Steady states analysis

In a first approach, FDA was applied using all variables

for evaluating the maximum capability of classification for

steady states. A 99.5% and 100% of correct classification

for the data in different steady states was obtained

with carbohydrate-based and protein-based wastewaters,

respectively. Figure 2 presents the territorial map for

different steady states discriminated by FDA for both

wastewaters. As it can be seen in the Figure 2A, the first

discriminant function (DF1) properly separates the different

steady states for carbohydrate-based wastewater. In con-

trast to the protein-based wastewater, the situation was

more complex (see Figure 2B). This situation required that

the first and the second discriminant functions in order to

achieve a complete separation of different steady states.

In a second step, the minimum number of variables,

which allows for the steady states properly being classified

Figure 1 | Example of pulse disturbance: OLR (- - -) and QBiogas (—).

618 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009

was determined. In this case, FDA was applied to a single

variable and to combinations of variables. Table 4 shows

the percentage of properly classification applying FDA to

one variable.

As it can be seen in Table 4, for carbohydrate-based

wastewater the best classification was achieved through

the variables of the gas phase: Phead, QBiogas, and QCH4.

Phead and QBiogas variables have a good correlation

(correlation coefficient-R 2 of 0.98), thus one of this

variables without significant loss of information can be

excluded. QCH4 variable is the product of QBiogas

and %CH4 variables. Bearing in mind, that methane is

the main end product of anaerobic digestion; QCH4

variable is very useful because it offers information on the

activity of methanogenic biomass. The biggest hit in the

classification of steady states by gas phase variables was

related to the rapid degradation of the carbohydrate-based

wastewater.

The substrate protein presented a worse classification of

the steady states when a single variable was used. The two

variables which perform the best classification were: TOCi

and TICi, related to the characteristics of the influent

wastewater. The following response variables which best

classify the steady states were: AP, TOCe, AGV and QCH4.

The minor success in the classification of steady states for

the protein-based wastewater using a single variable was

related to: the wastewater complexity, the mixture of

protein and ethanol in some of these experiments, the

largest number of steady states studied, and the narrowest

rank of OLR applied.

For both types of wastewater, H2 concentration on

biogas is the variable which most poorly classifies the

different steady states. This result differs from those reported

by Castellano et al. (2007) who worked with an ethanol-

based wastewater, finding that H2 concentration in the gas

phase presented one of the biggest classification capacities.

This result may be related to both the acidification degree

variability of carbohydrate-based wastewater before feeding

the reactor and with the complexity of protein-based

wastewater.

In order to improve the capacity of steady states

classification, FDA was applied to combination of two or

three variables. For the carbohydrate-based wastewater,

Figure 2 | Territorial map for steady states discriminated by FDA using all variables. (A) carbohydrate-based wastewater: centroids (A), and steady states: 1 (W), 2 (B) y 3 (K).

(B) protein-based wastewater: centroids (A), and steady states: 1 (W), 2 (X), 3 (B), 4 (K), 5 (O) y 6 (S).

Table 4 | Percentage of successful classification for steady states using a single

variable

Variable

Carbohydrate-based

wastewater

Protein-based

wastewater

PHead 89.9 58.3

%CH4 74.5 53.4

H2 53.6 46.7

QBiogas 91.4 59.0

QCH4 91.7 65.8

TOCi 89.4 78.8

TICi 85.1 82.6

TOCe 72.9 70.3

pHe 63.6 51.3

VFA 81.5 65.2

PA 81.4 73.9

619 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009

a 95% of success in classification of the steady states was

obtained using a two-variable combination, thus resulting in

five two-variable combinations: QBiogas-pHe, QCH4-pHe,

QCH4-VFA, QCH4-PA and VFA-PA. In opposition, a three-

variable combination was needed to ensure a correct

classification of the steady states (over 95%) for protein-

based wastewater, thus resulting in three three-variable

combinations: QBiogas-VFA-PA, Phead -VFA-PA and

QCH4-VFA-PA.

Disturbance analysis

Table 5 presents some results obtained with carbohydrate-

based wastewater for normalized slope response as the

main parameter. In Table 5, it can be noticed that for

the carbohydrate-based wastewater, the best indicators

for identifying the disturbance caused by a pulse overload

are CO and H2 in the gas phase and VFA in the liquid

phase; this behavior is typical in a reactor that has suffered a

organic overload (Hickey & Switzenbaum 1991; Jones et al.

1992; Ahring et al. 1995; Huang et al. 2000). For step-type

disturbance, the best indicators are H2 in gas phase

and VFA in liquid phase, respectively. For both type of

disturbances, pulse and step, pHe in liquid phase and %CH4

in gas phase are the worst indicators for the detection of

organic disturbances.

Table 6 shows the best indicators for each type of

wastewater and type of disturbance. For carbohydrate-

based wastewater the best indicators of organic disturbances

were both CO and H2 concentrations in the gas phase, this

result is similar to that found by Hickey & Switzenbaum

(1991), these authors found a clear relation between the CO

concentration in gas phase with the VFA accumulation in

liquid phase; in the liquid phase the best indicator was VFA

concentration, this fact is related with VFA accumulation

due to an imbalance between acidogenic and methanogenic

bio-processes (Ahring et al. 1995). On the opposite for

protein-based wastewater, the best indicator of organic

disturbances was VFA concentration in liquid phase,

followed by H2 concentration in gas phase.

Results integration

FDA technique combined with a phenomenological study

of the response to disturbances allows identifying the best

useful indicators to early diagnose of different process

states. However, other technical factors would be con-

sidered in order to determine the appropriate instrumenta-

tion for this process. In this sense, the variables should have

three main characteristics: low response delay, high

sensibility and low cost of both, sensor itself and its

operation-maintenance requirements. In addition, there

are other factors that probably will influence slightly on

the results, such as sampling location or sampling rate/

interval. Finally, according to the results of this research,

Table 6 | Best indicators for the identification of organic disturbances in both types of

wastewaters

Type of

disturbance

Carbohydrate-based

wastewater

Protein-based

wastewater

Pulse CO, H2, VFA VFA, H2, QCH4,QBiogas

Step H2, QBiogas,QCH4, VFA

H2, VFA, QCH4,QBiogas

Table 5 | Results for normalized slope response (h21) as main parameter for the

identification of organic disturbances with carbohydrate-based wastewater

Type of disturbance

Phase Variable Pulse Step

Gas %CH4 0.10 0.02

H2 1.74 0.24

CO 3.10 N.D.

Qbiogas 0.84 0.05

QCH4 0.78 0.05

Liquid TOCe 0.20 0.02

pHe 0.02 0.01

VFA 1.27 0.10

PA 0.04 0.02

N.D. Not Detected.

Table 7 | Levels of instrumentation for on-line monitoring, diagnosis and control of

anaerobic digestion of carbohydrate-based and protein-based wastewaters

Phase Basic Necessary Advisable

Gas QBiogasor QCH4

H2 concentration

Liquid pHe, temperatures,feed and recyclingflowrates

VFA and PA

620 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009

and considering the monitoring cost, three levels of

instrumentation are recommended for the two types of

wastewater (see Table 7).

CONCLUSIONS

According to the results of this research, the best indicators

for the two types of wastewaters, considering both steady

states and dynamic states are: Qbiogas or QCH4 and H2

concentration in the gas phase; VFA and PA in the

liquid phase.

For both carbohydrate-based wastewater and protein-

based wastewater, H2 concentration on biogas is the variable

which most poorly classifies the different steady states. But

for detection disturbances, H2 concentration on biogas is a

useful variable in both synthetic wastewaters. Nevertheless,

for carbohydrate-based wastewater, the H2 concentration

in the gas phase depends on the acidification degree of

substrate before feeding it into the reactor; this fact can cause

errors in the diagnosis of the state of the process.

The carbon monoxide concentration in the gas phase is

less useful for the detection of disturbances, as it was

sensible to pulse disturbances only for carbohydrate-based

wastewater.

ACKNOWLEDGEMENTS

This research was supported by the Xunta de Galicia

through the project SEDDAN (PGIDIT04TAM265006PR),

Spanish National R&D Program, and European Regional

Development Fund (ERDF) through the project ANACOM

(CTQ2004-07811-C02-01).

REFERENCES

Ahring, B. K., Sanderberg, M. & Angelidaki, I. 1995 Volatile fatty

acids as indicators of process imbalance in anaerobic

digestors. Appl. Microbiol. Biotechnol. 43(3), 559–565.

Angelidaki, I. & Sanders, W. 2004 Assessment of the anaerobic

biodegradability of macropollutants. Rev. Environ. Sci.

Biotechnol. 3(2), 117–129.

Bjornsson, L., Murto, M., Jantsch, T. G. & Mattiasson, B. 2001

Evaluation of new methods for the monitoring of alkalinity,

dissolved hydrogen and the microbial community in anaerobic

digestion. Water Res. 35(12), 2833–2840.

Castellano, M., Ruiz, G., Gonzalez, W., Roca, E. & Lema, J. M.

2007 Selection of variables using factorial discriminant analysis

for the state identification of an anaerobic UASB-UAF hybrid

pilot plant, fed with winery effluents. Water Sci. Technol.

56(10), 139–145.

Conklin, A. S., Chapman, T., Zahllerc, J. D., Stenseld, H. D. &

Fergusond, J. F. 2008 Monitoring the role of aceticlasts in

anaerobic digestion: activity and capacity. Water Res. 42(20),

4895–4904.

Costa, J. C., Alves, M. M. & Ferreira, E. C. 2009 Principal

component analysis and quantitative image analysis to predict

effects of toxics in anaerobic granular sludge. Bioresour.

Technol. 100(3), 1180–1185.

Garcia, C., Molina, F., Roca, E. & Lema, J. M. 2007 Fuzzybased

control of an anaerobic reactor treating wastewaters

containing ethanol and carbohydrates. Ind. Eng. Chem. Res.

46(21), 6707–6715.

Guwy, A. J., Hawkes, F. R., Hawkes, D. L. & Rozzi, A. G. 1997

Hydrogen production in a high rate fluidised bed anaerobic

digester. Water Res. 31(6), 1291–1298.

Hand, D. J. 1981 Discriminarion and Classification. John Wiley and

Sons Ltd, New York.

Hawkes, F. R., Guwy, A. J., Hawkes, D. L. & Rozzi, A. G. 1994

On-line monitoring of anaerobic digestion: application of a

device for continuos measurement of bicarbonate alkalinity.

Water Sci. Technol. 30(12), 1–10.

Hickey, R. & Switzenbaum, M. 1990 Behavior of carbon monoxide

as a trace component of anaerobic digester gases and

methanogenesis from acetate. Environ. Sci. Technol. 24(11),

1642–1648.

Hickey, R. & Switzenbaum, M. 1991 The response and utility of

hydrogen and carbon monoxide as process indicators of

anaerobic digesters subject to organic and hydraulic overloads.

J. WPCF 63(2), 129–140.

Huang, Y. H., Huang, G. H., Chou, S. & Cheng, S. S. 2000

Hydrogen as a quick indicator of organic shock loading in

UASB. Water Sci. Technol. 42(3–4), 43–50.

Jantsch, G. & Mattiasson, B. 2004 An automated

spectrophotometric system for monitoring buffer capacity in

anaerobic digestion processes. Water Res. 38(17), 3645–3650.

Jones, R. M., MacGregor, J. F., Murphy, K. L. & Hall, E. R. 1992

Towards a useful dynamic model of the anaerobic wastewater

treatment process: a practical illustration of process

identification. Water Sci. Technol. 25(7), 61–71.

Mathiot, S., Escoffier, Y., Ehlinger, F., Couderc, J. P., Leyris, J. P. &

Moletta, R. 1992 Control parameter variations in an anaerobic

fluidised bed reactor subjected to organic shockloads. Water

Sci. Technol. 25(7), 93–101.

Moletta, R. 1989 Comparaison de la response de differents

parametres a une surcharge organique d’un digesteur

anaerobie. Environ. Technol. Lett. 10, 173–184.

Molina, F., Ruiz-Filippi, G., Garcıa, C., Roca, E. & Lema, J. M. 2007

Winery effluent treatment at an anaerobic hybrid USBF pilot

plant under normal and abnormal operation. Water Sci.

Technol. 56(2), 25–31.

621 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009

Molina, F., Ruiz-Filippi, G., Garcıa, C., Lema, J. M. & Roca, E.

2009 Pilot-scale validation of a new sensor for on-line

analysis of volatile fatty acids and alkalinity in anaerobic

wastewater treatment plants. Environ. Eng. Sci. 26(3), 641–649.

Nakakubo, R., Moller, H., Nielsen, A. & Matsuda, J. 2008 Ammonia

inhibition of methanogenesis and identification of process

indicators during anaerobic digestion. Environ. Eng. Sci.

25(10), 1487–1496.

Pena, D. 2002 Multivariate Data Analysis (in spanish). McGraw-Hill

interamericana de Espana. S.A.U.

Ruiz-Filippi, G., Molina, F., Steyer, J. P., Vanrolleghem, P., Zaher, U.,

Roca, E. & Lema, J. M. 2005 Industrial scale validation of a new

titrimetric sensor for anaerobic digestion processes: comparison

of methodologies. In Proceedings of the 2nd IWA Conference

on Instrumentation, Control and Automation for Water and

Wastewater Treatment and Transport System. IWA Publishing,

Busan, Korea, pp. 133–138.

Slater, W. R., Merigh, M., Ricker, N. L., Labib, F., Ferguson, J. F. &

Benjamin, M. M. 1990 A microcomputer-based

instrumentation system for anaerobic wastewater treatment

processes. Water Res. 24(1), 121–123.

Switzenbaum, M. S., Giraldo-Gomez, E. & Hickey, R. F. 1990

Monitoring of the anaerobic methane fermentation process.

Enzyme Microb. Technol. 12(10), 722–730.

622 F. Molina et al. | Variables for monitoring and control of anaerobic digestion processes Water Science & Technology—WST | 60.3 | 2009