HUNGARIAN JOURNAL OF
INDUSTRY AND CHEMISTRY
Vol. 47(2) pp. 43–51 (2019)
hjic.mk.uni-pannon.hu
DOI: 10.33927/hjic-2019-19

RESIDENCE TIME DISTRIBUTION-BASED ANALYSIS OF AN
INDUSTRIAL-SCALE BIOGAS FERMENTER

ANDRÁS TANKOVICS1 , DÁVID TAKÁCS1 , JUDIT SZENDEFY2 , BÉLA CSUKÁS1 , AND MÓNIKA
VARGA*1

1Institute of Methodology, University of Kaposvár, Guba Sándor u. 40, Kaposvár, 7400, HUNGARY
2Biogas Plant, Magyar Cukor Zrt., Pécsi u. 10, Kaposvár, 7400, HUNGARY

Residence Time Distribution (RTD) measurement-based analysis of mixing conditions on an industrial-scale (13, 000
m3) anaerobic digester of pressed sugar-beet slices at Kaposvár Sugar Factory of Magyar Cukor Zrt. was studied.
The lithium salt tracing technique was applied, while the quantity of the lithium chloride tracer and the sampling of the
effluent were designed by a preliminarily studied simulation model of mixing. The lithium concentration at the outlet was
analysed by Inductively coupled plasma–optical emission spectroscopy (ICP-OES). Taking into account the geometrical
arrangement, the biogas flow produced and the cyclically changing recycle flow, various mixing models were generated
with different compartmentalization and flow structures by applying the method of Programmable Process Structures.
The simulation-based approximate identification of the mixing model was accomplished by a heuristic approach that
took into consideration multiple structures with changing mixing flows. A model with an advantageously small number of
compartments and parameters was sought which satisfies the measured RTD. The results suggest the intensive mixing
of upper levels with a poorly mixed lower level, which contributes to the long tail in the RTD. The actual set-up supports a
good horizontal distribution of the sugar-beet slices and the microbial biomass, while the limited degree of vertical mixing
helps to avoid the elutriation of the useful microbiome. The suggested mixing model will be combined with the formerly
elaborated model involving 9 bacterial groups.

Keywords: anaerobic digestion, mixing of digester, Residence Time Distribution, lithium tracing,
Programmable Process Structures

1. Introduction

The mixing of anaerobic digesters is a critical issue be-
cause it should support the uniform distribution of raw
materials to be digested and the bacterial biomass along
the cross-section of the unit. Moreover, the excessive sed-
imentation of the solid (bacterial) phase at the bottom
of the digester should be avoided. However, an unnec-
essarily high degree of mixing may elutriate the bacterial
biomass that decreases the effectiveness of transforma-
tions and may cause surplus environmental load.

The computational modelling of anaerobic digesters
was developed from the modelling of wastewater treat-
ment and degradation [1, 2] and from the ADM models
designed by IWA [3, 4]. A comprehensive review from
2013 is available [5].

Knowledge about mixing in large biogas digesters is
still in its infancy, so the objective of this study is to
broaden this by determining the minimum retention time
of substrates fed into anaerobic digesters and estimate
the distribution time of substrates before being extracted
from the investigated digester.

*Correspondence: varga.monika@ke.hu

Over the last century, compartmentalized models
were successfully used given the lack of computational
fluid dynamics (CFD) models available, e.g. for study-
ing an interesting scale-up problem, where the partially
mixed pilot performed better than the perfectly mixed
laboratory unit [6]. Moreover, at that time, a review com-
paring compartmentalization with CFD was published [7]
which concluded that CFD is the most scale–independent
method for mixing and scale-up studies, however, it is
hampered by limitations such as high computational de-
mands and the inadequacy of submodel consideration,
e.g. biokinetics.

In a recently published detailed review [8], the authors
concluded that “Compartmental models allow multi-scale
modelling with low computational time compared to
a full coupled model (e.g. reactive numerical simula-
tions). Thanks to these main characteristics, compart-
mental models are able to model complex full-size in-
dustrial systems. An effective compartment model could
handle multiple, multiphysics phenomenological models
(detailed kinetic reaction scheme, complex heat and mass
transfer model, population balance, etc.) that could not be
included in CFD analysis. Observed deviation between

https://doi.org/10.33927/hjic-2019-19
mailto:varga.monika@ke.hu


44 TANKOVICS, TAKÁCS, SZENDEFY, CSUKÁS, AND VARGA

fully detailed CFD model and compartment model show
very small results deviation despite of very significant re-
duction in calculation time (3 orders of magnitude).”

An interesting new example of a hypothesis-driven
compartmental model is given in [9].

Two full-scale digesters from a biogas plant (2, 000
m3 and 1, 500 m3) equipped with different mixing sys-
tems and filled with different substrates were investigated
by Kamarad et al. [10]. To characterize the substrate
distribution, solutions of lithium hydroxide monohydrate
were used in tracer tests at concentrations of 45 − 50 mg
Li+/kg TS in digesters. The tracer concentration in the
effluent of the digester was measured. Although the data
calculated by CFD methods were in very good agreement
with the results, a full comparison was not made

Kaposvár Sugar Factory of Magyar Cukor Zrt. de-
veloped an internationally straightforward anaerobic fer-
mentation technology to generate onsite used and surplus
energy with a reduced amount of waste emitted by pro-
ducing fuel gas of high methane content from the pressed
sugar beet slices. In a former PhD thesis [11], a detailed
simulation model was developed and validated for an ap-
proximately perfectly mixed pilot unit that took into con-
sideration 9 pseudogroups of bacteria by applying the
available earlier version of Direct Computer Mapping-
based modelling methodology.

The scaling up of the model to an industrial scale
[12] requires more detailed knowledge about the hydro-
dynamic conditions of the appropriately compartmental-
ized volume. The objective of this work was to measure
and analyse the Residence Time Distribution of a large in-
dustrial unit by a lithium tracer technique. The final goal
of this analysis was to embed the formerly developed de-
tailed digestion model into the compartmentalized mix-
ing model to enhance the investigated fermenter.

2. Experimental work

The industrial digester (see Fig. 1) was a cylindrical unit
with a diameter of 25 m and height of 28 m. The effluent
was removed from a volume of approximately 13, 000 m3

at a level of 20 m.
Spatially uniform feeding and appropriate mixing was

ensured by the recycle flow from a level 20 m in height
from the base of the unit and fed into the annularly placed
6 subsequent bottom segments with a prescribed cyclic
change of the flow. The pressed sugar beet slices were
also fed into this recycle flow via a screw feeder. Mixing
was enhanced significantly by the increasing upward flow
of the generated biogas. Moreover, three small mechan-
ical mixers ensured the raw materials and bacteria were
uniformly distributed in the lower third of the unit. The
slowly accumulating inorganic residue could be removed
from the middle of the bottom of the digester by means
of a slowly rotating agitator.

Figure 1: Schematic diagram of the industrial-scale fer-
menter of 13, 000 m3 in volume

2.1 Tracing and Measurement Technique

For the tracing of the flow, 40 kg of lithium chloride
p.a. was dissolved in 150 litres of tap water. Considering
the contamination of sugar beet slices with many ions,
lithium was chosen given the sensitivity of its measure-
ments. A blank sample was tested and a calibration pre-
pared by adding known amounts of lithium chloride to
the blank solution of realistic composition. This amount
was added to the sludge flow of fresh sugar beet slices at
a rate of 43 m3/h on average. The duration of tracing was
150 seconds, while the main recycle flow was switched
off during this time. Before and after the tracer inlet, the
recycle flow was maintained at 200 m3/h.

Differing from the geometrically arranged order of
annular segments 1-2-3-4-5-6, the actually applied or-
der of feeding was 1-2-(3)-6-4-5, while during the exper-
iments the inlet to segment #3 was closed (because of
a malfunctioning valve). Accordingly, the cyclic feeding
sequence was changed to 1-2-6-4-5 over 360 seconds.

The tracer was fed into segment 4, while the samples
were taken from the effluent and discharged from seg-
ment 5 at a height of 20 m. The draft samples of 1-2 litres
in volume were filtered before being analysed.

The concentration of Li+ was measured by the ICP-
OES spectrometer at Kaposvár University.

2.2 Modelling and Simulation Methodology

For the modelling and simulation of the flow structure of
the fermenter, the methodology of Programmable Process
Structures (PPS) was applied [9, 10].

In PPS (see Fig. 2), the locally programmable struc-
ture of process models could be generated from two gen-
eral meta-prototypes and from the standardized descrip-
tion of the actual process network automatically, resulting
in the dynamic structure of unified state and transition

Hungarian Journal of Industry and Chemistry



RESIDENCE TIME DISTRIBUTION-BASED ANALYSIS 45

Figure 2: Schematic diagram of the generation of the
process model by the method of Programmable Process
Structures

elements. The prototype elements, which describe the
functionalities of the model, could also be derived from
the general meta-prototypes. The freely editable, actual
prototype programs contain symbolic input, parameter
and output variables as well as a locally executable pro-
gram code. In various applications, many state and tran-
sition elements can be modelled with the same or similar,
reusable local programs (referred to as actual prototypes).
The state and transition elements of the actual model can
be parameterized and initialized concerning their case-
specific prototypes. Its execution, namely connection-
based communication between the state and transition
elements of the programmed structure, is solved by the
general-purpose kernel program of the method.

During the simulation, the actual elements are ini-
tialized by initial conditions and parameters, moreover,
the output values are recalculated stepwise by taking into
consideration input and parameter data according to the
associated local program prototype. The identified input
and output connections of the extensive/intensive prop-
erties and signals make the combined execution of the
balance-based and signal-based functionalities possible.

3. Results and Analysis

3.1 Experiments

The quantity of the lithium chloride tracer (40 kg) and the
sampling of the effluent during the measurement of the
RTD were designed by some preliminarily studied, ap-
proximate mixing models. The concentration of lithium
chloride in the effluent was measured every 0.5−2 hours
during the first period and daily or even less frequently
over the following 23 days (when production slowed
down by ending campaign). The data are summarized in
Table 1 and in Fig. 3.

3.2 Possible Flow Structures and Parameters

By taking into consideration the geometrical layout, the
produced biogas flow and cyclically changing recycle
flow, various mixing models with different compartmen-
talization and flow structures were generated by applying
the method of Programmable Process Structures.

Table 1: Measured concentrations of lithium chloride in
the effluent

Sample ID Date Time, hours Li+, mg/l

0 Blind 0 <0.100
1 2018.12.18 0.5 <0.100
2 2018.12.18 1 0.226
3 2018.12.18 2 0.347
4 2018.12.18 4 0.311
5 2018.12.18 6 0.237
7 2018.12.18 8 0.811
8 2018.12.19 26 0.627
9 2018.12.19 38 0.538

10 2018.12.20 48 0.625
11 2018.12.22 96 0.536
12 2018.12.24 144 0.401

12B 2018.12.26 192 0.516
13 2018.12.28 240 0.423
14 2019.01.02 360 0.325
15 2019.01.04 408 0.361
16 2019.01.07 480 0.384
17 2019.01.10 552 0.323

The most detailed compartmentalization (D1) is
shown in Fig. 4. Here the sludge zone of the process unit
was divided into three vertical layers which were divided
further into an inner cylinder, surrounded by an annulus
that was divided into six parts.

The mixing flows between the 21 (approximately per-
fectly mixed) compartments are as follows:

- Circflowmix: cyclically changing recycle flow, su-
perposed by a bidirectional mixing flow between the an-
nular slices above each other;

- Gasflowmix: bidirectional mixing flow, induced by
the upward biogas flow between the compartments above

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 5000 10000 15000 20000 25000 30000 35000 40000

C
o
n

c
e
n

tr
a
ti

o
n

, 
m

g
/L

Time, hour

Measured tracer concentration in effluent 

Figure 3: Measured tracer concentration of Li+ in the ef-
fluent of the industrial-scale fermenter (40 kg of lithium
chloride dissolved in 150 litres of tap water was added to
a sludge of sugar beet slices that entered segment #4 at a
flow rate of 43 m3/h for 150 seconds).

47(2) pp. 43–51 (2019)



46 TANKOVICS, TAKÁCS, SZENDEFY, CSUKÁS, AND VARGA

Figure 4: The most detailed schematic diagram of the fer-
menter (Scheme D1).

each other (for practical considerations, circflowmix and
gasflowmix can be integrated into a single flowmix);

- Permix: peripheral bidirectional mixing flow be-
tween the subsequent annular slices of the same layer ,
horizontally;

- Radmix: radial bidirectional mixing flow between
the annular slices and the inner cylinder of the same layer,
horizontally.

The flow rate of the biogas increased as the height of
the unit increased. Accordingly, gasflowmix was higher
between layers 2 and 3 than between 1 and 2.

It is worth mentioning that the outflow rate of the ef-
fluent was considerably less than the inflow rate of the
sludge composed of pressed sugar beet slices because a
significant proportion of the raw material was converted
into biogas that escaped through the upper gas dome.

In Scheme D2 (Fig. 5 which was finally proven to be
the best solution), the inner cylinder and the related ver-
tical and horizontal mixing flows were removed because
they slowed down the mixing in Scheme D1 by an unfea-
sible degree.

As a further simplification, the upper layer was not
decomposed at all. It should be noted that an interaction
between the changes in the structures and their parame-
ters exists. As an example, Scheme D3 (Fig. 6) behaved
in a similar way to D2 with the highest horizontal mixing
flows in the upper layer. Planned future research will fo-
cus on using less compartments which might also be ad-
vantageous in terms of embedding the detailed dynamic
model of anaerobic digestion into the flow structure. Ac-
cordingly, Scheme D4 (Fig. 7) was also tested, where
only the lower layer was decomposed.

As a marginal solution, the case of Scheme D5 (Fig.
8) was also studied with only vertical decomposition of
the process unit.

Figure 5: Schematic diagram of the best performing fer-
menter (Scheme D2).

3.3 Generation of the Simulation Models

The process network of the various schemes can be de-
fined by a general declarative program. A characteristic
part of the definitions of Scheme D2 is as follows:

states([c11],[sludge]).
states([c12,[sludge]).
...
states([c36,[sludge]).
states([env,[sludge,feed]).
transitions([c11],[flowmix,permix]).
transitions([c12],[flowmix,permix]).
...
transitions([c26],[flowmix,permix]).
transitions([c31],[permix]).
transitions([c32],[permix]).

Figure 6: Simplified schematic diagram of the fermenter
without decomposition of the upper layer (Scheme D3).

Hungarian Journal of Industry and Chemistry



RESIDENCE TIME DISTRIBUTION-BASED ANALYSIS 47

Figure 7: A more simplified schematic diagram of the fer-
menter (Scheme D4).

Figure 8: Trial in the absence of horizontal compartmen-
talization (Scheme D5).

...
transitions([c35],[permix,

feeding\_recirc]).
transitions([c36],[permix]).
dcode(sludge,[li]).
dcode(feed,[li]).

where states and transitions are declared by the state and
transition elements in the given compartments, respec-
tively. Meanwhile, the lowest level components of the
state elements are described by the dcode() predicates.

The transition-based representation of the flow struc-
ture is defined by the facts of predicate:

trans(TransitionName,Compartment,
InputComponents,InputSigns,
OutputComponents,OutputSigns).

For example,

trans(flowmix,[c13],
[n([c13],[sludge]),
n([c23],[sludge])],
[n([c13],[sludge]),
n([c23],[sludge])],[],[]).

generates bidirectional connections for mixing flows be-
tween compartments [c13] and [c23] of the first and
second layers, respectively. Similarly,

trans(permix,[c32],
[n([c32],[sludge]),
n([c33],[sludge])],
[n([c32],[sludge]),
n([c33],[sludge])],[],[]).

generates peripheral flows between compartments
[c32] and [c33] in the third layer.

The initial concentrations and parameters of the vari-
ous state and transition elements were recorded in an MS
Excel file from where they were transformed into a tex-
tual form of declarative predicates.

According to Fig. 2, the automatic generation of the
Programmable Process Structures was conducted:

• from the two general prototypes [9, 10],

• from the textual description of the actual process
network, and

• from the textual declaration of initial values and pa-
rameters.

The automatically generated structural model of Scheme
D2 is illustrated in Fig. 9.

3.4 Local Programming of the Generated
Process Structure

The functionalities of the flow-structure models can be
represented by locally executable programs embedded in
the prototype elements. The prototype elements can be
prepared from the copies of the meta-prototypes, e.g. the
local program for the calculation of flowmix is presented
below:

Figure 9: Programmable Process Structure of Scheme D2.

47(2) pp. 43–51 (2019)



48 TANKOVICS, TAKÁCS, SZENDEFY, CSUKÁS, AND VARGA

{ p r o g r a m ( ’
p e r m u t a t i o n ( I n p C o n c s , [

d ( [ Coord1 , s l u d g e , l i ] , [ C1 ] , g_m3 ) ,
d ( [ Coord2 , s l u d g e , l i ] , [ C2 ] , g_m3 ) ] ) ,

p e r m u t a t i o n ( P a r a m e t e r s , [
d ( v g a s , [ Vgas ] , m3_h ) ,
d ( v c i r c , [ V c i r c ] , m3_h ) ] ) ,

g ( d t , DT ) ,

p l u s f l o w ( Coord1 , V c i r c , V p l u s ) ,
DM1 i s ( Vgas + V p l u s ) / 6 0 * ( C2−C1 ) * DT ,
DM2 i s ( Vgas + V p l u s ) / 6 0 * ( C1−C2 ) * DT ,

OutComps = [
d ( [ Coord1 , s l u d g e , l i ] , [ DM1 ] , g ) ,
d ( [ Coord2 , s l u d g e , l i ] , [ DM2 ] , g ) ] ,

O u t S i g n s = [ ] ,
R e p o r t = [ ] .

p l u s f l o w ( [ C o o r d 1 ] , V c i r c , V p l u s ) :−
s u b _ a t o m ( Coord1 , 2 , 1 , 0 , Column ) ,
g ( s e l e c t e d , Column ) , ! .

p l u s f l o w ( _ , _ , 0 ) .

’ ) }

This program illustrates how the input, parame-
ter and output data are represented by the unified
d(Identifier,List_of_Values,Dimension)
triplets.

The input data for calculations originated from the re-
spective compartments Coord1 and Coord2. The pa-
rameters define the Vgas mixing flow associated with
the gas flow and the Vplus mixing flow generated by
the recycle upflow between the given vertical segments.
The actually active segment was determined by the inte-
ger value of Column in the g(selected,Column)
global predicate (which was actualized by another local
program, responsible for the cyclic switching of the recy-
cle flow).

The auxiliary clause plusflow() considers mixing
flow Vcirc only if the recirculation is actually associ-
ated with the given compartments, otherwise the surplus
mixing is equal to zero.

The OutComps output of the program forwarded the
changes in the amount of tracing component DM1 and
DM2 in the compartments Coord1 and Coord2, respec-
tively.

3.5 Evaluation of the RTD Measurements

The mass flow of the produced biogas was calculated
from the measured total volumetric flow rate of the three
parallel operating digesters, divided according to the in-
dividually measured loads of sugar beet slices of the par-
allel lines whilst taking into consideration the measured
composition of the biogas. The slightly changing output

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 1000 2000 3000 4000 5000 6000

C
o
n

c
e
n

tr
a
ti

o
n

, 
p

p
m

Time, min

Characteristic part of Shemes D1 and D5  

Measured values

D1 scheme

D5 scheme

Figure 10: Initial period of Scheme D1 that was insuffi-
ciently mixed and Scheme D5 that was overmixed.

flow of the effluent was calculated from the changing load
and biogas mass flows.

Regarding the identification and validation of the mul-
tiple structure, it must be emphasized that this task was
underdetermined. Considering the multiple interactions
between the structures and parameters, instead of a rigor-
ous optimization procedure, a heuristic trial and error ap-
proach was applied, controlled by the main features first
and then the values of the normalized root mean square
error (NRMSE) in the refinement. The evaluation was ef-
fectively aided by monitoring the change in concentra-
tions in each compartment.

First, D1 (being insufficiently mixed) and D5 (being
overmixed) were excluded as can be seen in Fig. 10. Af-
terwards, D2-4 were studied and based on the calculated
NRMSE values, D3 and D4 were stepwise rejected. Fi-
nally, the parameters of D2 were refined.

The simulated and measured data are illustrated in
Figs. Figs. 11–14 in addition to the calculated NRMSE
values.

Fig. 10 (and many trials using other parameters)
showed that even Schemes D3-5 were unable to model
the evident initial peaks of the measurements.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 5000 10000 15000 20000 25000 30000 35000

C
o
n

c
e
n

tr
a

ti
o

n
, 

p
p

m

Time, min

Simulated effluent concentration for scheme D3 

Simulated

Measured

Figure 11: Trial using horizontal compartmentalization in
two lower layers only (Scheme D3, NRMSE = 12.10 %).

Hungarian Journal of Industry and Chemistry



RESIDENCE TIME DISTRIBUTION-BASED ANALYSIS 49

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 5000 10000 15000 20000 25000 30000 35000

C
o
n

c
e
n

tr
a
ti

o
n

, 
p

p
m

Time, min

Simulated effluent concentration for scheme D2 

Simulated

Measured

Figure 12: Trial using horizontal compartmentalization in
three layers (Scheme D2, NRMSE = 12.08 %).

The horizontal decomposition in all layers (without
an inner tube) is able to express the main features using
many parameters (an example is shown in Fig. 11). De-
composition Scheme D2 (compartment volume is 685.3
m3) with improved parameterization describes the mea-
sured data better (see Fig. 12).

The change in the concentration in the various com-
partments (except for the traced bottom and output top
ones) is shown in Fig. 13. This shows that because of the
changing input bottom compartment of the recycle flow,
the initial peak in concentration does not appear in each
compartment.

The long tail is partly a consequence of the great dif-
ference between the input feed and output effluent flows
as well as of the limited degree of mixing between the
layers and annular zones (especially in the bottom layer).
It is also worth noticing the constant difference in concen-
tration in the compartments. Unfortunately, the continu-
ation of measurements was limited by the closing end of
the campaign. Moreover, were they to continue, the sensi-
tivity of the concentration measurements would decrease
far below 0.3 ppm.

The computation time, depending on the number of

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 5000 10000 15000 20000 25000 30000 35000

C
o
n

c
e
n

tr
a
ti

o
n

, 
p

p
m

Time, min

Effluent concentration for scheme D2 (modified parameters) 

Calculated

Measured

Figure 13: Scheme D2 with a feasible parameter set
(NRMSE = 21 %).

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

0 5000 10000 15000 20000 25000 30000 35000

C
o
n

c
e
n

tr
a
ti

o
n

, 
p

p
m

Time, min

Calculated concentrations in the compartments of scheme D2 

sludge [c22] li sludge [c23] li sludge [c24] li

sludge [c25] li sludge [c26] li sludge [c21] li

sludge [c32] li sludge [c33] li sludge [c35] li

sludge [c34] li sludge [c11] li sludge [c12] li

sludge [c13] li sludge [c15] li sludge [c16] li

sludge [c36] li sludge [c31] li

Figure 14: Change in concentration in the various com-
partments of the experiment shown in Fig. 12.

compartments and the flow structure, was between 2 and
90 minutes but was typically less than 10 minutes. A large
difference in the volume of the compartments leads to
an increase in computational efforts, while an equidistant
volume distribution accelerates computation.

4. Discussion

The results prove that the fermenter is thoroughly mixed
but by no means perfectly mixed. Accordingly, detailed
modelling needs to take into consideration hydrodynam-
ics, e.g. by using mixing flows between the compartments
of the volume.

It should be emphasized that an interaction between
the fermentation model and mixing exists because mixing
is considerably enhanced by the produced biogas. In due
course, this coupling has to be taken into consideration in
the final implementation of the model.

It is a specific feature of the Residence Time Distribu-
tion of anaerobic digesters that the output liquid effluent
flow is much less than the input load of the fresh sugar
beet sludge (because of the production of biogas). This
contributes to the slowly decreasing long tail of the RTD.

By considering the approximately identified parame-
ters, the mixing conditions can be characterized by the
following features:

• In the bottom layer, no bidirectional, horizontal mix-
ing flow occurs between the annular segments. The
horizontal mixing flow between the annular seg-
ments can also be neglected in the middle layer.
However, in the intensively bubbling upper layer,
significant mixing occurs characterized by a flow
rate of 200 m3/h between the adjacent segments.

• Regardless of the very limited horizontal mixing in
the bottom and middle layers, the appropriate distri-
bution of the fluid and solid phases is achieved by
the frequent change in location of the recycle flow
(and feed) between the bottom segments.

47(2) pp. 43–51 (2019)



50 TANKOVICS, TAKÁCS, SZENDEFY, CSUKÁS, AND VARGA

• The estimated mixing flow rate of the vertical
sludge, generated by the upward biogas flow, was
150 m3/h and 300 m3/h between the bottom & mid-
dle and middle & upper layers, respectively.

• The estimated vertical mixing flow rate in the chang-
ing active sites, generated by the recycle flow, was
200 m3/h between the vertically connected segments
(i.e. the mixing ratio was 0.5).

5. Conclusions

The Kaposvár Sugar Factory of Magyar Cukor Zrt. has
developed an internationally straightforward anaerobic
fermentation technology to generate onsite used and sur-
plus energy with decreased emissions of waste by pro-
ducing fuel gas of high methane content from pressed
sugar beet slices. The development of an improved com-
puter model as well as the analysis-based development of
the technology required more detailed knowledge of the
hydrodynamic conditions found within the process unit.
The objective of this work was to measure and analyse
the RTD of the appropriately compartmentalized indus-
trial unit of 13, 000 m3 in volume.

Spatially uniform feeding and appropriate mixing was
ensured by the recycle flow, fed into the annularly placed
six subsequent bottom segments with cyclic changes in
location. The pressed sugar beet slices were also fed into
this recycle flow via a screw feeder. Mixing was enhanced
significantly by the upward increasing upstream flow of
generated biogas and three mechanical agitators. By con-
sidering the geometrical arrangement, produced biogas
flow and cyclically changing recycle flow, various mixing
models were generated with different compartmentaliza-
tion and flow structures via the method of Programmable
Process Structures.

A lithium salt-based tracer technique was applied,
while the quantity of the lithium chloride tracer and
the sampling of the effluent during the measurement of
the Residence Time Distribution were designed by some
preliminarily studied simulation models of mixing. The
lithium chloride concentration at the outlet was initially
measured every 2 hours and daily or less frequently dur-
ing longer periods. The lithium chloride concentration
was analyzed by ICP-OES.

For the simulation-based approximate identification
of the mixing model, a heuristic approach was used that
took into consideration multiple structures with chang-
ing mixing flows. A model of an advantageously smaller
number of compartments and parameters was sought
which satisfies the measured Residence Time Distribu-
tion.

The results suggest the intensive mixing of upper lev-
els with a limitedly mixed lower part that contributes to
the long tail of the RTD. The applied arrangement sup-
ports the horizontal distribution of the sugar beet slices
to be digested and the multiple groups of bacteria. More-
over, limited vertical mixing helps to avoid the elutriation

of the bacteria and undigested organic materials into the
environment.

The suggested mixing model will be combined with
the formerly elaborated model of the anaerobic fermen-
tation process that consists of 9 groups of bacteria in our
future work.

Abbreviations

ADM Anaerobic Digestion Model
CFD Computational Fluid Dynamics
ICP-OES Inductively Coupled Plasma Optical

Emission Spectroscopy
IWA International Water Association
NRMSE Normalized Root Mean Square Error
PPS Programmable Process Structures
RTD Residence Time Distribution
TS Total Solid content

Acknowledgement

The authors are grateful for the financial support from the
project EFOP-3.6.1-16-2016-00007.

REFERENCES

[1] Batstone, D.J.; Keller, J.; Newell, R.B.; New-
land, M.: Modelling anaerobic degradation of com-
plex wastewater. I: model development. Biore-
sour. Technol., 2000, 75, 67–74, DOI: 10.1016/S0960-
8524(00)00018-3

[2] Batstone, D.J.; Keller, J.; Newell, R.B.; Newland,
M.: Modelling anaerobic degradation of complex
wastewater. II: parameter estimation and validation
using slaughterhouse effluent. Bioresour. Technol.,
2000, 75, 75–85, DOI: 10.1016/S0960-8524(00)00019-5

[3] Blumensaat, F.; Keller, J.: Modelling of two-stage
anaerobic digestion using the IWA Anaerobic Di-
gestion Model No. 1 (ADM1). Water Res., 2005, 39,
171–183, DOI: 10.1016/j.watres.2004.07.024

[4] Fezzani, B.; Cheikh, R.B.: Modelling of the
mesophilic anaerobic co-digestion of olive mill
wastewater with olive mill solid waste us-
ing anaerobic digestion model No. 1 (ADM1).
Bioresour. Technol., 2008, 99, 6565–6577, DOI:
10.1016/j.biortech.2007.11.035

[5] Lauwers, J.; Appels, L.; Thompson, I.P.; Degrčve,
J.; Impe, J.F.V.; Dewil, R.: Mathematical modelling
of anaerobic digestion of biomass and waste: Power
and limitations. Prog. Energ. Combust., 2013, 39,
383–402, DOI: 10.1016/j.pecs.2013.03.003

[6] Csukás, B.; Balogh, S.: Combining Genetic Pro-
gramming with Generic Simulation Models in Evo-
lutionary Synthesis. Computers in Industry, 1998,
36, 181–197, DOI: 10.1016/S0166-3615(98)00071-2

Hungarian Journal of Industry and Chemistry

https://doi.org/10.1016/S0960-8524(00)00018-3
https://doi.org/10.1016/S0960-8524(00)00018-3
https://doi.org/10.1016/S0960-8524(00)00019-5
https://doi.org/10.1016/j.watres.2004.07.024
https://doi.org/10.1016/j.biortech.2007.11.035
https://doi.org/10.1016/j.biortech.2007.11.035
https://doi.org/10.1016/j.pecs.2013.03.003
https://doi.org/10.1016/S0166-3615(98)00071-2


RESIDENCE TIME DISTRIBUTION-BASED ANALYSIS 51

[7] Guillard, F.; Tragardh, C.: Modeling of the Per-
formance of Industrial Bioreactors with a Dy-
namic Microenvironmental Approach: A Critical
Review. Chem. Eng. Technol., 1999, 22, 187–
195, DOI: 10.1002/(SICI)1521-4125(199903)22:3<187::AID-
CEAT187>3.0.CO;2-9

[8] Jourdan, N. et al.: Compartmental Modelling
in chemical engineering: A critical review.
Chem. Eng. Sci., 2019, 210, 115–196, DOI:
doi.org/10.1016/j.ces.2019.115196

[9] Norregaard, A. et al.: Hypothesis-driven compart-
ment model for stirred bioreactors utilizing com-
putational fluid dynamics and multiple pH sen-
sors, Chem. Eng. J., 2019, 356, 161–169, DOI:
10.1016/j.cej.2018.08.191

[10] Kamarad, L.; Pohn, S.; Bochmann, G.; Harasek, M.:

Determination of mixing quality in biogas plant di-
gesters using tracer tests and Computational Fluid
Dynamics. Acta Univ. Agric. Silvic. Mendelianae
Brun., 2013, 61(5), 1269–1278, DOI: 10.11118/ac-
taun201361051269

[11] Varga, M.: Economic optimization of sustainable
complex processes (model based optimization un-
der uncertain cost parameters for industrial scale
anaerobic fermentation of sugar beet slice). PhD
thesis, Kaposvár University 2009, (in Hungarian)

[12] Csukás, B.; Varga, M.: Préselt répaszelet
anaerob fermentációján alapuló biogáz előállítás
számítógépi modellezésének előzetes vizsgálata.
2008. Témajelentés. Folyamatinformatika Kutató –
Fejlesztő Bt., Kaposvár (in Hungarian)

47(2) pp. 43–51 (2019)

https://doi.org/10.1002/(SICI)1521-4125(199903)22:3<187::AID-CEAT187>3.0.CO;2-9
https://doi.org/10.1002/(SICI)1521-4125(199903)22:3<187::AID-CEAT187>3.0.CO;2-9
https://doi.org/doi.org/10.1016/j.ces.2019.115196
https://doi.org/doi.org/10.1016/j.ces.2019.115196
https://doi.org/10.1016/j.cej.2018.08.191
https://doi.org/10.1016/j.cej.2018.08.191
https://doi.org/10.11118/actaun201361051269
https://doi.org/10.11118/actaun201361051269

	Introduction
	Experimental work
	Tracing and Measurement Technique
	Modelling and Simulation Methodology

	 Results and Analysis
	 Experiments
	 Possible Flow Structures and Parameters
	 Generation of the Simulation Models
	 Local Programming of the Generated Process Structure
	Evaluation of the RTD Measurements

	 Discussion
	Conclusions