63

Chapter 3

Parameter estimation and

validation in Batch settling column
by coupling DAKOTA toolkit and
OpenFOAM

3.1 Introduction

In order to build a robust SST model, it is necessary to estimate the values of the
parameters of the hindered and compression settling functions. Batch settling test
are the most widespread method to evaluate settling velocities (Ramin et al., 2014a,
Griborio, 2004).

For this purpose, a mathematical algorithm is performed to evaluate an objective

function. The objective function qualifies the output/results of a certain variable of
the model and compare it to the physical measurements of the same variable.

In the well-posed 1D Bürger-Diehl framework (Bürger, Diehl, and Nopens, 2011),

used for activated sludge settling modeling, one can choose among different con-
stitutive functions for hindered and compression settling description. This is where
calibration becomes complex, due one can have different numbers of parameters
according to the chosen functions.

Different optimization processes can be carried out to find the correct set of param-
eters. Most of them are a global optimization algorithms.

Torfs et al. (2013) performed a Global Sensitivity Analysis (GSA) using a Brute Force
Monte Carlo method to calibrate 2 hindered settling models, and 1 compression
model. They found that parameters of the Vesilind (equation 1.6, Vesilind, 1968)
function are identifiable while one of the parameters of the Takacs equation (equa-
tion 1.7 Takács, Patry, and Nolasco, 1991) is unable to be identifiable. The compres-
sion parameters of the simplified function of DeClercq (equation 1.13, De Clercq et
64

al., 2008) are not able to describe different batch settling curves with a unique set
of parameters (Torfs et al., 2013). In summary, no unique set of parameters can be
found for the combined equations for hindered and compression settling.

Locatelli (2015) used an automatic differentiation tool to model sludge settling. He

chose a Vesilind-DeClercq (equations 1.6 and 2.5) coupling for hindered and com-
pression settling respectively in the Bürger-Diehl (Bürger, Diehl, and Nopens, 2011)
approach. Those four parameters have an important effect on the prediction of the
simulated sludge blanket height. The model with the calibrated parameters can be
accurate only within the three first hours of settling, beyond this time the SBH is
overestimated. Seven different parameter sets were tested and the quality of the es-
timation was measured by visually comparing the simulated and observed sludge
blanket height.

The HTC model (equation 1.15) calculates the solids flux in a domain of 60 horizontal
layers to represent the settling behavior in a batch column. Ramin et al. (2014a) used
only 3 parameters to estimate. The global calibration optimization method selected
was Markov Chain Monte Carlo (MCMC), in which a large number of iterations
(sometimes more than 100000) are made in order to find the best parameters set to fit
the experimental data. They reported no identification issues for the three calibrated
parameters. Even if MCMC is an accurate method, the main disadvantage of the
method is that it requires a large number of simulations making it computationally
intensive.

These calibration methodologies are usually performed using 1D models and the
associated numerical methods. The identified parameters could then be used within
a CFD simulation, usually performed by another software package with different
numerical methods.

One of the advantages of the OpenFOAM platform is that the same solver can be
used to perform 3D, 2D-axisymmetric, 2D or even 1D simulations. In practice, the
user has always to provide a 3D mesh but by defining specific "empty" or "wedge"
boundary conditions into the mesh, the solver adapts the numerical procedure to
the dimensions of the system. Therefore, the same solver and numerical methods
can be used for parameter estimation and validation in a batch settling column as
well as for simulation of a full-size clarifier. This makes the process easier and more
reliable.

DAKOTA® (https://dakota.sandia.gov/) is an open-source (GNU LGPL license)

toolkit that provides an interface between simulation codes and a variety of itera-
tive systems analysis methods, including optimization, uncertainty quantification...
It has applications spanning defense programs for climate modeling, computational
materials, nuclear power, renewable energy, and many others. DAKOTA® can there-
fore be coupled to OpenFOAM in order to perform the identification of the best set
of parameters for the settling model developped in chapter 2.
 65

The objective of this chapter is to show a local process of calibration and validation
of the settling model developed in OpenFOAM (2) using the sludge blanket height
(SBH) experimental information obtained in a batch column and the simulated SBH.
Calibration is made by using a 1D mesh created in OpenFOAM.

This study is using a local optimization method: this allows to lower the compu-
tational cost (as a CFD code is used). In addition, not enough experimental points
were available to perform a global analysis.

In order to illustrate this calibration and validation procedure, two sets of experi-
mental data were chosen, due to the extremely varying characteristics of the sludge
in the studied treatment plant. According to the number of experimental data points
available, a different settling model (hindered and compression functions) is chosen
for calibration. Indeed, the settling functions assessed in this study involve a dif-
ferent number of parameters. The choice of different functions therefore allowed to
adjust the degrees of freedom of the calibration process.

A non-linear squares method is performed in order to minimize an objective func-

tion. In such method, the model equation is fitted to the experimental data rather
than transforming it into a linear form (Sagnella, 1985). Estimation consist in finding
the optimal values for the parameters that can minimize a cost function or Quantity
of Interest (Qol). The cost function, often described as the sum of the squared errors
between the observed data and the simulated data, is minimized with respect to the
parameters.

In this approach, we will evaluate the quality/accuracy of the model outputs by

introducing a statistic called Nash Sutcliffe Efficiency (NSE). Hence, the NSE statis-
tic can be decomposed in three components: correlation, the bias and the relative
variability in the simulated and observed values to show the inherent systematic
problems due to calibration (Gupta et al., 2009).

Finally, validation of the model is carried out by using another set of batch sludge
settling experimental data and the settling velocities, and evaluating it through the
NSE statistic. With the found values of the settling parameters, the CFD model will
be validated in a full-scale clarifier.

3.2 Materials and Methods

3.2.1 Sludge settling Model

Equation 2.17 is the model of interest. Such model can be used for 1, 2 or 3 dimen-
sions. As stated before, this model consists of 1 general PDE, where one can choose
66

among different hindered settling velocities (~υhs ) and effective solids stress ( σe ) func-
tions to describe the settling behavior. Two different sets of experimental data are
used to calibrate the model, thus the model 2.17 is used in two configurations:

• Vesilind equation (Vesilind, 1968), and the simplified function of DeClercq for
effective solids stress (equation 3.1).

• Diehl equation (equation 2.20) and simplified function of DeClercq for effective
solids stress (equation 3.1).


0 for 0 ≤ αd < Xcrit /ρd
σe (αd ) = (3.1)
λ(α − Xcrit
for αd ≥ Xcrit /ρd
d ρd )

In the OpenFOAM code, equation 2.2 is expressed as the volume fraction of the
dispersed phase (αd ) and not the sludge concentration (X) (see chapter 2).

Parameter λ indicates the magnitude in which the settling velocity is slowed down.
The compression function is calculated with equation 2.18, but is only active when
the sludge volume fraction crosses the critical volume fraction defined as Xcrit /ρd .

3.2.2 Experimental Test on Batch settling column

For the calibration analysis different sludge samples were grabbed at different sea-
sons (January to October 2018) from the aerated tank of the WRRF of Achenheim
(WRRF details in chapter 4). Batch settling tests were performed in order to obtain
the settling velocities and the sludge blanket heights using an Ultrasonic Transducer
device developed by ICube laboratory (Abda et al., 2009; François et al., 2016; Pal-
lares et al., 2017).

The batch settling column consisted of a cylindrical transparent PEXIGLAS column

of 1m height and 0.4m diameter with a total volume of 126 liters (figure 3.1 right).
Prior to sedimentation test, the sludge mixture was homogenized by a reversible
pump system working in closed-loop. Air bubbles must be removed from the en-
tire column, since they can disturb the velocity measurements. Wall effects can be
neglected due to the large diameter of the cylinder.

Determination of the sludge properties

Prior to data acquisition, the homogeneous activated sludge was sampled to deter-
mine the TSS, Volatile Suspended Solids (VSS) and sludge densities. The TSS and
the VSS were determined following the procedure AFNOR NFT 90-105, 1997 (Qual-
ité de l’eau - Dosage des matières en suspension). The sludge floc density was measured
by pycnometer, centrifuging the bulk sludge following the procedure described in
Locatelli (2015). The properties were measured three times for each dilution.
 67

F IGURE 3.1: Ultrasonic transducer device position (left) and Exper-

imental batch column with settled sludge from January 16th 2018
(right)

Table 3.1 presents a summary of the sludge properties for all the batch tests from Jan-
uary to October (without dilution). The sludge floc density is an intrinsic property
that is not affected by the dilution.

TABLE 3.1: Sludge properties of the Achenheim WRRF

Date TSS (kg.m−3 ) VSS (kg.m−3 ) NVSS (%) Density (kg.m−3 )

January 4.16 ± 0.06 2.63 ± 0.03 36.68 ± 1.52 1018.2 ± 5.9
February 5.39 ± 0.03 3.32 ± 0.03 38.44 ± 0.48 1000.2 ± 2.0
April † 5.54 ± 0.020 3.41 ± 0.007 38.49 ± 0.10 1010.4 ± 4.2
August ‡ 2.84 ± 0.003 1.18 ± 0.003 58.57 ± 0.49 1012.7 ± 0.8
October † 4.55 ± 0.03 2.05 ± 0.07 54.88 ± 1.39 1001.1 ± 1.6
† Dilutions were made during these tests
‡ Not a representative value, maintenance at the WRRF was done during the
measurements.

Ultrasonic transducer description

The data obtained from the Ultrasonic transducer device is based on the Doppler ef-
fect. The transducer sends an acoustic signal to the activated sludge suspension and
it receives back an echo called backscattered signal. From this signal, the position and
velocity of a particle (scatterer) can be known. This is a non-invansive technique for
sludge settling velocity recording. The device is placed over the surface of settling
column (see figure 3.1 left) to perform vertical measurements.

The transducer emits a Pulse Repetition Frequency (PRF) at an emission frequency

( f 0 ), then the acoustic beam will be propagated in a conical shape measurement
68

TABLE 3.2: Properties and parameters of the ultrasonic transducer,

can be fixed to the desired measures.

Sensor diameter 1 cm
Maximum Pulse Repetition Frequency + 300 Hz
Maximum Pulse Repetition Frequencye f f * 15 Hz
Maximum measurable velocity 0.001 m/s
Central frequency 1.565 MHz
Carrier frequency 1.5-1.25 MHz
Spatial Resolution 2 mm
Time Resolution 0.66 s
+ Number of ultrasonic bursts emitted per second
* Number of emitted ultrasonic bursts recorded per second

cell (spatial resolution). After emitting the ultrasonic pulse, the electronic system
switches to “receiving” mode. The acoustic wave propagates along the beam axis
and each scatter that crosses the beam will diffuse an echo towards the transducer.
The backscattered signal (echo) is composed, at a given time, by the sum of the echoes
of all scatterers located in the measurement cell. The scatters/particles will induce
a frequency shift in the backscattered signal, i.e., the Doppler shift f D . In each cell,
the information obtained from several pulses is processed by the instrument to es-
timate the projection of the velocity vector over the beam axis. The features of the
Ultrasonic device are summarized in table 3.2.

To get the information for the sludge blanket height, the values of amplitude are
needed. The strength of the backscattered signal increases sharply giving a higher
value in the amplitude signal. Thus, the amplitude will show a high value when the
ultrasound burst reach the top of the sludge blanket (François et al., 2016). Higher
the backscattered intensity is, higher the particle concentration (Thorne and Hanes,
2002). Therefore, the interface of the clear water and the sludge or SBH, can be
estimated by relating the height where the sharp intensity jump of the amplitude
sign was found.

To obtain the settling velocity of the particles ~υs within the sludge blanket, the Doppler
shift frequency ( f D ) is needed, equation 3.2 correlates both variables:

C fD
~υs = (3.2)
2 fc

Where C is the sound speed and f c the carrier frequency (Hz). The settling velocity
can be obtained by measuring f D at different depths. A variation of 1°C in water
temperature will produce a variation of 0.1 and 0.3 % in the particle velocity and cell
position. Thus, a correction by the speed of sound at 20°C has to be done.

The average time for velocities and amplitude data acquisition was 1 hour. Separate
tests were also carried out for more than 5 hours. Particles settling velocity and the
amplitude (backscattered intensity) of the signal were recorded every 0.15 sec.
70

The Courant-Friedrichs-Lewy condition (CFL) condition (equation 3.3) is fixed to

0.9, to ensure the convergence of the simulation. This condition assures the informa-
tion of a cell traveling directly to its neighbor and do not skip it.

∆t
CFL = d (3.3)
∆z

Where: d is the velocity magnitude, ∆T is the time step and ∆z is the length between
mesh elements.

3.2.4 DAKOTA® optimization process

DAKOTA® is a software developed by U.S. Sandia National Labs and stands for De-
sign and Analysis toolkit for Optimization and Terascale Application. This software
allows for model optimization, uncertainty quantification, parameter study and de-
sign of experiments. It can be used with its own syntax or it can be coupled to an
external software (acting like a black box tool) to perform the optimization process.

Optimization process with DAKOTA®

DAKOTA® can perform an optimization process just by setting-up an input file,

with the experimental and model output information, and choosing the optimiza-
tion model/code.

Dakota has different optimization methods (gradient-based, efficient global opti-

mization, nonlinear least squares, etc.) to find the response for the objective func-
tion. The model to be optimized can be set directly into DAKOTA® environment or
it can be linked to another software (see next section). One can set the initial guess
and domain of the values of the parameters where the optimization method should
seek. Finally, a series of calculations are performed to obtain the objective function
and make a new iteration (if necessary) with a new set of parameters automatically
calculated by DAKOTA®.

Figure 3.3 illustrates the closed-loop of an optimization process using DAKOTA en-
vironment. DAKOTA already contains algorithms to perform an estimation process,
such as the Rosenbrock function, which is a test problem used to evaluate the charac-
teristics of an optimization process (convergence, precision).

Optimization process coupling DAKOTA® and OpenFOAM®

Coupling DAKOTA and OpenFOAM® mainly consists in running an OpenFOAM

solver in order to get a response or output, in which later, DAKOTA will read to
analyze the response. If the response has not a desirable value then DAKOTA will
 73

Optimization problems through the NLS, involve the minimization of an objective

function f(x) such as:

n
f (x) = ∑ [ Hsim,i (θ ) − Hobs,i ]2 (3.4)
i =1

This objective function f ( x ), called sum Sum of Squared Errors (SSE), is defined as
the difference between experimental data (Hobs ) and the model predictions (Hsim ) in
a particular location or time step.

Here, θ would be the set of the model parameters that are being calibrated.

The NL2SOL gradient based method

The NL2SOL method looks for a local minimum, the gradient acts as a compass and
always points downhill. The method starts with an initial guess of the parameters
values and then moves to a set of parameters that can minimize the cost function.
An iteration process is created to improve the performance of such initial guess to
reach the best fit.

To illustrate this problem, only the compression parameters: λ and Xcrit are con-
sidered, but the calibration process include also the hindered settling parameters
(equations 2.2 and 2.20). With the initial guess, the model function is expressed as a
Taylor series expansion near the initial values (λ0 and Xcrit0 ), the expansion can be
represented in a linear parameter equation:

df df
Yi − f (λ, Xcrit · t) = f (λ0 , Xcrit0 , ·ti ) + ( λ − λ0 ) + ( Xcrit − Xcrit0 ) (3.5)
dλ dXcrit

Where Tsim is the independent experimental variable assumed to be free of error for
n pairs of data points and Yi the experimental value. Values can be obtained for
(λ − λ0 ) and ( Xcrit − Xcrit0 ) by solving the difference between the experimental and
predicted values with the initial parameter guess.

The new obtained values are used for the next initial guess and repeated several
times. In each iteration the SSE (equation 3.4) is calculated and evaluated until a
converging criterion is achieved. Convergence is reached when the change in the
value of the SSE from one iteration to another is below a tolerance value (here a
value of 10−5 is used).

Thus, the NL2Sol algorithm in Dakota®, uses the Gauss-Newton gradient method to
minimize the cost function. Assuming that the first derivative of [ Hsim (θ ) − Hobs ]2
tends to zero and thus the Hessian matrix of the second derivative of f ( x) can be
74

approximated by using only the first derivative of [ Hsim (θ ) − Hobs ]2 . This approxi-
mation is corrected by a secant update method.

Confidence intervals on estimation

The 95 % confidence intervals are computed as the optimal value of the estimated
parameters ± a t-test statistic times the standard error of the estimated parameter
vector. The standard error is a linearization involving the matrix of the derivatives
of the model with respect to the derivatives of the estimated parameters (Adams et
al., 2014).

3.2.6 Evaluating the estimated parameters

The Nash-Sutcliff Efficiency (NSE) normalized statistic evaluates the quality of the
estimation by determining the relative magnitude of the residual variance compared
to the measured data variance, thus it is expressed as:

n
2
∑ ( Hiobs − Hisim )
i =1
NSE = 1 − n (3.6)
2
b obs )
∑ ( Hiobs − H
i =1

bi,obs is the mean of the measured data for the constituent being evaluated.
Where H

NSE can range from ∞ to 1. Moriasi D. N. et al. (2015) developed this evaluation
criteria based on measures for watershed hydrological modeling (nutriments, flow
and sediments). To determine the quality of the calibrated parameters one can take
such criteria for the NSE, as it is shown in table 3.3

TABLE 3.3: Evaluation criteria for estimated parameters

Criteria NSE Value

Very Good > 0.8
Good 0.6 ≤ NSE ≤ 0.8
Satisfactory 0.5 < NSE < 0.6
Not Satisfactory ≤ 0.5
78

is used to determine the limits of the hindered zone. With a linear regression, the
values of the slopes of the settling curves are obtained. The hindered velocity values
are shown in table 3.6.
TABLE 3.6: Measured settling velocities at different sludge concentra-
tions with the samples taken in October

Concentration (Kg·m−3 ) ~υhse (m·s−1 )

2.08 1.765.10−3
3.05 1.089.10−4
4.54 6.817.10−4

Comparing the velocity values at similar concentrations, (for the same WRRF sludge
but in different season), it results that velocity for X0 = 4.54 Kg·m−3 is 20 times
higher than the velocity for the sludge at X0 = 4.62 Kg·m−3 . This increase in the
velocity can be due to the fact that the sludge of October has a higher percentage of
NVSS.

For this experimental data, the calibration/validation process is made with configu-
ration two (section 3.2.1), i.e., the coupling of equation 2.20 and equation 3.1. Thus,
5 parameters are calibrated. The initial guess for the values are listed in table 3.7.

TABLE 3.7: Initial guess of the parameters and their limits

Parameter Initial Guess Upper bound Lower bound

~υ0 (m·s−1 ) 0.009 0.1 0.001
q 1.5 2 0.5
X̄ (Kg·m−3 ) 0.7 1 0.1
λ ( m2 · s − 2 ) 0.02 0.01 0.001
Xcrit (Kg·m−3 ) 5 6 4.6

3.3.2 Calibration and Validation on sludge blanket height

April data

Calibration process for the complete settling curve, (estimation of the 4 settling pa-
rameters at the same time) was done taking the results of the measured sludge blan-
ket at the initial concentration of X0 = 3.95 Kg·m−3 for one hour of settling. The
simulated SBH is where a high volume fraction gradient is observed. In the simu-
lation results, the value of such height was obtained at the layer where the sludge
concentration is X = 0.9 Kg·m−3 . The values of the estimated settling parameters
and their confidence intervals are shown in table 3.8.

Calculating the NSE for the complete settling curve yields a value of 0.995 which
indicates a good quality of the estimated parameters according to Moriasi D. N. et
al. (2015) criteria. If the same statistic is applied separately for both zones, hindered
and compression, values of 0.987 and 0.945 are obtained respectively. Thus, even if
80

a local minimum of the cost function is found, the estimated values perform a good
simulation of the batch settling behavior within the first hour of settling. Hence, the
low uncertainty (red dotted lines in figure 3.8a) of the parameters ensures the good
definition of the estimation process.

The log-log plot of the experimental data (figure 3.8b) shows a linear trend which
indicates that the SBH moves according to a power law. The simulated SBH shows a
marked breakdown around 1800 s caused by the constant sludge critical concentra-
tion. Hence, after 4200 s, the simulated SBH does not follow a linear trend, suggest-
ing that compression parameters should vary with time.

To validate the 1D model the estimated parameters are used to model batch settling
behavior at two different initial concentrations. The estimated parameters seem to
reproduce the same behavior as the experimental data within the hindered zone
at X0 = 4.62 Kg·m−3 (figure 3.9a). Within the first hour the model can predicts
accurately the settling behavior. After this time, the models over predicts the sludge
height, indicating again that compression parameters should vary with time (figure
3.9b).

Validation at the initial concentration of X0 = 5.54 Kg·m−3 is not included due to the
large lag-phase during the settling experiment where it is complex to determine a
true hindered settling velocity. Within this experiment, the compression parameters
are not truly validated in batch column due to the lack of measurement points. But
for hindered settling zone the model makes a good agreement with experimental
data when using the estimated parameters.

October data

Calibration process was done in the same manner as in the previous campaign but
using a Diehl’s function (equation 2.20) for hindered settling modeling. Thus, the
estimation is made simultaneously for 5 parameters. The values of the estimated
settling parameters and their confidence intervals are reported in table 3.9. The ini-
tial concentration was set to X0 = 4.54 Kg·m−3

TABLE 3.9: Values for the estimated settling parameters and their
confidence intervals for October data

Parameter Estimated Value Confidence intervals

v 0 ( m · s−1 ) 0.00926 ±1.57.10−4
q 1.36 ±0.01
X̄ (Kg·m−3 ) 0.7146 ±0.0105
λ(m2 ·s−2 ) 0.01671 ±2.00.10−4
Xcrit (Kg·m−3 ) 4.6 ±0.0992

Calculating the NSE for the complete settling curve yields to a value of 0.982 which
indicates a good quality of the estimated parameters. The NSE for the hindered and
84

sludge blanket for these additional initial concentrations. The parameters of the
hindered settling can accurately describe the settling zone.

The log-log plot (figure 3.11b) illustrates again the power law behavior of the settling
curves, and that the estimated parameters can accurately describe the hindered zone
marked by the curvy trend of the line. The model can also reproduce accurately the
compression curve within the first hour. However, the trend of the curve is to remain
constant with time once in the compression regime.

For this sludge, it is observed that a unique set of compression parameters can accu-
rately describe the sludge blanket height during the first settling hour for all sludge
concentrations. However those parameters fails to predict an accurate SBH after one
hour. This unsuccessful prediction is expected as the compression parameters do not
change in time.

Ramin et al. (2014a) introduced a model with a varying critical concentration which
is estimated in function of the initial concentration. Even if their model seems to
overcome this problem, the approximation of Xcrit being dependant of the initial
concentration is not feasible in a full-size clarifier where the initial concentration is
not know and a sludge blanket height is physically hard to track. (Refer to chapter 4
to see the problems when measuring the sludge blanket height in a full SST).

Locatelli (2015) implemented a time varying critical concentration model. The time
dependant critical concentration was described with an empirical equation. The
sludge blanket model predictions were successful for different parameters sets, but
only within the first 3 hours of batch settling.

3.3.3 Validation on settling velocities for October data

The validation on the settling velocity in column presented here, only concerns the
initial concentration at X0 = 4.54 Kg·m−3 at two different times. The experimental
velocity profiles are compared to the calibrated model, using the found values, and
a model that do not use the compression function. The model without compres-
sion function employs the same calibrated parameters for the Diehl function used to
simulate the hindered settling.

Both models do not follow the distribution of the settling velocities within the dis-
crete and hindered zone, hence the ending of the hindered zone (represented by the
vertical continuous line on figure 3.12-right) is overestimated. This is expected, the
models do not calculate the individual particles velocities and thus the distribution
is impossible to predict including the distribution in the hindered zone (Locatelli,
2015, Torfs et al., 2017). Only in the bottom of the column the model with the com-
pression function predicts accurately the settling velocities, where such velocities
mostly depend on the local sludge concentration.
 85

However, after 44 minutes of settling (figure 3.12 right). The model do not predict
a linear behavior of the velocities within the compression zone, and a discontinu-
ity between the clear zone and the compression zone is created. This is due to the
constant critical concentration through time.

Despite the fact that the prediction of the individual settling velocities are not in-
cluded in both models, the model including the compression function, estimates
better settling velocities when compared to a model that do not use the compression
function (figure 3.12 black dotted line).

To evaluate the accuracy of the model, the Mean Absolute Error (MAE) is obtained,
lower is the value closer the predicted values are to the measured ones. Such statistic
measures how close the predicted values are to the measured ones by summing the
absolute differences of the values, i.e.

n
∑ |~υsim − ~υobs |
i =1
MAE = (3.7)
n

where: ~υsim and ~υobs are the simulated and measured settling velocities respectively.
At 5 minutes, the model within the compression function, shows a MAE of 3.68x10−4
within the compression part. Compared to the MAE (8.75x10−3 ) of the model with-
out compression function, the latter is clearly less accurate when predicting the par-
ticles settling velocities within a batch column.

The same evaluation made with the velocity profiles at 44 minutes is performed.
The MAE are 3.15x10−4 and 8.58x10−3 for the model with compression and without
compression function respectively. The model including the compression function is
still more accurate for the predictions of the settling velocities even at longer times.

3.4 Conclusions

DAKOTA software was successfully coupled to a 1D model performed in Open-

FOAM®. A local minimum gradient based method for parameter optimization was
used. This allows to have a less computationally intensive process and obtain accu-
rate values for the estimated parameters. DAKOTA constitutes a powerful tool that
can be used also to perform a global optimization process opening a new opportu-
nity for developers to study deeply the compression parameters in the OpenFOAM
model.

The two calibration processes were made with different couples of settling functions
(hindered + compression): this was because there was a lack of experimental points
for some datasets. Calibration performed with equations of Vesilind (1968) and sim-
plified form for σe of Declercq was chosen because only two sets of experimental
 87

velocity (higher is the concentration, lower is the settling velocity). The settling ve-
locity is highly affected by the content in NVSS. Faster velocities were found when
the content of NVSS was higher at similar MLSS concentrations. This behavior lead
us to realize two calibration/validation processes with the sludge of Achenheim
WRRF.

Identification problems can occur within the model, due to the fact that values for
the optimized parameters depend on the initial guess. This is because a gradient
based method tries to find a local minimum quantity of interest or cost function.
However, the estimated parameters showed a good accuracy, the quality has been
measured through the NSE statistic for the different settling zones and the complete
settling curve, the values showed a very good evaluation (NSE > 0.75). Using a
global sensitivity analysis was not within the scope of this study.

The flocculation state is a possible phenomenon that can explain the time-dependant
compression parameters, this was included in a multiclass 1D settling model by
Torfs et al. (2016). However, the same methodology would be hardly compatible
with a drift-flux approach in CFD since to this stage there is no code to model the
velocity of the different class of particles.

Nevertheless, it was decided to use the estimated parameters from the batch settling
tests and 1D simulations presented here to simulate the intermittent operation of
the full-scale WRRF (clarifier feeding and recirculation every 20 minutes). They can
predict an accurate sludge blanket height within the first hour of settling which is in
the order of magnitude of the sludge residence time within the clarifier.
 89

Chapter 4

Hydrodynamic study of the

full-scale Clarifier

4.1 Introduction

Within suspended growth biological processes, Secondary Settling Tanks (SST) must
achieve sludge-water separation, biomass recycling and storage in case of hydraulic
overloading. Hence, SST govern effluent quality in terms of suspended solids and
indirectly in terms of biokinetic processes (Torfs et al., 2015a). Modelling this unit
process is therefore essential to achieve an optimal operation of WRRF.

When using 1D models based on the Bürger-Diehl framework (Locatelli, 2015) (Torfs
et al., 2017) to describe sludge settling behaviour, adding a compression function re-
sulted in improved predictions of sludge blanket height in a settling column. In Torfs
et al. (2015a), a 1D continuous flow simulation of a conventional activated sludge
process is performed. It reveals that adding compression as constitutive function,
greatly improves the sludge blanket height prediction when high loads are present
into the clarifier, hence predicting a more realistic sludge concentration in the bio-
logical reactor.

Since the 70’s, CFD has been used increasingly for analysis and design of water and
wastewater treatment. Its use for mass transport modelling was visualized 20 years
ago (Samstag et al., 2012). The advantage of such approach is that one can have an
insight of the internal behavior of the tank. Secondary sedimentation was one of
the first unit processes to be modelled using Computational Fluid Dynamics (CFD)
(Samstag et al., 2016). However, to date, few CFDstudies incorporate a mechanistic
compression equation to describe settling tanks and no validation of such a model
exists at full-scale.

To optimize SST geometry and operation, CFD modelling is of great interest as it

allows to capture the complex hydrodynamics within the clarifier. Even if it is con-
sidered that particles settling and the arriving flow is constant and uniform, regions
90

with high circulation exist and the flow field deviates from ideal uniform distribu-
tion (Tamayol, Firoozabadi, and Ashjari, 2010).

Some WRRF often operate with an intermittent flow, i.e. the inlet and recirculation
flows are discontinuous, and depend on the period of the day. This is the case for
small WRRFs, e.g., in France, the 95% of the WRRF based on CAS technology have a
capacity less than 9000 People Equivalent.

In small WRRF the inlet flow is controlled by a pumping station operating with
on/off control according to the water level in the sump. Hence, the recirculation
pump often also works discontinuously: the return activated sludge RAS flow-rate
is constant but the pump operates only several minutes (5 to 30 minutes) per hour
according to the inlet and RAS ratio. Thus, the discontinuous feeding and extraction
is likely to impact both sludge blanket height and sludge inventory in the settling
tank.

The objective of this chapter is to employ the CFD solver described in chapter 2 (Valle
Medina and Laurent, 2020) including: hindered settling and mechanistic compres-
sion equations to simulate a full-scale SST operating discontinuously.

The small Achenheim WRRF, has a sequential flow, i.e., depending on the upstream
flow coming from the combined sewer network, level sensors will trigger one or two
pumps to feed the tanks. Water is fed when the level on the pump station rises 1.3m
and it stop when level is up to 0.8m.

Therefore, this intermittent behavior may affect the prediction of the sludge blanket
height as well as the RAS concentration and the quality of the ESS. However, the
quality of the ESS is not included in this approach since the model (described before
in chapter 2) does not include the discrete settling modeling.

For this hydrodynamic simulation, the conditions of April and October 2018 experi-
mental campaigns are presented. These experimental campaigns were held in order
to gather data about the sludge blanket height and particles settling velocity.

The simulations with different sludge concentrations and boundary conditions were
made to validate the data obtained in April and October. Thus, the cases were set as
follows:

• April. First, a simulation with constant inflow (Qin ) and RAS flow (Qr ) was
performed until the simulation reached the convergence. Departing from this
state, another simulation with dynamic inflow and RAS flow during 24 hours
was performed. Validation is made at different radial positions, only during
the time were the experimental data was done, i.e. from 9:00-11:00 am.

• October. Also, a first simulation was performed to set the basis of the dy-
namic scenario, such simulation is finished when a constant SBH and RAS are
92

chemical phosphorus removal with iron chloride (FeCl3 ). Wastewater comes from a
combined sewage network of domestic places and some wineries.

The process of wastewater treatment is as follows:

• Wastewater arrives first to a pumping station where level sensors control pumps
operation based on on/off control.

• Then, water passes to an automatic screening to retain solids larger than 1cm.

• Water is conducted to a dissolved air flotation tank for grit and greases re-
moval.

• Wastewater arrives to a cylindrical biological reactor composed of two zones:

one external aerobic zone for carbon and nitrogen removal (alternate aeration
for nitrification/denitrification), and an internal anaerobic zone for enhanced
biological phosphorus removal . Iron chloride is added to the mixture for PO43−
precipitation.

• The mixed liquor suspension passes through a degasser before entering the
secondary clarifier

• Treated water is finally released in the receiving aquatic medium which is the
Bruche canal.

Effluent quality limits should fulfill the Arreté du 21 juillet 2015, which is the French
national regulation, and should also fulfill the local regulation for discharging the
treated wastewater into the water sources (see table 4.1).

The dimensions and design parameters of Achenheim WRRF are listed on table 4.2.
Image 4.2 shows an aerial view of the Achenheim WRFF. The red circle encloses the
secondary settling tank. The yellow circle confines the biological treatment.

TABLE 4.1: French national and local effluent quality limits concen-
trations. All limits are in mg·L−1

Biological Oxygen Demand BOD5 25

Chemical Oxygen Demand COD 100
Total Suspended Solids TSS 30
Ammonium N − NH4 10
Total Phosphorus 2

Achenheim Clarifier

The cylindro-conical clarifier is 21.4m of diameter, the height at the center and near
the external walls is 3.8m and 3m respectively (Figure 4.3). The inflow (Qin ), enters
through a center feed well at design average velocity of 0.0109m·s−1 . The inlet is
limited by an internal baffle in order to reduce the velocity and turbulence of the
mixture to allow quiescent conditions. An external baffle is placed near the outlet to
 93

TABLE 4.2: Achenheim WRRF design parameters and biological re-

actor dimensions

Maximum Peak flow in dry weather 165 m3 ·h−1

Maximum Peak flow in wet weather 220 m3 · h − 1
Average flow in dry weather 114 m3 · h − 1
COD maximum load capacity 1022 kg·d−1
Volume of the aerobic tank 3085 m3
Volume of the anaerobic and contact tank 375 m3
Hydraulic retention time in the biological tanks 53.2 h
Sludge concentration in biological treatment 4−6 Kg·m−3

F IGURE 4.2: Aerial view of the Achenheim WRRF

94

avoid floating particles to escape into the treated stream. The sludge at the bottom of
the clarifier is conducted to the center by a scrapper (which velocity is 3 revolutions
per hour). The RAS flow (Qr ) is designed to be 1.15 times Qin . The design variables
of the clarifier are listed in table 4.3
TABLE 4.3: Clarifier dimensions and design parameters

Clarifier volume 1200 m3

Maximum peak RAS flow rate in dry weather 190 m3 ·h−1
Average RAS flow rate in dry weather 130 m3 ·h−1
Diameter 21.4 m
Surface area 330 m2
Design Hydraulic retention time 3.87 h
Maximum peak surface overflow rate 0.61 m · h−1
Average peak surface overflow rate 0.31 m · h−1

4.2.2 Experimental campaigns carried on SST

Several experimental campaigns were carried out from January to October 2018 in
order to study the behavior of the sludge blanket and the settling velocities inside
the clarifier. The measurement campaigns can be divided in two arrangements:

1. Punctual measurements

2. Continuous measurements.

Punctual measurements

Four experimental campaigns: one in January, one in February, one in April and one
in August were done using the following methodology:

For each campaign the same ultrasonic transducer used for batch experiments (chap-
ter 3), was employed to track the vertical velocity and amplitude of the particles in
the clarifier. The ultrasonic transducer device was placed, vertically, over the surface
of the water at three radial distances: 3.5m, 7.4m and 8.6m from the inlet.

To perform the measurements at 7.4m and 8.6m, the ultrasonic device was held on
a vertical aluminum tube which is fixed to 4m long horizontal tube, we will call this
ensemble “big T”. The horizontal tube was fixed as well, to a heavy base standing
outside of the clarifier (figure 4.4). The length of the horizontal tube was adjustable.
The “big T” was pushed manually into the clarifier just right after the skimmer of
the clarifier has passed by. Once the “big T” stopped shaking from the former move-
ment, the data acquisition began. The acquisition time lasted around 20 minutes
which is the time for the skimmer to perform a complete tour of the clarifier. In this
manner, the settling process is not perturbed by a sudden stop of the skimmer.
 99

F IGURE 4.7: Layout of the Peacok UVP system placed in the skimmer
of the clarifier

4.2.3 Sludge rheology measurements

Sludge rheology is an important characteristic impacting the description of the fluid

domain (the viscous tensors intervene in the momentum equations). The sludge vis-
cosity was determined experimentally by using an AR2000 (TA Instruments) rheome-
ter. Such rheometer uses a double plate geometry with a gap of 2.0mm which rotates
progressively until a maximum shear rate (Locatelli, 2015).

The rheology test was only made with the sludge samples from April. Thus, the vis-
cosity of same samples used for the batch experiment (section 3.3.1) was measured.
The samples were sieved at 2mm prior to testing. The initial operating conditions
for the rheometer are in table 4.6.
TABLE 4.6: Rheometer Characteristics

Setting Parameter
Plate diameter 2cm
Gap 2mm
Temperature 20°C
Pre-shear duration 1 min
Accelaration mode Linear
Maximum shear rate 500s−1
Duration of the ramp 5 min

The evolution of the shear stress with respect to the shear rate for one of the sludge
samples can be observed in figure 4.8. The point corresponding to the intercept of the
curve is considered as the yield stress (τ) and the slope of the curve is the apparent
viscosity (µ0 ). The sludge presents a viscoelastic behavior at the beginning of the
test. After crossing the shear stress of 100 s−1 the behavior is plastic, thus the sludge
can be considered as a plastic fluid.

The apparent viscosity and the yield stress can be both related to the sludge con-
centration in an exponential equation (figure 4.9). Table 4.7 shows the values of the
 101

TABLE 4.7: Parameters for the viscosity models

Parameter Value
C1 0.0017Pass
C2 0.0852L·g−1
A1 0.0019Pas
A2 0.432L·g−1

4.2.4 CFD Simulation Description

Sludge Settling Model

The settling model used for the simulations is the one described in section 2.1.3. The
objective is to validate the model, with the calibrated parameters found in the batch
settling column, with the measurements taken in the full-size clarifier. April and
October experimental campaigns conditions are simulated using the same model
functions and parameters described in section 3.3.1.

Sludge Rheology Model

Sludge viscosity is modeled by a plastic model, i.e. the yield stress of the sludge is
not considered. The Bokil and Bewtra (1972) equation 4.2, is used to describe the re-
lation between the apparent viscosity (µ0 ) and the MLSS concentration. µ0 increases
exponentially with the MLSS concentration. One advantage of the equation is that it
has only two parameters and is valid for sludge concentrations above 0.7 kg·m−3 .

The values of the parameters C1 and C2 used for the simulations are presented in
table 4.7. However, in the OpenFOAM code, equation 4.2 is computed in function of
the volume fraction of the mixture (α) and is based 10.

The sludge density used for the simulations was 1010 Kg·m−3 (based on the values
obtained experimentally). The considered water density is 998 Kg·m−3.

Geometry meshing

To reduce the computational cost of the simulation, an axisymmetric representation

(known as 2D or quasi3D model) was chosen. The same dimensions illustrated in
figure (4.3) are used to build the geometry and mesh. The scrapper is not included
in the mesh.

To construct the mesh, again the snappyHexMesh tool from OpenFOAM was used.
First a rectangular 3D block of the half part of the clarifier (from the middle to the
external wall) was built. Perfect squared corners are needed in the intersection of
three limits therefore refinement is done in the external and internal baffles (figure
104

The p-value is calculated to determine if the differences between the variances are
likely or unlikely. For this test, if the p-value is > 0.05 then the variances are not
different and we consider that the meshes do not produce different results.

Table 4.8, shows the p-value when comparing the variances of the different couples
of the three different meshes for both concentration profile and RAS concentration.
The found p-values are higher than 0.05, thus no significant differences are found
when changing the mesh. Therefore, the coarsest mesh can be used in order to re-
duce the computational time.

TABLE 4.8: P-values for the t-student test to compare the variances of
the calculated RAS volume fraction and the volume fraction profile.

Couple of meshes RAS Volume fraction Volume fraction profile

7970-8897 0.9264 0.9787
7970-12816 0.7796 0.9692
8897-12816 0.7130 0.9479

Thus, the axisymmetric mesh of 7970 cells, which average cell dimensions are 16.34cm
in the horizontal direction and 4.48cm in the vertical direction is used. The smallest
cells (refinements near the baffles) dimensions are: 1.51 and 0.56cm in the horizontal
and vertical axes respectively.

Turbulence Model

The same buoyancy k-epsilon model described in section 1.6.2 is used for all the
simulation cases. Buoyancy effect can occur due to the sediment-induced density
differences (Lakehal et al., 1999). Indeed, under the influence of the gravity, a particle
having a density higher than the surroundings will settle faster, until the buoyancy
force equals the drag force. Therefore, an important force shaping the flow field is
the buoyancy, this is produced when the density ratio between the phases is low and
the drag between them is high (Brennan, 2001).

General boundary conditions for the simulation cases

Boundary conditions have to be set to each one of the patches constituting the geom-
etry of the 2D axisymmetric clarifier (figure 4.10). Table 4.11 summarizes the input
set to the base case. The description of those boundary conditions appear below the
table.

The initial sludge concentrations are 5.54 Kg·m−3 for April simulations and 4.54 Kg·m−3
for October simulations.

The boundary condition for the velocity of the incoming flow (~υin ) depends on the
configuration and is calculated as the ratio of Qin and the sectional area of the inlet
Ain
 105

A base case for both campaigns was simulated. For this, a constant velocity is im-
posed at the inlet and the removal patches. The velocity of the fluid is calculated
using the average measured flows for Q and Qr taken from the WRRF during the
days where the experimental tests were carried on. The simulation was over when
the convergence was reached (a constant SBH and RAS through time was observed)

The selected values appear on table 4.9

TABLE 4.9: Average measured flows during the campaigns carried on

in April and October

April October
Q m3 · h − 1 58 52
Qr m3 ·h−1 62 63

A brief summary of the different cases discussed within this chapter is on table 4.10:

TABLE 4.10: Summary of the simulated cases for validation

Type of
Type of experimental data Settling model
Case Hydrodynamic
acquisiton for validation Coupling
simulation
Vesilind and
Section 4.3.1 No applicable Constant flow
Simplified form of DeClercq
Vesilind and
Section 4.3.1 Punctual Measurements Dyanmic flow
Simplified form of DeClercq
Diehl and
Section 4.3.2 No applicable Constant flow
Simplified form of DeClercq
Diehl and
Section 4.3.2 Continuous Measurements Dyanmic flow
Simplified form of DeClercq
 TABLE 4.11: Boundary conditions for the base case (continuous inflow). In brackets () the name of the variable in OpenFOAM

106
Volume Fraction Velocity Pressure Kinetic turbulent Dissipation rate Kinematic
Patch
(alpha.sludge) (U) (p_rgh) energy (k) of k (epsilon) Viscosity (nut)
Inlet Fixed value Fixed value zeroGradient Fixed value Fixed value Calculated
Outlet zeroGradient zeroGradient Fixed Value to 0 zeroGradient zeroGradient zeroGradient
Removal zeroGradient Fixed Value zeroGradient zeroGradient zeroGradient zeroGradient
Baffles and Walls zeroGradient Non-slip condition zeroGradient kqRWallFunction epsilonWallFunction nutkWallFunction
Free surface Symmetry
Front and Back Wedge

Fixed Value. It is a specified value of the variable. It must be always expressed in international system units.

ZeroGradient. This boundary condition extrapolates a quantity to the patch from the nearest cell value: the quantity is developed in space and its
gradient is equal to zero in the normal direction of the boundary.

No-Slip. It is an alternative to the zero fixedValue boundary condition for velocity. There is no difference between them. It is applied only in the
wall patch.

WallFunctions. When wall turbulence modeling, the distance from the wall to the cell centers next to the wall is stored as part of the patch.
kqRWallFunction, acts as a zero-gradient condition at the wall. epsilonWallFunction, calculates epsilon and such values are added into the matrix
to act as a constraint, and nutkWallFunction is a condition for kinematic viscosity

Calculated. It is a condition in which OpenFOAM is calculating automatically the estimated value derived from other fields.

Symmetry. This is a symmetry plane.

Wedge. Used for axisymmetric cases, the geometry is specified as a wedge of small angle (e.g. < 5°) and 1 cell thick running along the plane of
symmetry, straddling one of the coordinate planes (OpenFOAM Manual). OpenFOAM understand that radial coordinates should be employed.
 107

To simulate the dynamic flow behaviour, within the simulation the values of Q and
Qr changed in time. Flow values where chosen according to the real behavior of
the WRRF flows during the experimentation days for both April (figure 4.12a) and
October (figure 4.12b) campaigns.

F IGURE 4.12: Clarifier and Recirculation flow during a) April 6th 2018
and b) 17th to 19th October 2018

In OpenFOAM a fixed value for velocity (U) has to be set for each time when the
inflow/RAS flow is changing, therefore for the simulations held in April and Octo-
ber, the velocity values are set according to figure 4.12a and figure 4.12b respectively.
The rest of the boundary conditions are set according to table 4.11.
108

4.3 Results

4.3.1 Punctual measurements results (April 2018)

Experimental Data results

Figure 4.13 shows the evolution of the amplitude during the test in the clarifier when
the ultrasonic transducer was placed at 8.6 m away from the inlet. There is a clear
line in the graphic with high values of amplitude (0.0011V) that indicates the bottom
of the clarifier position at 3.1 m. Another constant line is observed at 1.8 m, with
smaller values of amplitude (≃ 0.0002V) indicating the sludge blanket height.

The maximum particle settling velocities values found were around 0.001 12 m·s−1
whatever the transducer position and the measurement depth.

The blue "curtain" all over the depth of the clarifier observed after 15 minutes is
maybe due to particle dispersion cause by the inertia when the extraction pump is
off. This coincides exactly when the pump is inactive.

Figure 4.14 shows the evolution of the settling velocities measured at 8.6m from the
clarifier inlet. During the first 15 minutes, the sludge particles show high velocities
towards the bottom of the tank. Those velocities are increased probably because
of the recirculation pump, which is active during this period. After 15 minutes,
the particles seem to settle slower and particles velocities seem to be more disperse
within the blanket and the clear zone. This moment coincides with the stopping of
the extraction pump.

During all the test, a periodic behavior with negative-positive velocities is observed
more less every 30 seconds. Positive velocities indicates a direction towards the bot-
tom of the clarifier. Thus, particles are settling and rising within the sludge blanket.
The behaviour might be produced by the density currents (differences in densities
of the water-sludge) created in the surface of the SBH.

The mean settling velocities found within the sludge blanket measured at 8.6m are
around 0.0005 m·s−1 which are a hundred times higher than those found in batch
column. This indicates that not only the particles settling velocity takes action within
the sedimentation process, but also the incoming velocity of the mixture, density
currents and/or the advective transport due to recirculation flow.

Figures 4.15 and 4.16 show the amplitude and settling velocities of the sludge parti-
cles respectively, measured at 3.5m away from the inlet. The sludge blanket surface
is at 1.7m. The bottom of the tank was not observed during this measurement.

During the first 7 minutes, particle velocities are observed at the bottom of the tank.
The recirculation pump was active during this time. After 7 minutes, the velocities
at the bottom of the tank seem to be more dispersed in the vertical direction, and the
 F IGURE 4.14: Evolution of the settling velocities measured at 8.6m from the clarifier inlet
110
 111

velocities within the blanket seem to be lower. This, also coincides with the inactive
recycling pump period.

The same trend is observed as the measurements taken at 8.6m. Periodic positive
and negative velocities every 30 seconds are observed. The skimmer was stopped
during the measurement, this behavior cannot be attributed to the motion of it. Thus,
the rising and settling of the particles maybe attributed also to the density currents
created by the incoming flow.

Within this zone settling velocities seem to be higher than those measured near the
wall. Maximum values are about 0.001 m·s−1 , maybe due to the incoming sludge
and the extraction zone that disturbs the blanket.

Constant inflow simulations results

In order to initiate dynamic simulations, uniform and constant boundary conditions

at the inlet and recirculation were first imposed in order to reach the convergence of
the simulations of the settling tank.

The convergence was reached when the variations of the RAS and the SBH from one
time step to other were less than 5%. The clarifier at the beginning of the simula-
tion is supposed to be empty. The simulation converged after 10 hours of continu-
ous flow. The average estimated RAS concentration was 8.66 Kg·m−3 and the final
sludge blanket height was 1.37m. Checking the mass balance the difference between
the inlet flux and the removal flux is near 17% which is acceptable.

The sludge blanket height is uniform along the horizontal axis and some waves ap-
pear in the zone near the inlet. In the inlet zone the sludge goes down and comes into
contact with the thickened sludge. This, produces the fluid to decelerate and shift
upward producing the waves in the upper part of the sludge blanket. The sludge
particles start to settle causing an accumulation at the bottom of the clarifier (figure
4.17).

As the fluid is coming inside the tank, it goes directly towards the bottom (figure
4.18a). However, a velocity gradient is found between the inlet and the removal
patches, just next to the internal wall. This velocity gradient is generated by the flow
extraction made at the recirculation. Once the flow reaches the bottom it starts to
spread in the horizontal direction.

The buoyant effect term (represented by Cg in the turbulence model) makes a rise
up of the SBH while it is thickening. (figure 4.18a). Energy is dissipating along the
clarifier producing a slow down in the velocity of the fluid towards the external part.
(figure 4.18b).

The sludge concentration profile has a smoother gradient in the vertical axis near the
inlet, and a higher concentration is reached at the bottom of the tank (figure 4.19a).
 F IGURE 4.15: Evolution of the amplitude measured at 3.5m from the clarifier inlet
112
F IGURE 4.16: Evolution of the settling velocities measured at 3.5m from the clarifier inlet

113
 125

On figure 4.28 the particle velocity evolution results show that flocs within the dis-
crete zone achieve high velocities. This maybe due to the incoming fluid motion. The
negative velocities indicates that particles are moving towards the bottom. However,
during these days, the trend of positive and negative velocities every 30 seconds
which was observed in April is not present in October.

And, probably the movement of the skimmer is also affecting the settling velocity of
the particles, since the transducer is attached directly to the skimmer and following
its path. It is tricky to establish the velocities of the particles within the blanket due
to its thinness, compared to the total depth of the clarifier (3.5m).

F IGURE 4.28: Evolution of the particles settling velocities measured

( f 0 = 1.65MHz) during 51 hours. Copyright (Ubertone)

Within this experimental test, it is observed that almost all the particles velocities
are negative, which indicates that they travel towards the bottom of the clarifier.
The negative/positive particles velocity trend within the sludge blanket it is not
observed here. One possible reason is that Peacok UVP transducer was recording
the particles velocity data every 45 seconds, while the positive/negative velocity
trend was observed every 30 seconds.

CFD model in continuous constant flow mode

Similarly to the previous case, before simulating the true hydrodynamic behavior of
the clarifier of Achenheim, a base simulation was performed in order to reach the
convergence. The inlet flow was set to Qin = 115 m3 ·h−1 and the RAS flow was set to
Qr = 63 m3 ·h−1 . The evolution of the predicted sludge blanket height at 3.5m away
from the inlet was extracted. In section (4.3.2), the comparison of the simulated SBH
to the experimental data is presented.
126

F IGURE 4.29: a) Sludge concentration and b) Velocity field after 36h

of simulation with the sludge and flow values of October
 127

Figure 4.29a shows the sludge concentration distribution of the clarifier after 36h
of simulation using the sludge characteristics of October. A thin sludge blanket is
formed just near the inlet. At a radial distance of 3.5m, the blanket height is barely
20cm. The sludge concentration at the bottom slope is 22 Kg·m−3 which is a high
concentration. However, it was long to get the steady state; variations of concen-
tration from 4.3 kg·m−3 to 10.7 kg·m−3 were observed during the 36h of simulation,
indicating an average RAS concentration of 5 Kg·m−3 in periods of 5 hours.

The high variations in the RAS concentration, indicates that sludge has not a con-
stant thickening. This can be explain by the continuous recirculation extraction
which is producing a short-circuiting at the inlet zone and diluting the sludge ac-
cumulated at the bottom (figure 4.29b).

CFD model validation using the dynamic boundary conditions

The objective of this campaign was to measure the SBH continuously for a longer
period and compare with the simulation results.

The experimental sludge height was extracted from the amplitude profiles at a fre-
quency of 1.64MHz. The measured SBH (red dotted line in Figure 4.30) was deter-
mined setting an amplitude limit value of 0.0001V. Then, a moving average with a
period of 10 was calculated to obtain representative heights.

To simulate the intermittent flow of Achenheim clarifier the inflow and recirculation
rates conditions for the CFD model, were set accordingly to figure 4.12b. The simu-
lated SBH was measured at the same point where the ultrasonic device was placed
in the full-size clarifier and compared to the measured SBH (figure 4.31). For the
simulated SBH, the height was considered where the concentration in the surface of
the blanket is equal to the initial concentration.

Figure 4.30 shows the simulated and measured SBH during the 51 hours of the ex-
perimental test with respect to the incoming and recirculation flows. In general, the
measured sludge blanket height is low, barely 60cm from the bottom of the tank at
the maximum point.

It is clear that the model is able to reproduce the SBH variations as the sludge blan-
ket is sensitive to both flows (inlet and recirculation). However, the model tends to
be overestimate these fluctuations. For instance, a high peak is observed in the sim-
ulation around 8am in October 18th (figure 4.30b). It coincides with the stop of the
recycling pump and while the incoming flow is still significant. Before this event,
the model under-predicts the SBH, while this is not really the case afterwards.

The model can predict at some point an accurate sludge blanket. However higher
peaks and moments with no sludge blanket are predicted. The model uses an ap-
proximate real flow, i.e. we extracted the flow data from graphs obtained directly
 129

from the WRRF. The extraction method for the flow data has done visually which
may create some uncertainty in the values. This can explain why in some points the
simulated sludge blanket do not agree with the measured one.

In figure 4.31, it is seen that the intermittent flow simulation seems to predict a better
trend of the SBH than the constant flow simulation. The model with constant flow
is under-estimating the SBH.

The intermittent flow model seems to be more sensitive to the changing loads spe-
cially when the recirculation flow is off. However, the model can represent a good
height when the recirculation flow is on. The particles in the clarifier are disturbed
by the scrapper motion, making possible an elevation of the blanket. This movement
can be captured by the transducers, and thus lump the real settling process. One pos-
sible explanation why some elevations (peaks) do not coincide with the experimen-
tal data, is that the model does not consider the scraper motion and this may have
an impact on the fate of solids and velocity field (swirl motion, increased particle
flow towards the removal, De Clercq, 2003). The variations in the predicted sludge
concentration are noticeable (figure 4.32). However, the average sludge concentra-
tion during the 3 days remain in 9.8 Kg·m−3 . Different sludge concentrations are
expected during the day, at some moments such concentration can rise 30 Kg·m−3 ,
and sometimes no sludge is thickening. The intermittent behavior of the clarifier, do
not allow for a constant sludge concentration at the removal. This variant sludge
concentration may impact the performance of the biological reactor and the sludge
processing line.

A look inside the dynamics of the SST (figure 4.33), shows that effectively, a thin
sludge blanket is created inside the tank. Since no thickened sludge blanket is found,
the incoming sludge goes immediately towards the removal. Different reasons can
explain this behavior:

• Lower average incoming flows (Qin = 115 m3 ·h−1) compared to those in April
(Qin = 120 m3 ·h−1) and the same RAS flow (Qr = 63 m3 ·h−1) do not allow the
sludge to accumulate at the bottom.

• The high mineral content in the sludge makes to settle faster and thus no thick-
ening is carried on.

Unfortunately, the comparison of the measured settling velocities and the simulated
ones are not presented here. Indeed, we analyzed the values, and it is hard to com-
pare the measured values due to the thinness of the sludge blanket.

4.4 Conclusions

To validate the model, different experimental campaigns were carried out from Jan-
uary to October 2018. Only two campaigns were able to obtain satisfactory results,
132

F IGURE 4.33: CFD simulation of a) sludge concentration and b) mix-

ture velocity at 3:30pm on October 18th
 133

in April and October.

The continuous measurements during 51 hours revealed that the sludge blanket
within the clarifier has a dynamic behaviour, finding height values between 0.6 and
0.2m. The punctual measurements, carried on during 20 minutes at different radial
distances, showed that particle settling is possibly disturbed by the density currents
generated by the incoming flow. Indeed, the particles velocities rise and descend
within the sludge blanket every 30 seconds, even at distances far away from the
inlet.

From the campaigns carried on April and their respective simulations the following
conclusions can be determined:

• From the continuous flow case, we observed that in the external zone of the
clarifier, quiescent conditions are created. Indeed, through the velocity and
concentration vertical profiles, a similar behavior to the batch settling profiles
is obtained. This can be also corroborated with the lower particles velocities
measured at a radial distance of 8.6m when they are compared to the particles
velocities measured at 3.5m.

• The measured particle velocities in the clarifier are higher compared to those
found in batch experiments (1.10− 5 m·s−1 ).From the comparison between the
measured particles velocities and the simulated vertical convective and set-
tling velocities, it was observed that the settling velocity of particles is lumped
by the convective vertical velocity of the mixture. Thus, the convective veloc-
ity of the mixture drives the particle motion within the sludge blanket in the
clarifier.

From the data extracted in October we can conclude that:

• Simulating a constant inflow and recirculation flow during 51 hours in the

clarifier, the SBH will be underestimated when compared to the experimental
data.

• cfd Simulations showed that SBH dynamics can be accurately represented by

setting the same operating conditions of the clarifier, i.e.,, the inlet and recir-
culation flow rate were changing according to the flow values measured in the
WRRF. Even if there is a later response in the prediction of higher peaks of the
SBH, one can see that simulated average sludge blanket heights are similar to
the measured ones.

It was observed that sludge properties, in particular the mineral content, produce
different results in the measured and simulated sludge blanket and RAS concentra-
tion. Having similar initial concentrations (5.54Kg.m−3 in April and 4.54Kg.m−3 in
October), the thickening of the sludge blanket is different. During April campaign
the SBH was in average 1.34m, while in October it was in average 0.5m.
134

Different couplings of settling phenomenological functions from used to describe the

Achenheim sludge settling depending on the season where the data was extracted
form the WRRF. For April sludge, the Vesilind-DeClercq’s simplified form (equa-
tions 1.6 and 1.13 respectively) functions are used. The choice was made due to the
lack of data to validate first the batch settling test data. This coupling may lead to an
overestimation of the SBH within the clarifier, however such overestimation is more
remarkable after two hours of settling. In Achenheim clarifier, the sludge is contin-
uously extracted (every 20-30 minutes), thus the risk of SBH overestimation can be
neglected. Anyway, results have shown that this coupling can predict an accurate
sludge blanket height within the clarifier.

To validate the model using the data of Achenheim clarifier in October, a different
settling model was used. The couple Diehl-DeClercq’s simplified form (equations
1.8 and 1.13 respectively) is employed. The change was made because we wanted
to test from the beginning this approach since Torfs et al. (2016) have found that
this couple predicts a more accurate sludge blanket height in batch settling. This
coupling predicts as well accurate blanket height results inside the clarifier.

One perspective of this study would be the measurement of the sludge blanket
height during the rainy events on the Achenheim WRRF to see deeply the impact
of the compression function used for these simulations.

Rheological parameters have also an impact but this was not investigated during
this study, this will be discussed in the next chapter 5.

In brief, by using the CFD enhanced solver described in (2) (Valle Medina and Lau-
rent, 2020) and the calibrated parameters found in chapter 3, different CFD simu-
lations were performed and revealed satisfactory results for the prediction of the
sludge blanket height and the particles velocity profile.
 135

Chapter 5

CFD case studies

CFD has become a powerful tool for prediction, optimization and analysis of the
hydrodynamics inside a WRRF. CFD allows to obtain a glassbox overview of the
settling behavior, in which one can observe the behavior of the sludge mixture in-
side the clarifier. CFD models, being a 3D approach, are normally used to study the
hydrodynamics of the clarifier when a physical part is changed (i.e., shape, baffle po-
sitions and dimensions, inlet surface... (Flamant et al., 2004, Griborio, 2004, Xanthos
et al., 2011, Das et al., 2016).

The aim of this chapter is to present different scenarios by changing some variables
in the CFD settling model. On the one hand, the aim is to assess model responses to
changes in model parameters and/or functions (compression parameters, rheology).
On the other hand, simulating a wet weather condition allows to evaluate if the
model is able to capture the expected SBH and RAS dynamics in these conditions.

5.1 Case studies presentation

The different case studies include the variations on:

• The parameter λ of compression function in equation 5.1.

• The critical concentration above which compression starts Xcrit in equation

2.18.

• A change in the rheological submodel.

• A extremely high hydraulic load.

Like in the previous chapters, an axi-symmetric 2D CFD approach is used to allow

faster results. The solver settings are as follows:

• Continuous simulations i.e. no intermittent flow is present in the inlet and

recirculation;

• Convergence of the simulation is considered when the RAS concentration and

the SBH are constant through time;
136

• The sludge settling model is the coupling of Vesilind function (equation 2.2)
and constant solid stress parameter (equation 5.1) for hindered and compres-
sion settling respectively. This choice was made because we departed from
the validated case of the clarifier of Achenheim using the experimental data in
April. During this month a true sludge blanket was seen, therefore within this
case the impact of changing the compression parameters will be more remark-
able.

• The buoyancy k-ǫ turbulence model is used.

• The inlet sludge concentration is (X0 ) of 5.54 Kg·m−3

However, the case with extremely high hydraulic condition has a different setup:
the settling velocity model is chosen to be the couple of equations 2.20 and 5.1. As
in the previous case, the departing point was the validated case for Achenheim clar-
ifier using the data in October. During this month, a low sludge blanket height was
observed, and thus by a CFD approach we want to demonstrate that the clarifier has
the capacity of retain solids even at high hydraulic loads. The inlet sludge concen-
tration was set to 4.54 Kg·m−3 . The incoming flow was set to a value higher than
the maximal flow capacity of the WRRF. For all the cases, the sludge blanket height
SBH was measured at the after 36h of simulation. In the simulations the SBH was
calculated as the limit between the settled sludge and the clear water, i.e. where the
concentration in the upper part of the blanket is equal to the initial concentration
(X0 ).

The velocity, concentration and compression function (dComp ) profiles at 5.4m away
from the inlet, and the RAS concentration are measured after 36h of simulation, ex-
cept for the case with a different rheology model where the profiles were measured
at three different radial positions.

Table 5.1 shows a summary of the different cases discussed in the following sections

TABLE 5.1: Summary of the analyzed cases

Modified Parameters
Case Objective
Conditions
To observe the impact of the effective
λ for effective solids
Section 5.2 solids stress parameters in the prediction
stress
of the SBH and RAS concentration
To review the impact of critical
Critical concentration
Section 5.3 concentration value in the prediction
for compression function
of the SBH and and RAS concentration
To understand the hydrodynamics inside
Rheology model
Section 5.4 the clarifier when the viscosity of the
for sludge viscosity
sludge is changed
To analyse the impact of a high hydraulic
Section 5.5 Flow initial condition load in the clarifier performance using a
settling model including compression
 137

5.2 Impact of compression parameter λ

In the compression function, the effective solids stress (σe′ ) accounts for the property
of activated sludge to thicken due to the permanent contact between the flocs and
to resist to deformation (Buscall and White, 1987, Aziz et al., 2000). De Clercq (2006)
determined in batch experimentation, that the effective solids stress has high values
when the initial concentration is high.

To reduce the number of compression parameters within the CFD solver/code, the
parameter λ is set equivalent to the derivative of the effective solid stress (σe′ ), and it
is only valid when the sludge concentration crosses the critical concentration (Xcrit ).
Thus, the primitive of (σe′ ) becomes:


0 for 0 ≤ αd < Xcrit /ρd
σe (αd ) = (5.1)
λ(α − Xcrit
for αd ≥ Xcrit /ρd
d ρd )

Thus, the parameter λ indicates the magnitude of the slow-down of the settling ve-
locity when the compression is active

Through the calibration process described in chapter 3, we obtained different set of

parameters that could describe the settling behavior. We set the limits of parameter
λ between 0.01 and 0.1 m2 ·s−2 and the estimated parameter set gave satisfactory
results within the first hour of batch settling.

To highlight its influence on the CFD simulation results, a value of 1 m2 ·s−2 is im-
posed. If the value is higher this will indicate that the settling velocities within
the sludge blanket shall be lower than those found in the base case where λ =
0.046 m2 ·s−2 (section 3.3.1) and therefore the prediction of the SBH should be higher.

The case when λ is 1 m2 ·s−2 is compared to the base case at a time step of 36h. Fig-
ure 5.1a shows the sludge concentration distribution in the clarifier at such time. It is
observed that the average concentration found at the removal patch is 6.18 Kg·m−3 .
This average predicted RAS concentration is lower compared to the predicted value
for the base scenario (8.66 Kg·m−3 ). The predicted concentration within all the clari-
fier is similar (maximal concentration of 6.5 Kg·m−3 ).

Since, the effective solids stress (σe ) is analogous to a dispersion term (equation 2.1),
its higher value in the simulation yields to a relatively homogeneous solids distribu-
tion along the horizontal and vertical axis of the blanket.

The measured SBH at the middle of the clarifier is higher (1.66m) when λ is equal to
1 m2 ·s−2 , than when the value of λ is 0.046 1 m2 ·s−2 . A higher value in the compres-
sion function, due to a higher value in the effective solids stress, makes that settling
velocity to be slower and then an elevation in the SBH is expected.
138

F IGURE 5.1: Concentration distribution after 36h of simulation when

a) λ = 1 m2 ·s−2 and b) λ = 0.046 m2 ·s−2
140

compression function is active after 30 minutes. This delay is expected as the sludge
needs more time to thicken to reach the 8 Kg·m−3 .

Surprisingly, the higher value of Xcrit predicts a higher sludge blanket (figure 5.3a)
(2m against 1.34m). This behevior was not expected as a higher compression limit
would lead to a lower sludge blanket. One hypothesis to explain this is as follows:
as compression is only active for a high concentration, the sludge thickens faster in
the upper parts of the sludge blanket (hindered settling only). This solids flux com-
ing to the bottom yields to a higher concentration that crosses the Xcrit value. The
higher MLSS concentration leads to a higher viscosity as computed by the Bokil and
Bewtra (1972) relation (equation 1.40). As the shear stress at the bottom is low, this
restrains the flow to reach out the removal and induces the observed SBH increase.
The motion of the scrapper at the bottom of the tank may overcome this viscosity
problem by breaking the shear stress of the sludge.

In figure 5.4a and 5.4b the concentration profile and compression function profile
are shown respectively. In contrast to the previous case, when the value of Xcrit
is 8 Kg·m−3 , the difference between the beginning of the compression zone (0.95m,
figure 5.4b) and the sludge blanket (2m, figure 5.4a) do not coincide. The former
behavior can be explained by the fact that compression function is constant through
time depending only on the local concentration. The limit of the compression zone
will rise only when the concentration in each cell will be higher than 8 Kg·m−3 .

When Xcrit = 8 Kg·m−3 :

• The concentration profile (figure 5.4a) clearly displays a concentration trend

similar to the one observed in a batch column. This confirms that in this zone
of the clarifier quiescent conditions are present. Due to the delay of sludge to
cross the critical concentration, sludge accumulation is higher and thus the pre-
diction of the concentration at the bottom of the clarifier is higher (9.8 Kg·m−3 )
than the simulated one with the reference scenario (8.8 Kg·m−3 ).

• As expected, the values of the compression function (figure 5.4b) are lower to
those of the reference case. Within the sludge blanket, settling velocities are
lower and produce the small values for dComp.

• The velocity profile (figure 5.4c), again show the trend of the settling velocity
expected in a batch column. A constant velocity between 1.9–1.42m is observed
indicating that sludge is in hindered regime.

Table 5.2 shows the differences of the values obtained for the SBH and the RAS con-
centration between the reference case and the study cases. In this simple approach,
it is observed that λ will have a bigger impact on the prediction of the sludge con-
centration, while Xcrit will impact on the prediction of the SBH.
 143

Where PC and PE are parameters for the viscosity model. The parameters are set
into the transportProperties file (see Apprendix A).

For the Bingham approach, the apparent viscosity (µapp ) of the mixture is calculated
using the yield stress (τ0 ), the plastic viscosity (µ0 ) and the rate of strain (equivalent
to the velocity gradient) (equation 5.3). The yield stress is calculated from the Dick
and Ewing (1967) equation (equation 2.24) but noted in base 10 and in function of
the volume fraction.

τ0
µapp = µ0 + (5.3)
|γ̇|

The parameters used for the Bokil and Bewtra (1972) and Dick and Ewing (1967)
equations are those in table 4.7, which describes the rheology characteristics of the
sludge sample taken in April.

For this case, the sludge blanket height and the concentration and settling velocity
vertical profiles, are extracted at three different radial positions. The settling ve-
locity profiles of the Bingham plastic model are compared to the base case and the
experimental ones.

5.4.2 Results: concentration profiles

The RAS concentration after 36h of simulation is 10.5 Kg·m−3 . Figure 5.5 shows
the sludge concentration distribution and the calculated sludge blanket height at
different radial positions.

Compared to the base case, only at the middle of the tank (Rd = 5.4m) the SBHs are
estimated similar for both cases. In the base case, no yield stress is considered and
thus the mixture can flow freely at the bottom of the tank and thus reach the external
wall more rapidly producing an homogeneous elevation in the horizontal direction
of the clarifier (figure 5.5b).

In figure 5.5a it is observed that the distribution of the sludge blanket is not uniform
in the horizontal axis. The yield stress of the sludge makes that the sludge starts
to thicken near the inlet zone. With the slow incoming flow, the shear stress of the
sludge is hard to break producing a thickened blanket that moves hardly to the ex-
ternal wall. Thus, this makes a high elevation in the blanket near the inlet zone and
decreasing at the external part of the clarifier. High sludge concentrations within the
sludge blanket are found near the external wall. By considering a yield stress for
the activated sludge, the mixture offers resistance to deformation while moving to-
wards the outlet, and thus a more concentrated sludge can be found in a zone where
no shear rates are present.
148

5.4.4 Results: impact of the viscosity in the compression function

As stated in the former case, the high viscosity of the activated sludge induced lower
predicted settling velocities as with less viscous sludge. Thus, if settling velocities
within the hindered zone are becoming slower, the same trend can be observed with
the compression function. In figure 5.10a the low values of the compression function
are present esspecially near the external wall where the motion of the incoming fluid
do not disturb the sludge settling. In the plastic model approach the sludge has no
yield stress to beat, therefore the sludge settling is becoming faster producing in
some areas a lower sludge blanket (5.10b)

F IGURE 5.10: Compression function distribution within the clarifier

for a) Bingham plastic model and b) Plastic model

Care must be taken, if the Bingham type model is chosen to simulate the rheology
of the sludge, when a slow flow is coming into the settling tank. The predictions
exaggerate the elevation of the sludge blanket at the inlet, which may block the free
transit of the fluid predicting a short-circuiting in the clarifier.

If a Bingham-plastic model is chosen for activated sludge modeling in a CFD ap-

proach, one should consider to include the movement of the scrapper at the bottom
of the tank (Lakehal et al., 1999). This will help to overcome the yield stress of the
150

By adding the compression function in the model, a more realistic prediction in the
sludge blanket and RAS concentration is expected, than using an equation only ac-
counting for the hindered settling during wet weather conditions (Torfs et al., 2015b).

The following case simulates a high hydraulic scenario where the surface overflow
rate (SOR) is higher compared to a dry weather scenario. Thus, the values of Qin and
Qr were set to 538 m3 ·h−1 and 288 m3 ·h−1 respectively. The sludge concentration at
the inlet Xin is set to 4.54 Kg·m−3 . Table 5.3 makes a summary of the operation
parameters for the wet weather and reference scenarios.

TABLE 5.3: Operation parameters of the clarifier in two scenarios.

High hydraulic load Normal flow

Q(m3 ·h−1 ) 250 52
Qin (m3 ·h−1 ) 538 115
Xin (Kg·m−3 ) 4.54 4.54
Inlet mass flux (Kg·h−1 ) 2443 522
Q r ( m3 · h − 1 ) 288 63
Recirculation factor 1.15 1.21

Figure 5.12b shows the distribution of the sludge within the clarifier when a higher
hydraulic load comes into the clarifier. After 36 hours of simulation, the maximum
reached height for the blanket is 0.8m and the RAS concentration is 7.33 Kg·m−3 ,
those values are higher compared to the dry scenario after (figure 5.12a). Indeed, a
higher sludge inventory inside the tank is expected as more sludge is coming into
the clarifier.

For both scenarios, the maximum concentrations predicted at the bottom of the tank
are high. This is due to the high settling velocity of the sludge. The high mineral
content of the sludge causes a very high hindered settling velocity which may lump
the compression effect producing a higher sludge accumulation.

TABLE 5.4: Simulation results of the clarifier in two scenarios.

High hydraulic load Normal flow

Xr (Kg·m−3 ) 7.33 4.94
Recirculation mass flux (Kg·h−1 ) 2052 311
Mass inside the tank (Kgn) 5800 730

In the wet weather scenario, it is more noticeable that baffles acts as an energy dis-
sipator (figure 5.13a). Hence, a higher sludge blanket is expected as more solids are
entering the tank. The high elevation of the sludge blanket, makes that the fluid goes
towards the external wall dissipating the energy while it is crossing the tank. Less
risk of the incoming fluid moving towards the removal is presented. In contrast, in
the reference case the fluid goes immediately towards the bottom of the tank causing
a risk of short-circuiting (figure 5.13b).
152

F IGURE 5.13: Fluid velocity distribution during constant a)high hy-

draulic load scenario, b)normal flow
 153

It was observed, that even at a high hydraulic load, the prediction in the SBH is still
low. From the statement of Patziger, 2016, the simulated SOR (0.7 m·h−1 ) in this high
hydraulic load scenario, has not yet cross the critical SOR, and thus, the incoming
flow can transit in quiescent conditions without risk of short-circuiting. It should be
noticed that, in France, SSTs are often designed with a SOR of 0.6 m·h−1 calculated
on the peak flow.

This indicates that the WRRF can probably operate at higher hydraulic loads. The
clarifier is able to retaine the higher incoming sludge mass while not compromising
effluent quality.

Hence, it noticeable that sludge is not really thickening near the removal zone in the
normal flow scenario. The sludge tends to thicken more when a high hydraulic load
is presented. This not thickening behaviour may be explained by the fast settling
of the sludge produced by the high content in minerals. The sludge is then easily
washed out through the continuous removal.

5.6 Conclusions

These sections showed different CFD scenarios where some parameters of the com-
pression function in the settling model, the rheological approach or the flow at the
inlet were modified. The cases revealed how important and useful CFD can be to
understand the hydraulic behaviour of a settling tank.

In general, a change in the compression parameters will predict different sludge

blanket heights, the distribution of the sludge inside the clarifier and the concentra-
tion at the removal of the clarifier.

The value for parameter λ was increased by 20 times the value used for the reference
case. This resulted in a high sludge dispersion within the blanket zone along the
clarifier. This predicts failure in the clarifier due to the no thickened sludge the
removal.

The Xcrit was increased 1.45 times the value used for the reference case. This change
over-predicted the sludge blanket height. The simulation presented here, the trend
of the SBH was to rise within the first 36 hours of continuous settling. The simula-
tion seems to estimate a extremely high SBH when the steady state will be reached.
Nevertheless, at this time the predicted sludge blanket height is already elevated
compared to the measured sludge blanket in section (4.3.1).

In the 1D models, it was observed that exponential functions describing hindered

settling may lump the effect of the compression function in the prediction of the
sludge blanket height (Torfs et al., 2016). However, in the CFD model it was ob-
served that the compression function has also a high impact in the distribution of
the solids, which is not really captured by a 1D model.
154

Indeed, the value of λ will have a greater impact on the sludge concentration predic-
tion along and at the removal of the clarifier. The parameter Xcrit will have a more
noticeable impact on the sludge blanket height prediction.

The WRRF of Achenheim can operate at high hydraulic loads without compromis-
ing the effluent quality. Through the CFD rainy event case it has been seen that the
clarifier can store the exceeded incoming sludge and thicken it without affecting too
much the sludge concentration at the removal and/or the treated water quality. The
sludge properties are an important variable for the sludge settling accumulation. A
high content in mineral content will lead to a fast particle settling, and thus a non-
thickened sludge blanket with high sludge concentrations at the bottom. This would
show that the compression function is not active.

For the sludge from Achenheim, the plastic model fits better to simulate its hydro-
dynamics. The experimental results showed that a uniform sludge blanket surface
can be found all along the axis, and that the settling velocities are higher near the
inlet and not in the external wall.

It was demonstrated that a plastic sludge flows better to the hopper due to gravity
(higher bottom concentration near the removal), which is an expected behavior in
a clarifier. In this case, the settling velocities are affected by the type of rheology
model employed, the Bingham plastic model predicts lower settling velocities than
the plastic model.

The compression function effect is almost imperceptible when a Bingham plastic

model is used, particularly in zones in quiescent conditions. The estimated settling
velocities are very small at the bottom of the blanket, and in zones where there is no
vertical concentration gradient.

These simulations highlight the importance of calibration and validation of the CFD
model applied to SST. Indeed, changing parameters of the settling/compression
function and the structure of the rheological submodel have significant impact on
model predictions.

Here, the choice of the turbulence model (buoyant k-ǫ) was based on Brennan (2001)
suggestions. The impact of turbulence modeling was not in the scope of this re-
search. However, other simulations performed with our solver demonstrated that
the standard k-ǫ model do not predict a true limit between the clear phase and the
settled sludge (data not shown), and the prediction of the SBH was excessively high.
This area obviously deserves more research.

Thanks to this exercise, we can support the importance of using Computational

Fluids Dynamics models as an optimization/prediction/control tool. By chang-
ing some variables, we analyzed the possible consequences in the prediction of the
sludge at the removal, sludge concentration or possible short-circuiting within the
tank.
 155

Conclusions and Perspectives

Conclusions

This study presented the development of a modified drift-flux solver for the Open-
FOAM® open-source platform, called now compressionFluxFoam. The solids trans-
port equation includes an extra second-order term that accounts for compression. A
simplified constitutive function of De Clercq et al. (2008) was used. An addtional
constitutive functions to describe hindered settling was included: the power-law ex-
pression of Diehl (2015). These developments integrate the most up-to-date knowl-
edge of activated sludge sedimentation mechanisms that were surprisingly never
implemented in the present form in a CFD code.

Within this study we confirmed, in accordance to the experimental test carried on

by Locatelli (2015), that the content of mineral matter in activated sludge deeply im-
pacts the settling velocity. Indeed, the batch settling test for the same sludge showed
very different settling velocities, even at similar initial concentrations. For example,
−4
the sludge with an initial concentration of 4.54 Kg·m−3 settles at 6.81 × 10 m · s−1
which is almost 4 times higher to the settling velocity when the initial sludge con-
centration is 4.16 Kg·m−3 . The high mineral content (55% of NVSS) of the sludge at
4.54 Kg·m−3 makes it to settle faster.

For this reason, it was not possible to gather all the sludge batch test in just one
representative model for hindered settling. Hence, it was observed that the con-
tent in mineral mineral was increasing from 37% of NVSS in January to (55% of
NVSS) in October. The experimental campaigns carried on during this study had
different results, not expected, but interesting as well. The campaign where a no-
table sludge blanket was observed and particles velocities were successfully tracked,
showed an interesting behavior of the velocities within the sludge blanket. Maybe
this behaviour is due to the density currents produced by the different phases, that
disturbs the quiescent conditions inside the sludge blanket.

Unfortunately, we were not be able to observe this behaviour in the next campaigns.
The NVSS content was increasing in the next months, reaching a value of 55% in the
last campaign. This content obviously affect on the settling velocity and thus, the
performance of the clarifier. The fast particles settling process observed can explain
that a very thin sludge blanket was observed. Thereby, it was difficult to validate
156

the model with the particles settling velocities in this campaign carried out during
51 hours.

The CFD model was able to reproduce the batch settling behaviour within the first
hour. The parameters describing this batch behaviour were accurate enough to use
them later for the axi-symmetric clarifier simulations. Hence, the prediction on
the particle settling velocities was improved by adding the compression function.
Results in the axi-symmetric clarifier were satisfactory. Indeed, the sludge blan-
ket height was reproduced accurately, for both measurements in 20 minutes and 48
hours.

It was shown that in the full-scale clarifier, the settling velocities of the particles are
eclipsed by the convective vertical flux of the fluid (mixture). Experimentally, it was
observed that the particles velocities are in order of magnitude between 1.10−3 m·s−1
and 1.10−4 m·s−1 while in batch the settling velocity of the particles are less than
1.10−5 m·s−1 within the sludge blanket. The CFD model could prove indeed, that in
the clarifier the settling velocities of the particles are lumped by the vertical velocity
of the mixture.

The rheology model in the hydrodynamic modelling of activated sludge settling

plays an important role. Using a Bingham plastic approach will result in a highly
viscous sludge that can damp flow fluctuations. It is recommended to use this rhe-
ology model when one is sure to break the sludge yield stress, i.e., including the
effect of the scrapper.

Perspectives

The CFD model still needs to improve its performance. Several studies have pointed
out that the critical concentration is not constant with time. Locatelli (2015) described
this behavior by an empirical exponential equation. However, from a physical point
of view, the critical concentration should not be dependant of the initial/inlet sludge
concentration. Thus, experimental evidence has shown that changes in the floccu-
lation state of particles influence the compression (Torfs et al., 2015b). Including
an equation for each class/size of particle like the approach described in Torfs et al.
(2016), would improve the prediction of the sludge blanket height and thus, the RAS
concentration will be better estimated. Hence, the prediction in the effluent quality
will take part within the outputs of the CFD model and thus gives an estimation of
the performance of the clarifier.

The intermittent flow of the small WRRFs, shows that they have strongly differ-
ent operational conditions than the bigger ones (PE > 90 000). Indeed, the inac-
tive/active periods of both recirculation and inlet pumps affect the performance of
 157

the clarifier. CFD SST modellers should also look into this small WRRFs to test and
improve the existing models.

In this work, we have coupled DAKOTA software to OpenFOAM for parameter

estimation. This tool offers promising perspectives for:

• using global sensitivity and estimation algorithms for an improved calibration

process in batch settling and full-scale SST simulation;

• improving geometrical design of the tank. DAKOTA can be used for automatic
shape optimization. Within SST, several geometrical features can be modified
to improve the performances: size of the internal baffles, baffles distance from
the inlet, scrapper rotation and configuration, tank size, effluent dissipation in-
let configuration. The possibilities are unlimited to optimize the performance
of SSTs.

The CFD solver can also be improved by adding new rheological models. Currently,
within the new solver, OpenFOAM only accounts for the plastic and Bingham-Plastic
models to describe sludge viscosity. Other models type like Cross, Herschel-Buckley,
Carreau, should be included in order to study also the effects of adding the shear-
thinning (pseudo-plastic behaviour) in the hydrodynamics of the activated sludge.
This open perspectives to use this solver in other unit processes that SSTs where
density gradients and rheological properties can affect significantly the hydrody-
namic properties like the biological reactor itself (especially in membrane bioreac-
tors), anaerobic digesters, sludge thickening processes...
 159

Appendix A

Setting up a base case in

OpenFOAM

The mixture model approach is a model where all the participating phases are treated
as a mixture, i.e., in which the fluid exhibit mean properties of density and viscosity.
Thus, the mixture model code in OpenFOAM is called "DriftFluxFoam" or "com-
pressionFluxFoam". The solver is applicable for fluids with small scale interfaces
(see image A.1), fluids found in settling tanks, cyclone separators, bubbles in heat
exchangers, anular flow in refineries. (Márquez Damián, 2013)

F IGURE A.1: Representation of short and long geometrical scales in

a bubbly flow. a) Long scale interfaces, b) short scale interface, c)
presence of both. Source from Márquez Damián, 2013.

To simulate a basic case in OpenFOAM three main directories are needed: 0, constant
and system). The content of each directory depends on the code approach, i.e., for
the compressionFluxFoam code the contents are as follows:

• The 0 directory contains the variables or fields, that are used for the resolution
of the fluid’s motion.

• The constant directory contains all the parameters and constant values that af-
fect the fluid’s motion and the mesh of the employed geometry

• The system directory which contains the controls of the simulations and the
discretization schemes (numeric methods, time step size, time interval writing,
etc.)
160

Figure A.2 to have a complete view of all the directories and files used for a sludge
settling compression case.

F IGURE A.2: General overview and main content of a simulation case

in OpenFOAM

A.1 The zero directory

After the mesh is created in snappyHexMesh, which is the own OpenFOAM meshing
tool, one must set the initial boundary conditions at 0 folder. However, meshes
created with other tools can be used as well.

For all files the dimensions and initial values for the boundaries and internal fields
shall be filled. An example of the epsilon boundary conditions file is shown in figure
A.3. Dimensions in OpenFOAM are always in international system in this order:
kilogram, meter, second, Kelvin, mole, ampere and candela

When simulating a turbulent flow with compressionfluxFoam, six fields or variables

are needed, the volume fraction (alpha.sluge), the field velocity (U), the pressure
(pr gh), the parameters for the turbulent model, in this case (k and ǫ) and the turbu-
lent viscosity (nut).

1. alpha.sludge. One must set the value of the initial volume fraction at the inlet
or internal field. The volume fraction is calculated through equation 2.6. This
is a dimensionless variable.

2. epsilon. If the k-epsilon turbulence model is chosen, the rate at which the turbu-
lence kinetic energy is converted into thermal internal energy (epsilon) initial
condition is needed. Units are m2 .s−3 .
 161

F IGURE A.3: Initial conditions for the epsilon file, dimensions, inter-
nalField and boundaryField shall be filled to start a case

3. k. The kinetic energy generated by the fluid is set up in this file. It has m2 .s−2
units.

4. nut. Stands for the turbulent viscosity field used for the turbulence model. The
units are m2 .s−1 .

5. pr gh. The pressure equation is solved for p_rgh, which is the dynamic pres-
sure, and is equal to the total pressure minus the hydro static pressure (ρ*gh).
It has the pressure unities of kg.m−1 .s−2 .

6. nut. Velocity field is a vector, therefore 3 components (x, y and z) and the sense
(positive or negative) should be filled. If the value of velocity is set to 0 for the
walls it indicates that a no slip condition is imposed. Velocity units are m.s−1

There exist a large variety of boundary conditions according to the field and the
related patch, but they will be explained in chapter for4 when the main case is set
up.

A.2 The constant directory

The directory contains the physical properties and constants used for the different
models: rheology, sedimentation and turbulence and the boundaries for the mesh
(figure A.4).

The description of each file/directory within the folder is as follows:

162

F IGURE A.4: Extract from the boundary file showing the different
types of boundaries and names for the mesh

1. polyMesh. Eight different files are included in the directory. The declaration
of the limits of the mesh is done in the boundary file. The names of the limits
(inlet, outlet, wall, free surfaces. . . ) and the type of boundary (patch, wall,
empty, wedge or symmetry) shall be the same as the ones used for the initial
conditions in the zero directory. OpenFOAM make difference between capital
and small letters. The rest of the files contains the number of points, cells and
faces of the unstructured mesh.

2. transportProperties. The file contains basically the parameters for the rheology
model and the settling velocity model. The rheology model for the activated
sludge can be chosen between Plastic or Bingham. The velocity settling model
can be chosen among Vesilind, Takacs or Diehl equations for hindered settling.
For compression settling the inputs are the values of alpha for effective solids
stress and the critical concentration. See Figure A.5.

3. turbulenceProperties. The turbulence model (RANS, LES or laminar) is declared

in this file. CompressionFluxFoam suggest the using of the buoyant k-epsilon
model, since a density gradient may be created by a variation in the composi-
tion of the mixture. According to OpenFOAM manual the 1/Prt coefficient is
replaced by Cg to provide control of the model.

4. g. This corresponds to the value of the gravity (9.81 m.s−1 ). The sense and
the direction of the gravity force according to the mesh orientation must be
indicated.
 163

F IGURE A.5: Illustration of how to set the parameters for the settling
velocity model

A.3 The system directory

The system directory contains three main files (controlDict, fvSchemes and fvOp-
tions) that will be discussed next. However, it can contain also pre and postpro-
cessing tools for extracting points, velocity or pressure profiles, refining or editing
meshes among other functions. Refer to the OpenFOAM manual for more details in
the different postProcessing tools.

1. controlDict. The input for the simulation time and data writing is defined in this
file (figure A.6). Start and stop time must be set. One advantage in OpenFOAM
is that if a simulation is stopped before the ending time it can be restarted in
the last written time step by using the latestTime option.

The data results are written into csv format and the written time can be set in
the writeInterval option. A new directory will be created for each time Step
and the values of the variables introduced in the time 0 directory and new
variables, such as Udm, dComp and sigma, will be written.

2. fvSchemes.The numerical schemes are specified for the first and second time
derivatives, gradient normal to a cell face, gradient, divergent, Laplacian, and
interpolations terms. When the word “default” is used, OpenFOAM uses the
specified numerical method for all variables. If a variable should be solved
with a different scheme, then it can be specified as in figure A.7.

Interpolations in the variables are made from point to point of the unstructured
mesh is made from cell centre to face centre in OpenFOAM. The surface normal
gradient is evaluated at the cell face from two adjacent cells.

For more details of the different numerical schemes please refer to OpenFOAM
manual.

3. fvSolution. The solutions, tolerances and algorithms are set in within the fvSo-
lution file. Other sub-sections can be found in the file are solvers, relaxation-
Factors and PIMPLE.
164

In the section solver the linear solver for each discretised equation for each
variable (alpha.sludge, U, pr gh, k, epsilon) is defined. Different solvers can
be used, Preconditioned (bi-)conjugate gradient (PCG), Stabilized Precondi-
tioned (bi-) conjugate gradient (PBiCGStab), smoother, generalized geometric-
algebraic multi-grid (GAMG) and diagonal. Refer to OpenFOAM manual for
more information about the solver usage. The tolerances are an indicator of
the accuracy of the simulation, they assure that the residual is small enough
and they must be specified for each solver.

The relaxation factors improve stability in simulation. The Under-relaxation

(alpha) limits how a variable changes from one iteration to the next. The choice
in the value of alpha, ranging from 0 to 1, should be high enough to move
quickly into the next iteration but low enough to ensure stability into the iter-
ative process.

The PIMPLE algorithm combines the PISO and SIMPLE algorithm, i.e., it looks
for the steady-state in each time step, after this, the outer correction loops en-
sure that explicit parts of the equations are converged. If the tolerance criterion
is reached within the steady-state calculation, the outer correction loop is left
and the computation moves on until the end time of the simulation. Large
courant numbers (≫ 1) can be used and therefore time step highly increase.
Refer to Holzmann, 2016 book for further details about the PIMPLE algorithm.

A.4 Processing

Just after setting up all the variables and boundary conditions in the 0 directory, the
parameters in the constant directory, and the time controls and numerical schemes,
one just must type compressionFluxFoam in a terminal to run the case.

Further, OpenFOAM provides the option to run in parallel in order to reduce the
computational time. A file called decomposeParDict has to be integrated into the sys-
tem directory indicating the number of the processors one wants to employ.

Then in a terminal one first types decomposePar and then mpirun -np X compression-
FluxFoam -parallel where X corresponds to the number of processors indicated in the
decomposeParDict file.

CompressionFluxFoam is a derivation of driftFluxFoam, thus the PIMPLE Algo-

rithm is used. To assure the converge of the simulations the residuals of the variables
are monitored, the smallest value is, the more accurate the solution.

The advantage of the PIMPLE Algorithm is that higher Courant Numbers can be
used without compromising the accuracy and convergence of the simulation. Figure
A.8 shows a results of the residuals monitored during the simulation. This can be
done by using foamMonitor -l postprocessing/residuals/0/residuals.dat.
 165

F IGURE A.6: Extract from a controlDict file and the different inputs.

F IGURE A.7: Specified numerical scheme for the gradient of al-

pha.sludge.
166

F IGURE A.8: Specified numerical scheme for the gradient of al-

pha.sludge.

A.5 Post-Processing

For the postprocessing data view or analysis there are different tools one can use,
the main one is the graphical interface (figure A.9) called Paraview and the user
can access by just typing paraFoam in a terminal. Paraview offers a large set of
options where one can see the velocity fields, extract information about data points
and values of variables in each cell, make different cuts of the geometry to analyze
a specific part of the geometry, obtain velocity, pressure or volume fraction profiles,
get the evolution of the variables in time, etc.

F IGURE A.9: Paraview graphical interphase.

 167

Bibliography

Abda, F. et al. (2009). “Ultrasonic device for real-time sewage velocity and suspended
particles concentration measurements”. en. In: Water Science and Technology 60.1,
pp. 117–125. ISSN: 0273-1223. DOI: 10.2166/wst.2009.281. URL: /wst/article/
60/1/117/15850/Ultrasonic-device-for-real-time-sewage-velocity (visited
on 07/03/2019).
Adams, B.M. et al. (2014). Dakota, A Multilevel Parallel Object-Oriented Framework for
Design Optimization, Parameter Estimation, Uncertainty Quantification, and Sensitiv-
ity Analysis: Version 6.0 User’s Manual. Vol. Version 6.3. Sandia Technical Report
SAND2014-4633,
Ahmed, S. R., G. Ramm, and G. Faltin (1984). “Some Salient Features Of The Time-
Averaged Ground Vehicle Wake”. In: SAE Technical Paper. SAE International. DOI:
10.4271/840300. URL: https://doi.org/10.4271/840300.
Al-Sammarraee, M. and A. Chan (2009). “Large-eddy simulations of particle sedi-
mentation in a longitudinal sedimentation basin of a water treatment plant. Part
2: The effects of baffles”. In: Chemical Engineering Journal 152.2–3, pp. 315–321.
ISSN : 1385-8947. DOI: 10 . 1016 / j . cej . 2009 . 01 . 052. URL: https : / / www .
sciencedirect.com/science/article/pii/S1385894709000667 (visited on 01/27/2017).
Anderson, Norval E. and R. H. Gould (1945). “Design of Final Settling Tanks for Ac-
tivated Sludge [with Discussion]”. In: Sewage Works Journal 17.1, pp. 50–65. ISSN:
00969362. URL: http://www.jstor.org/stable/25029971.
Aziz, A. A. et al. (2000). “The characterisation of slurry dewatering”. en. In: Water
Science and Technology 41.8, pp. 9–16. ISSN: 0273-1223, 1996-9732. DOI: 10 . 2166 /
wst.2000.0136. URL: https://iwaponline.com/wst/article/41/8/9/8836/The-
characterisation-of-slurry-dewatering (visited on 10/02/2019).
Biggs, C.A., P. Lant, and M. J. Hounslow (2003). “Modelling the effect of shear history
on activated sludge flocculation”. In: Water Science & Technology 11, pp. 251–257.
Bingham, E. C. (1916). “An investigation of the laws of plastic flow”. en. In: Bulletin
of the Bureau of Standards 13.2, p. 309. ISSN: 0096-8579. DOI: 10.6028/bulletin.304.
URL : https://nvlpubs.nist.gov/nistpubs/bulletin/13/nbsbulletinv13n2p309_
A2b.pdf (visited on 10/12/2019).
Bokil, S.D. and J.K. Bewtra (1972). “Influence of mechani-cal blending on aerobic
digestion of waste activated sludge.” In:
Boller, Markus (1997). “Small wastewater treatment plants - a challenge to wastewa-
ter engineers”. en. In: Water Science and Technology 35.6, pp. 1–12. ISSN: 0273-1223,
168

1996-9732. DOI: 10.2166/wst.1997.0237. URL: https://iwaponline.com/wst/

article/35/6/1-12/5868 (visited on 10/19/2019).
Bratby, J and G.V.R. Marais (1976). “A Guide for the Design of Dissolved air (Pres-
sure) Flotation Systems for Activated Sludge Processes.” In: Water SA, 2 (2), pp. 86–
100.
Brennan, D. (2001). The Numerical Simulation of Two Phase Flows in Settling Tanks. Uni-
versity of London. URL: https://books.google.fr/books?id=nWw-nQEACAAJ.
Brouckaert, C. J. and C. A. Buckley (1999). “The use of Computational Fluid Dynam-
ics for improving the design and operation of water and wastewater treatment
plants”. In: Water Science and Technology 40.4–5, pp. 81–89. ISSN: 0273-1223. DOI:
10 . 1016 / S0273 - 1223(99 ) 00488 - 6. URL: https : / / www . sciencedirect . com /
science/article/pii/S0273122399004886 (visited on 01/27/2017).
Burckbuchler, Marie and Stephan Fischer (2017). “A miniature UVP hardware ap-
plied to environmental monitoring”. In: HydroSenSoft Symposium Proceedings.
Buscall, Richard and Lee R. White (1987). “The consolidation of concentrated sus-
pensions. Part 1.—The theory of sedimentation”. In: Journal of the Chemical Society,
Faraday Transactions 1: Physical Chemistry in Condensed Phases 83.3, pp. 873–891.
Bürger, R., W.l. Wendland, and F. Concha (2000). “Model Equations for Gravitational
Sedimentation-Consolidation Processes”. en. In: ZAMM - Journal of Applied Math-
ematics and Mechanics / Zeitschrift für Angewandte Mathematik und Mechanik 80.2,
pp. 79–92. ISSN: 1521-4001. DOI: 10.1002/(SICI)1521- 4001(200002)80:2<79::
AID- ZAMM79> 3.0.CO;2- Y. URL: http://onlinelibrary.wiley.com/doi/10.
1002/(SICI)1521-4001(200002)80:2<79::AID-ZAMM79>3.0.CO;2-Y/abstract
(visited on 01/27/2017).
Bürger, Raimund, Julio Careaga, and Stefan Diehl (2017). “A simulation model for
settling tanks with varying cross-sectional area”. en. In: Chemical Engineering Com-
munications 204.11, pp. 1270–1281. ISSN: 0098-6445, 1563-5201. DOI: 10.1080/00986445.
2017 . 1360871. URL: https : / / www . tandfonline . com / doi / full / 10 . 1080 /
00986445.2017.1360871 (visited on 10/13/2019).
Bürger, Raimund, Stefan Diehl, and Ingmar Nopens (2011). “A consistent modelling
methodology for secondary settling tanks in wastewater treatment”. In: Water Re-
search 45.6, pp. 2247–2260. ISSN: 0043-1354. DOI: 10.1016/j.watres.2011.01.020.
URL : https://www.sciencedirect.com/science/article/pii/S0043135411000352
(visited on 01/27/2017).
Bürger, Raimund, Kenneth H. Karlsen, and John D. Towers (2005). “A Model of Con-
tinuous Sedimentation of Flocculated Suspensions in Clarifier-Thickener Units”.
In: SIAM Journal on Applied Mathematics 65.3, pp. 882–940. ISSN: 0036-1399. URL:
https://www.jstor.org/stable/4096201 (visited on 05/22/2019).
Bürger, Raimund, Ricardo Ruiz-Baier, and Héctor Torres (2012). “A Stabilized Finite
Volume Element Formulation for Sedimentation-Consolidation Processes”. en. In:
SIAM Journal on Scientific Computing 34.3, B265–B289. ISSN: 1064-8275, 1095-7197.
 169

DOI : 10.1137/110836559. URL : http://epubs.siam.org/doi/10.1137/110836559

(visited on 10/12/2019).
Bürger, Raimund et al. (2012). “On reliable and unreliable numerical methods for
the simulation of secondary settling tanks in wastewater treatment”. In: Com-
puters & Chemical Engineering 41, pp. 93–105. ISSN: 0098-1354. DOI: 10 . 1016 / j .
compchemeng.2012.02.016. URL: https://www.sciencedirect.com/science/
article/pii/S0098135412000725 (visited on 01/27/2017).
Bürger, Raimund et al. (2013). “A consistent modelling methodology for secondary
settling tanks: a reliable numerical method”. en. In: Water Science & Technology
68.1, p. 192. ISSN: 0273-1223. DOI: 10 . 2166 / wst . 2013 . 239. URL: http : / / wst .
iwaponline.com/cgi/doi/10.2166/wst.2013.239 (visited on 10/31/2017).
Cacossa, Kenneth F. and David A. Vaccari (1994). “Calibration of a compressive grav-
ity thickening model from a single batch settling curve”. en. In: Water Science and
Technology 30.8, pp. 107–116. ISSN: 0273-1223, 1996-9732. DOI: 10.2166/wst.1994.
0392. URL: https://iwaponline.com/wst/article/30/8/107-116/27982 (visited
on 10/13/2019).
Carlsson, Bengt (1998). “An introduction to sedimentation theory in wastewater
treatment”. en. In: Systems and Control Group Uppsala University, p. 7.
Cho, S. H. et al. (1993). “Settling velocity model of activated sludge”. In: Water Re-
search 27.7, pp. 1237–1242. ISSN: 0043-1354. DOI: 10.1016/0043-1354(93)90016-B.
URL : http://www.sciencedirect.com/science/article/pii/004313549390016B
(visited on 01/27/2017).
Cole, R.F. (1968). Experimental Evaluation of the Kynch Theory. University of North Car-
olina at Chapel Hill. URL: https://books.google.fr/books?id=K4PLoQEACAAJ.
Concha, F. and M. C. Bustos (1991). “Settling velocities of particulate systems, 6.
Kynch sedimentation processes: batch settling”. In: International Journal of Mineral
Processing 32.3, pp. 193–212. ISSN: 0301-7516. DOI: 10.1016/0301-7516(91)90068-
T. URL: http://www.sciencedirect.com/science/article/pii/030175169190068T
(visited on 07/31/2019).
Copp, J.B. (2001). The COST Simulation Benchmark: Description and Simulator Manual.
Dahl, Claus (1993). Numerical modelling of flow and settling in secondary settling tanks.
Aalborg University, Department of Civil Engineering.
Das, Shankhadeep et al. (2016). “Improving the performance of industrial clarifiers
using three-dimensional computational fluid dynamics”. In: Engineering Applica-
tions of Computational Fluid Mechanics 10.1, pp. 130–144. ISSN: 1994-2060. DOI: 10.
1080/19942060.2015.1121518. URL: http://dx.doi.org/10.1080/19942060.
2015.1121518 (visited on 01/27/2017).
De Clercq, B. (2003). Computational Fluid Dynamics of Settling Tanks: Development of
Experiments and Rheological, Settling, and Scraper Submodels. Belgium: Dept. Applied
Mathematics, Biometrics and Process Control, Ghent University. URL: https://
books.google.fr/books?id=-cACtwAACAAJ.
170

De Clercq, Jeriffa (2006). “Batch and continuous settling of activated sludge: in-depth
monitoring and 1D compression modelling”. eng. PhD thesis. Ghent University.
URL : http://lib.ugent.be/fulltxt/RUG01/001/376/220/RUG01- 001376220_
2010_0001_AC.pdf.
De Clercq, Jeriffa et al. (2005). “Detailed spatio-temporal solids concentration pro-
filing during batch settling of activated sludge using a radiotracer”. In: Water Re-
search 39.10, pp. 2125–2135. ISSN: 0043-1354. DOI: 10.1016/j.watres.2005.03.
023. URL: https://www.sciencedirect.com/science/article/pii/S0043135405001193
(visited on 01/27/2017).
De Clercq, Jeriffa et al. (2008). “Extending and calibrating a mechanistic hindered
and compression settling model for activated sludge using in-depth batch experi-
ments”. In: 42, pp. 781–791. DOI: 10.1016/j.watres.2007.08.040.
Derlon, Nicolas et al. (2017). “Batch settling curve registration via image data mod-
eling”. en. In: Water Research 114, pp. 327–337. ISSN: 00431354. DOI: 10.1016/j.
watres.2017.01.049. URL: https://linkinghub.elsevier.com/retrieve/pii/
S0043135417300568 (visited on 10/14/2019).
Dick, Richard I. and Ben B. Ewing (1967). “The Rheology of Activated Sludge”. In:
Journal (Water Pollution Control Federation) 39.4, pp. 543–560. ISSN: 0043-1303. URL:
https://www.jstor.org/stable/25035907 (visited on 10/10/2018).
Diehl, S. (2015). “Numerical identification of constitutive functions in scalar nonlin-
ear convection–diffusion equations with application to batch sedimentation”. In:
Applied Numerical Mathematics. Fourth Chilean Workshop on Numerical Analysis
of Partial Differential Equations (WONAPDE 2013) 95, pp. 154–172. ISSN: 0168-
9274. DOI: 10.1016/j.apnum.2014.04.002. URL: https://www.sciencedirect.
com/science/article/pii/S0168927414000531 (visited on 01/27/2017).
Ding, A., M.J. Hounslow, and C.A. Biggs (2006). “Population balance modelling of
activated sludge flocculation: Investigating the size dependence of aggregation,
breakage and collision efficiency”. en. In: Chemical Engineering Science 61.1, pp. 63–
74. ISSN: 00092509. DOI: 10.1016/j.ces.2005.02.074. URL: https://linkinghub.
elsevier.com/retrieve/pii/S0009250905004410 (visited on 10/13/2019).
Eikelboom, Dick H. (2000). Process Control of Activated Sludge Plants by Microscopic
Investigation. en. Google-Books-ID: wE8C7SVySbMC. IWA Publishing. ISBN: 978-
1-900222-29-7.
Ekama, G. A. and P. Marais (2004). “Assessing the applicability of the 1D flux the-
ory to full-scale secondary settling tank design with a 2D hydrodynamic model”.
In: Water Research 38.3, pp. 495–506. ISSN: 0043-1354. DOI: 10 . 1016 / j . watres .
2003.10.026. URL: https://www.sciencedirect.com/science/article/pii/
S0043135403005797 (visited on 01/27/2017).
Ekama, GA (1986). “Sludge settleability and secondary settling tank design proce-
dures”. In: Wat. Pollut. Control 85, pp. 101–113.
Ekama, G.A. et al. (1997). Secondary Settling Tanks: Theory, Modelling, Design and Op-
eration. Vol. 6. Richmond, UK: International Association on Water Quality, STR.
 171

Flamant, O. et al. (2004). “Experimental Analysis and Simulation of Settling Pro-

cess”. In: Process Safety and Environmental Protection 82.4, pp. 312–318. ISSN: 0957-
5820. DOI: 10 . 1205 / 095758204323162337. URL: https : / / www . sciencedirect .
com/science/article/pii/S0957582004711787 (visited on 01/27/2017).
François, Pierre et al. (2016). “Experimental study of activated sludge batch set-
tling velocity profile”. In: Flow Measurement and Instrumentation 48, pp. 112–117.
ISSN : 0955-5986. DOI: 10 . 1016 / j . flowmeasinst . 2015 . 08 . 009. URL: https :
//www.sciencedirect.com/science/article/pii/S0955598615300091 (visited
on 01/27/2017).
Fuchs, Andreas and Gernot Staudinger (1999). “Characterising the clarification of
the supernatant of activated sludges”. In: Water Research 33.11, pp. 2527–2534.
ISSN : 0043-1354. DOI: 10 . 1016 / S0043 - 1354(98 ) 00491 - 6. URL: http : / / www .
sciencedirect.com/science/article/pii/S0043135498004916 (visited on 08/01/2019).
G. Chebbo Ph.D. and M.-C. Gromaire, Ph.D. (2009). “VICAS—An Operating Pro-
tocol to Measure the Distributions of Suspended Solid Settling Velocities within
Urban Drainage Samples”. In: Journal of Environmental Engineering 135.9. DOI: 10.
1061/(ASCE)0733-9372(2009)135:9(768). URL: http://ascelibrary.org/doi/
abs/10.1061/%28ASCE%290733- 9372%282009%29135%3A9%28768%29 (visited on
01/27/2017).
Gao, Haiwen and M. K. Stenstrom (2018). “Evaluation of three turbulence models
in predicting the steady state hydrodynamics of a secondary sedimentation tank”.
In: Water Research 143, pp. 445 –456. ISSN: 0043-1354. DOI: https://doi.org/10.
1016/j.watres.2018.06.067. URL: http://www.sciencedirect.com/science/
article/pii/S004313541830527X.
Ghawi, Ali and J. Kriš (2012). “Improvement performance of secondary clarifiers
by a computational fluid dynamics model”. In: Slovak Journal of Civil Engineering
19.4, pp. 1–11. ISSN: 1210-3896. DOI: 10.2478/v10189- 011- 0017- 9. URL: https:
//www.degruyter.com/view/j/sjce.2011.xix.issue- 4/v10189- 011- 0017-
9/v10189-011-0017-9.xml (visited on 01/27/2017).
Gong, M. et al. (2011). “Development of a flocculation sub-model for a 3-D CFD
model based on rectangular settling tanks”. en. In: Water Science and Technology
63.2, pp. 213–219. ISSN: 0273-1223, 1996-9732. DOI: 10.2166/wst.2011.035. URL:
https://iwaponline.com/wst/article/63/2/213/13783/Development-of-a-
flocculation-submodel-for-a-3D (visited on 10/13/2019).
Goula, Athanasia M. et al. (2008). “A CFD methodology for the design of sedimen-
tation tanks in potable water treatment: Case study: The influence of a feed flow
control baffle”. In: Chemical Engineering Journal 140.1–3, pp. 110–121. ISSN: 1385-
8947. DOI: 10.1016/j.cej.2007.09.022. URL: https://www.sciencedirect.com/
science/article/pii/S1385894707006250 (visited on 01/27/2017).
Greenshields, Chris (2017). OpenFOAM User Guide. OpenFOAM. URL: https://cfd.
direct/openfoam/user-guide/.
172

Griborio, Alonso (2004). “Secondary Clarifier Modeling: A Multi-Process Approach”.

PhD thesis. University of New Orleans Theses and Dissertations.
Griborio, Alonso and Alex J. McCorquodale (2006). “Optimum Design of your Cen-
ter Well: Use of a CFD Model to understand the Balance between Flocculation
and improved Hydrodynamics”. en. In: Proceedings of the Water Environment Fed-
eration 2006.13, pp. 263–280. ISSN: 1938-6478. DOI: 10.2175/193864706783710587.
URL : http://www.ingentaconnect.com/content/10.2175/193864706783710587
(visited on 10/13/2019).
Grijspeerdt, Koen, Peter Vanrolleghem, and Willy Verstraete (1995). “Selection of
one-dimensional sedimentation: Models for on-line use”. en. In: Water Science and
Technology. Modelling and Control of Activated Sludge Processes 31.2, pp. 193–
204. ISSN: 0273-1223. DOI: 10 . 1016 / 0273 - 1223(95 ) 00192 - P. URL: http : / /
www.sciencedirect.com/science/article/pii/027312239500192P (visited on
10/18/2019).
Guo, Lisha et al. (2009). “An experimental study of low concentration sludge set-
tling velocity under turbulent condition”. In: Water Research 43.9, pp. 2383–2390.
ISSN: 0043-1354. DOI: 10 . 1016 / j . watres . 2009 . 02 . 032. URL: https : / / www .
sciencedirect.com/science/article/pii/S0043135409001353 (visited on 01/27/2017).
Gupta, Hoshin V. et al. (2009). “Decomposition of the mean squared error and NSE
performance criteria: Implications for improving hydrological modelling”. en. In:
Journal of Hydrology 377.1-2, pp. 80–91. ISSN: 00221694. DOI: 10.1016/j.jhydrol.
2009.08.003. URL: http://linkinghub.elsevier.com/retrieve/pii/S0022169409004843
(visited on 09/10/2018).
Holzmann, Tobias (2016). Mathematics, Numerics, Derivations and OpenFOAM®. en.
Holzmann CFD. DOI: 10.13140/RG.2.2.27193.36960. URL: http://rgdoi.net/
10.13140/RG.2.2.27193.36960 (visited on 03/20/2019).
Hreiz, Rainier et al. (2019). “CFD Investigation of the effects of bubble aerator lay-
outs on hydrodynamics of an activated sludge channel reactor”. en. In: Environ-
mental Technology 40.20, pp. 2657–2670. ISSN: 0959-3330, 1479-487X. DOI: 10.1080/
09593330.2018.1448001. URL: https://www.tandfonline.com/doi/full/10.
1080/09593330.2018.1448001 (visited on 10/18/2019).
Härtel, L. and H. J. Pöpel (1992). “A Dynamic Secondary Clarifier Model including
Processes of Sludge Thickening”. en. In: Water Science and Technology 25.6, pp. 267–
284. ISSN: 0273-1223, 1996-9732. DOI: 10 . 2166 / wst . 1992 . 0128. URL: https : / /
iwaponline.com/wst/article/25/6/267-284/27799 (visited on 10/18/2019).
Isenmann, Gilles (2016). “Approche Euler-Lagrange pour la modélisation du trans-
port solide dans les ouvrages de décantation”. fr. In: URL: https://tel.archives-
ouvertes.fr/tel-01490384 (visited on 10/09/2019).
Ishii, M. and K. Mishima (1984). “Two-fluid model and hydrodynamic constitutive
relations”. In: Nuclear Engineering and Design 82.2–3, pp. 107–126. ISSN: 0029-5493.
DOI : 10.1016/0029-5493(84)90207-3. URL: https://www.sciencedirect.com/
science/article/pii/0029549384902073 (visited on 01/27/2017).
 173

Ishii, Mamoru and Takashi Hibiki (1975). Thermo-fluid Dynamics of Two-Phase Flow.
en. Springer US. ISBN: 978-0-387-28321-0. URL: https://www.springer.com/gp/
book/9780387283210 (visited on 03/27/2019).
Jareteg, Adam, Alejandro Lopez, and Olivier Petit. “Coupling of Dakota and Open-
FOAM for automatic parameterized optimization”. en. In: p. 22.
Jimenez, Jose et al. (2015). “High-rate activated sludge system for carbon manage-
ment – Evaluation of crucial process mechanisms and design parameters”. en. In:
Water Research 87, pp. 476–482. ISSN: 00431354. DOI: 10.1016/j.watres.2015.07.
032. URL: https://linkinghub.elsevier.com/retrieve/pii/S0043135415301342
(visited on 10/10/2019).
Judd, S.J. and C. Judd (2011). The MBR Book. URL: https://www.scopus.com/inward/
record.uri?eid=2-s2.0-84949176564&partnerID=40&md5=f0059fe0a2915fe12c94ead73ff6c47e.
Kahane, R, T Nguyen, and M.P Schwarz (2002). “CFD modelling of thickeners at
Worsley Alumina Pty Ltd”. en. In: Applied Mathematical Modelling 26.2, pp. 281–
296. ISSN: 0307904X. DOI: 10 . 1016 / S0307 - 904X(01 ) 00061 - 0. URL: https : / /
linkinghub.elsevier.com/retrieve/pii/S0307904X01000610 (visited on 10/12/2019).
Kahane, R.B., M. P Schwarz, and R.M. Johnston (1997). Proceedings of the Computa-
tional Fluid Dynamics in Minerals and Metal Processing and Power Generation, CSIRO.
Melbourne, Australia. URL: http : / / www . cfd . com . au / cfd _ conf97 / papers /
kah020.pdf.
Khalili-Garakani, Amirhossein et al. (2011). “Comparison between different models
for rheological characterization of activated sludge”. In: Iranian Journal of Environ-
mental Health Science & Engineering 8, pp. 255–264.
Kim, H. S. et al. (2005). “Study of flow characteristics in a secondary clarifier by nu-
merical simulation and radioisotope tracer technique”. In: Applied Radiation and
Isotopes 63.4, pp. 519–526. ISSN: 0969-8043. DOI: 10.1016/j.apradiso.2005.03.
016. URL: https://www.sciencedirect.com/science/article/pii/S0969804305000916
(visited on 01/27/2017).
Kinnear, D.J. and K. Deines (2001). “Acoustic Doppler current profiler: clarifier ve-
locity measurement. Atlanta, USA.” In: Proceedings WEFTEC 2001,
Koopman, Ben and Keith Cadee (1983). “Prediction of thickening capacity using
diluted sludge volume index”. en. In: Water Research 17.10, pp. 1427–1431. ISSN:
00431354. DOI: 10 . 1016 / 0043 - 1354(83 ) 90274 - 9. URL: https : / / linkinghub .
elsevier.com/retrieve/pii/0043135483902749 (visited on 10/18/2019).
Krebs, Peter et al. (1996). “Influence of inlet and outlet configuration on the flow in
secondary clarifiers”. In: Water Science and Technology. Water Quality International
’96 Part 3: Modelling of Activated Sludge Processes; Microorganisms in Activated
Sludge and Biofilm Processes; Anareobic Biological Treatment; Biofouling 34.5,
pp. 1–9. ISSN: 0273-1223. DOI: 10 . 1016 / 0273 - 1223(96 ) 00622 - 1. URL: http :
//www.sciencedirect.com/science/article/pii/0273122396006221 (visited on
08/01/2019).
174

Kynch, G. J. (1952). “A theory of sedimentation”. en. In: Transactions of the Faraday So-
ciety 48.0, pp. 166–176. ISSN: 0014-7672. DOI: 10.1039/TF9524800166. URL: https:
//pubs.rsc.org/en/content/articlelanding/1952/tf/tf9524800166 (visited
on 07/31/2019).
Lakehal, Djamel et al. (1999). “Computing shear flow and sludge blanket in sec-
ondary clarifiers”. In: Journal of Hydraulic Engineering 125.3, pp. 253–262. URL:
http : / / ascelibrary . org / doi / abs / 10 . 1061 / (ASCE ) 0733 - 9429(1999 ) 125 :
3(253) (visited on 03/24/2017).
Larsen, P (1977). On the Hydraulics of Rectangular Settling Basins. Tech. rep. Lund,
Sweden.: Dept of Water Res. Engrg. Lind University.
Laurent, J. et al. (2014). “A protocol for the use of computational fluid dynamics as
a supportive tool for wastewater treatment plant modelling”. en. In: Water Science
and Technology 70.10, pp. 1575–1584. ISSN: 0273-1223, 1996-9732. DOI: 10 . 2166 /
wst.2014.425. URL: http://wst.iwaponline.com/content/70/10/1575 (visited
on 01/27/2017).
Li, Ben and M. K. Stenstrom (2014). “Research advances and challenges in one-
dimensional modeling of secondary settling Tanks – A critical review”. In: Wa-
ter Research 65, pp. 40–63. ISSN: 0043-1354. DOI: 10.1016/j.watres.2014.07.007.
URL : https://www.sciencedirect.com/science/article/pii/S004313541400503X
(visited on 01/27/2017).
Locatelli, F. et al. (2015). “Detailed Velocity and Concentration Profiles Measurement
During Activated Sludge Batch Settling Using an Ultrasonic Transducer”. In: Sep-
aration Science and Technology 50.7, pp. 1059–1065. ISSN: 0149-6395. DOI: 10.1080/
01496395.2014.980002. URL: http://dx.doi.org/10.1080/01496395.2014.
980002 (visited on 01/27/2017).
Locatelli, Florent (2015). “Activated sludge batch settling : from experimental inves-
tigation using an ultrasonic transducer to 1D modelling, sensitivity analysis and
parameter identification”. Theses. Université de Strasbourg. URL: https://tel.
archives-ouvertes.fr/tel-01404066.
Locatelli, Florent et al. (2013). “Impact de la loi de comportement rhéologique sur la
sédimentation en batch de boues activées : approche expérimentale et numérique”.
fr. In: La Houille Blanche 4, pp. 31–36. ISSN: 0018-6368, 1958-5551. DOI: 10.1051/
lhb / 2013030. URL: http : / / dx . doi . org / 10 . 1051 / lhb / 2013030 (visited on
01/27/2017).
Mancell-Egala, William A. S. K. et al. (2016). “Limit of stokesian settling concen-
tration characterizes sludge settling velocity”. In: Water Research 90, pp. 100–110.
ISSN: 0043-1354. DOI: 10 . 1016 / j . watres . 2015 . 12 . 007. URL: https : / / www .
sciencedirect.com/science/article/pii/S0043135415304073 (visited on 01/27/2017).
Mancell-Egala, William A. S. K. et al. (2017). “Settling regimen transitions quan-
tify solid separation limitations through correlation with floc size and shape”.
In: Water Research 109, pp. 54–68. ISSN: 0043-1354. DOI: 10 . 1016 / j . watres .
 175

2016.10.080. URL: https://www.sciencedirect.com/science/article/pii/

S0043135416308399 (visited on 01/27/2017).
Manninen, Mikko, Veikko Taivassalo, and Sirpa Kallio (1996). “On the mixture model
for multiphase flow”. en. In: VTT Publications. Espoo Technical Research Centre of
Finland, 288. 67 p. URL: https://docs.google.com/document/d/185xhiG3cKxMz2IDOdADQlDFMEz5ZuGirS
edit?usp=drive_web&ouid=105725106488157656131&usp=embed_facebook (vis-
ited on 03/27/2019).
Marchisio, D.L. Fox (2013). “Computational Models for Polydisperse Particulate and
Multiphase Systems”. In:
Matko, T. et al. (1996). “Recent Progress in the Numerical Modelling of Wastewater
Sedimentation Tanks”. In: Process Safety and Environmental Protection 74.4, pp. 245–
258. ISSN: 0957-5820. DOI: 10.1205/095758296528590. URL: http://www.sciencedirect.
com/science/article/pii/S0957582096706274 (visited on 01/27/2017).
McCorquodale, J.A. (1976). Hydraulic study of the circular settling tanks at the West
Windsor Pollution Control Plant. Windsor, Ontario, Canada.: Rep., University of
Windsor,
Moriasi D. N. et al. (2015). “Hydrologic and Water Quality Models: Performance
Measures and Evaluation Criteria”. In: Transactions of the ASABE 6, pp. 1763–1785.
ISSN: 21510032, 21510040. DOI : 10.13031/trans.58.10715. URL : http://elibrary.
asabe.org/abstract.asp?aid=46548&t=3&dabs=Y&redir=&redirType= (visited
on 08/23/2018).
Moukalled, F., L. Mangani, and M. Darwish (2016). The Finite Volume Method in Com-
putational Fluid Dynamics. Vol. 113. Fluid Mechanics and Its Applications. Cham:
Springer International Publishing. ISBN: 978-3-319-16873-9 978-3-319-16874-6. DOI:
10.1007/978-3-319-16874-6. URL: http://link.springer.com/10.1007/978-
3-319-16874-6 (visited on 04/18/2018).
Márquez Damián, Santiago (2013). “An extended mixture model for the simultane-
ous treatment of short and long scale interfaces”. eng. In: URL: https://bibliotecavirtual.
unl.edu.ar/tesis//handle/11185/489 (visited on 10/12/2019).
Nopens, Ingmar (2005). “Modelling the activated sludge flocculation process: a pop-
ulation balance approach”. nl. OCLC: 66401508. PhD thesis. S.l.: s.n.]
Nopens, Ingmar et al. (2015). “Population balance models: a useful complementary
modelling framework for future WWTP modelling”. en. In: Water Science and Tech-
nology 71.2, pp. 159–167. ISSN: 0273-1223, 1996-9732. DOI: 10.2166/wst.2014.500.
URL : https://iwaponline.com/wst/article/71/2/159/18775/Population-
balance-models-a-useful-complementary (visited on 10/13/2019).
Otterpohl, R. and M. Freund (1992). “Dynamic Models for Clarifiers of Activated
Sludge Plants with Dry and Wet Weather Flows”. en. In: Water Science and Tech-
nology 26.5-6, pp. 1391–1400. ISSN: 0273-1223, 1996-9732. DOI: 10.2166/wst.1992.
0582. URL: https://iwaponline.com/wst/article/26/5- 6/1391- 1400/26988
(visited on 10/13/2019).
176

Pallares, A. et al. (2017). “Suspended Sediment Monitoring: Comparison between

Optical and Acoustical Turbidity”. In: International Conference on Urban Drainage.
Parker, D. S., D. J. Kinnear, and E. J. Wahlberg (2001). “Review of Folklore in Design
and Operation of Secondary Clarifiers”. en. In: Journal of Environmental Engineer-
ing 127.6, pp. 476–484. ISSN: 0733-9372, 1943-7870. DOI: 10 . 1061 / (ASCE ) 0733 -
9372(2001)127:6(476). URL: http://ascelibrary.org/doi/10.1061/%28ASCE%
290733-9372%282001%29127%3A6%28476%29 (visited on 10/18/2019).
Parker, Denny S., Warren J. Kaufman, and David Jenkins (1972). “Floc Breakup in
Turbulent Flocculation Processes”. eng. In: Journal of the Sanitary Engineering Di-
vision 98.1, pp. 79–99. URL: https://cedb.asce.org/CEDBsearch/record.jsp?
dockey=0127407 (visited on 10/13/2019).
Patziger, M. (2016). “Computational fluid dynamics investigation of shallow circular
secondary settling tanks: Inlet geometry and performance indicators”. In: Chemical
Engineering Research and Design 112, pp. 122–131. ISSN: 0263-8762. DOI: 10.1016/j.
cherd.2016.06.018. URL: https://www.sciencedirect.com/science/article/
pii/S0263876216301502 (visited on 01/27/2017).
Patziger, M. et al. (2012). “Influence of secondary settling tank performance on sus-
pended solids mass balance in activated sludge systems”. In: Water Research 46.7,
pp. 2415–2424. ISSN: 0043-1354. DOI: 10 . 1016 / j . watres . 2012 . 02 . 007. URL:
http : / / www . sciencedirect . com / science / article / pii / S0043135412000954
(visited on 10/03/2019).
Phillips, O. M. (1957). “On the generation of waves by turbulent wind”. en. In: Jour-
nal of Fluid Mechanics 2.05, p. 417. ISSN: 0022-1120, 1469-7645. DOI: 10 . 1017 /
S0022112057000233. URL: http : / / www . journals . cambridge . org / abstract _
S0022112057000233 (visited on 07/31/2019).
Pitman, AR (1984). “Operation of biological nutrient removal plants”. In: Theory,
Design and Operation of Nutrient Removal Activated Sludge Processes 11.111, p. 16.
Plósz, Benedek Gy. et al. (2007). “One-dimensional modelling of the secondary clarifier-
factors affecting simulation in the clarification zone and the assessment of the
thickening flow dependence”. en. In: Water Research 41.15, pp. 3359–3371. ISSN:
00431354. DOI: 10 . 1016 / j . watres . 2007 . 03 . 007. URL: http : / / linkinghub .
elsevier.com/retrieve/pii/S0043135407001820 (visited on 03/24/2017).
Ramalingam, K. et al. (2009). “Development and Validation of a 3-Dimensional Com-
putational Fluid Dynamics (CFD) Model for Rectangular Settling Tanks in New
York City Water Pollution Control Plants”. In: Water Practice and Technology 4.1,
wpt2009015. URL: http : / / wpt . iwaponline . com / content / 4 / 1 / wpt2009015 .
abstract (visited on 07/25/2017).
Ramin, Elham et al. (2014a). “A new settling velocity model to describe secondary
sedimentation”. In: Water Research 66, pp. 447–458. ISSN: 0043-1354. DOI: 10.1016/
j . watres . 2014 . 08 . 034. URL: https : / / www . sciencedirect . com / science /
article/pii/S0043135414006010 (visited on 01/27/2017).
 177

Ramin, Elham et al. (2014b). “Modelling of secondary sedimentation under wet-

weather and filamentous bulking conditions”. PhD thesis. DTU Environment.
Ramkrishna, D (2000). Population Balances: Theory and Applications to Particulate Sys-
tems in Engineering. London, U.K.: Academic Press.
Rao, M. Anandha (2014). Rheology of Fluid, Semisolid, and Solid Foods. Food Engineer-
ing Series. Boston, MA: Springer US. ISBN: 978-1-4614-9229-0 978-1-4614-9230-6.
DOI : 10.1007/978-1-4614-9230-6. URL: http://link.springer.com/10.1007/
978-1-4614-9230-6 (visited on 10/18/2019).
Ratkovich, N. et al. (2013). “Activated sludge rheology: A critical review on data
collection and modelling”. In: Water Research 47.2, pp. 463–482. ISSN: 0043-1354.
DOI : 10.1016/j.watres.2012.11.021. URL: https://www.sciencedirect.com/
science/article/pii/S0043135412008330 (visited on 01/27/2017).
Sagnella, G.A. (1985). “Model fitting, parameter estimation, linear and non-linear
regression”. en. In: Trends in Biochemical Sciences 10.3, pp. 100–103. ISSN: 09680004.
DOI : 10.1016/0968- 0004(85)90261- 0. URL: https://linkinghub.elsevier.
com/retrieve/pii/0968000485902610 (visited on 09/07/2019).
Samstag, R. W. et al. (2016). “CFD for wastewater treatment: an overview”. In: Water
Science and Technology 74.3, pp. 549–563. ISSN: 0273-1223. DOI: 10.2166/wst.2016.
249. URL: http://wst.iwaponline.com/content/74/3/549.
Samstag, Randal W. et al. (2012). “Field and CFD Analysis of Jet Aeration and Mix-
ing”. en. In: Proceedings of the Water Environment Federation 2012.12, pp. 4113–4139.
ISSN : 19386478. DOI: 10 . 2175 / 193864712811708301. URL: http : / / openurl .
ingenta.com/content/xref?genre=article&issn=1938- 6478&volume=2012&
issue=12&spage=4113 (visited on 10/09/2019).
Sari Erkan, Hanife, Nouha Bakaraki Turan, and Güleda Önkal Engin (2018). “Chap-
ter Five - Membrane Bioreactors for Wastewater Treatment”. In: Comprehensive
Analytical Chemistry. Ed. by Dotse Selali Chormey et al. Vol. 81. Fundamentals
of Quorum Sensing, Analytical Methods and Applications in Membrane Bioreac-
tors. Elsevier, pp. 151–200. DOI: 10 . 1016 / bs . coac . 2018 . 02 . 002. URL: http :
//www.sciencedirect.com/science/article/pii/S0166526X18300035 (visited
on 07/23/2019).
Smethurst, G (1992). Basic Water Treatment for Application Worldwide. Second. Thomas
Telford.
Stamou, Anastasios and Anthoula Gkesouli (2015). “Modeling settling tanks for wa-
ter treatment using computational fluid dynamics”. In: pp. 745–762. DOI: 10.2166/
hydro.2015.069.
Stricker, A.-E., I. Takács, and A. Marquot (2007). “Hindered and compression set-
tling: parameter measurement and modelling”. en. In: Water Science & Technology
56.12, p. 101. ISSN: 0273-1223. DOI: 10.2166/wst.2007.786. URL: http://wst.
iwaponline.com/cgi/doi/10.2166/wst.2007.786 (visited on 10/31/2017).
178

Sweeney, Dennis J., Thomas A. Williams, and David R. Anderson (2019). Statistics -
Experimental design. en. URL: https://www.britannica.com/science/statistics
(visited on 04/30/2019).
Taebi-Harandy, Amir and Edward D. Schroeder (2000). “Formation of density cur-
rents in secondary clarifier”. In: Water Research 34.4, pp. 1225–1232. ISSN: 0043-
1354. DOI: 10.1016/S0043-1354(99)00261-4. URL: http://www.sciencedirect.
com/science/article/pii/S0043135499002614 (visited on 07/31/2019).
Takács, I., G. G. Patry, and D. Nolasco (1991). “A dynamic model of the clarification-
thickening process”. In: Water Research 25.10, pp. 1263–1271. ISSN: 0043-1354. DOI:
10 . 1016 / 0043 - 1354(91 ) 90066 - Y. URL: http : / / www . sciencedirect . com /
science/article/pii/004313549190066Y (visited on 04/23/2018).
Tamayol, A., B. Firoozabadi, and M. A. Ashjari (2010). “Hydrodynamics of Sec-
ondary Settling Tanks and Increasing Their Performance Using Baffles”. en. In:
Journal of Environmental Engineering 136.1, pp. 32–39. ISSN: 0733-9372, 1943-7870.
DOI : 10.1061/(ASCE)EE.1943- 7870.0000126. URL: http://ascelibrary.org/
doi/10.1061/%28ASCE%29EE.1943-7870.0000126 (visited on 03/09/2019).
Tao, Junjie et al. (2012). “Composition of Waste Sludge from Municipal Wastewater
Treatment Plant”. In: Procedia Environmental Sciences. 2011 International Confer-
ence of Environmental Science and Engineering 12, pp. 964–971. ISSN: 1878-0296.
DOI : 10.1016/j.proenv.2012.01.372. URL: http://www.sciencedirect.com/
science/article/pii/S1878029612003738 (visited on 07/24/2019).
Tarpagkou, Roza and Asterios Pantokratoras (2013). “CFD methodology for sedi-
mentation tanks: The effect of secondary phase on fluid phase using DPM cou-
pled calculations”. en. In: Applied Mathematical Modelling 37.5, pp. 3478–3494. ISSN:
0307904X. DOI: 10 . 1016 / j . apm . 2012 . 08 . 011. URL: https : / / linkinghub .
elsevier.com/retrieve/pii/S0307904X12004672 (visited on 10/18/2019).
Tchobanoglous, G, F Burton, and D Stensel (2003). Wastewater Engineering: Treatment
and Resource. 4th ed. Metcalf & Eddy, Inc. New York, N.Y.: McGraw-Hill.
Tebbutt, T. H. Y. (2001). Fundamentos de control de la calidad del agua. es. Google-Books-
ID: q49iPAAACAAJ. Limusa. ISBN: 978-968-18-3317-6.
Thorne, P.D. and Hanes (2002). “A review of acoustic measurement of small-scalesediment
processes.” In: Continental shelf research, 22(4), pp. 603–632.
Torfs, E. et al. (2015a). “Impact on sludge inventory and control strategies using the
benchmark simulation model no. 1 with the Bürger-Diehl settler model”. eng. In:
Water Science and Technology: A Journal of the International Association on Water Pol-
lution Research 71.10, pp. 1524–1535. ISSN: 0273-1223. DOI: 10.2166/wst.2015.122.
Torfs, Elena et al. (2013). “Towards improved 1-D settler modelling: calibration of
the Bürger model and case study”. eng. In: URL: http://lup.lub.lu.se/record/
4174878 (visited on 07/03/2019).
 179

Torfs, Elena et al. (2015b). “Impact of the flocculation state on hindered and compres-
sion settling: experimental evidence and overview of available modelling frame-
works”. eng. In: Systems Analysis and Integrated Assessment, 9th IWA Symposium,
Papers. Gold Coast, QLD, Australia: International Water Association (IWA), p. 4.
Torfs, Elena et al. (2016). “Concentration-driven models revisited: towards a uni-
fied framework to model settling tanks in water resource recovery facilities”. en.
In: Water Science and Technology, wst2016485. ISSN: 0273-1223, 1996-9732. DOI: 10.
2166/wst.2016.485. URL: http://wst.iwaponline.com/content/early/2016/
10/20/wst.2016.485 (visited on 01/27/2017).
Torfs, Elena et al. (2017). “On constitutive functions for hindered settling velocity in
1-D settler models: Selection of appropriate model structure”. In: Water Research
110, pp. 38–47. ISSN: 0043-1354. DOI: 10 . 1016 / j . watres . 2016 . 11 . 067. URL:
https://www.sciencedirect.com/science/article/pii/S0043135416309290
(visited on 01/27/2017).
Ueberl, Judith (1995). Verbesserung der Absetzwirkung von Nachklärbecken. Deutsch.
Zurich.
Ungarish, Marius (1995). “Hydrodynamics of Suspensions.” en. In: Springer-Verlag
290, pp. 406–408. ISSN: 1469-7645, 0022-1120. DOI: 10.1017/S0022112095212576.
URL : https : / / www . cambridge . org / core / journals / journal - of - fluid -
mechanics/article/hydrodynamics-of-suspensions-by-m-ungarish-springer-
1993-322-pp-dm-138/A9CEDC1EA5F9554CD524C365C50DD4DC (visited on 10/12/2019).
Vahedi, Arman and Beata Gorczyca (2012). “Predicting the settling velocity of flocs
formed in water treatment using multiple fractal dimensions”. In: Water Research
46.13, pp. 4188–4194. ISSN: 0043-1354. DOI: 10.1016/j.watres.2012.04.031. URL:
https://www.sciencedirect.com/science/article/pii/S0043135412002862
(visited on 01/27/2017).
– (2014). “Settling velocities of multifractal flocs formed in chemical coagulation
process”. en. In: Water Research 53, pp. 322–328. ISSN: 00431354. DOI: 10 . 1016 /
j.watres.2014.01.008. URL: https://linkinghub.elsevier.com/retrieve/
pii/S004313541400030X (visited on 10/14/2019).
Valle Medina, M.E. and J. Laurent (2020). “Incorporation of a compression term in
a CFD model based on the mixture approach to simulate activated sludge sedi-
mentation”. en. In: Applied Mathematical Modelling 77, pp. 848–860. ISSN: 0307904X.
DOI : 10.1016/j.apm.2019.08.008. URL: https://linkinghub.elsevier.com/
retrieve/pii/S0307904X19304962 (visited on 10/07/2019).
Van Marle, C. and C. Kranenburg (1994). “Effects of Gravity Currents in Circular
Secondary Clarifiers”. en. In: Journal of Environmental Engineering 120.4, pp. 943–
960. ISSN: 0733-9372, 1943-7870. DOI: 10 . 1061 / (ASCE ) 0733 - 9372(1994 ) 120 :
4(943). URL: http://ascelibrary.org/doi/10.1061/%28ASCE%290733- 9372%
281994%29120%3A4%28943%29 (visited on 08/01/2019).
Vanderhasselt, A. et al. (1999). Sludge Settling Characterisation With An Automated Set-
tlometer.
180

Verloop, W. C. (1995). “The inertial coupling force”. In: International Journal of Mul-
tiphase Flow 21.5, pp. 929 –933. ISSN: 0301-9322. DOI: https : / / doi . org / 10 .
1016/0301-9322(95)00029-W. URL: http://www.sciencedirect.com/science/
article/pii/030193229500029W.
Versteeg, H.K. and W. Malalasekera (2007). An Introduction to Computational Fluid
Dynamics: The Finite Volume Approach. Longman Scientific & Technical. ISBN: 978-
0-582-21884-0. URL: https://books.google.fr/books?id=7UynjgEACAAJ.
Vesilind, P. A. (1968). “Theoretical considerations: Design of prototype thickeners
from batch settling tests.” In: Water & Sewage Works, pp. 115–302.
Von Sperling, Marcos (2007). Activated sludge and aerobic biofilm reactors. en. Biolog-
ical wastewater treatment series Marcos von Sperling ; Vol. 5. OCLC: 255801712.
London: IWA Publ. ISBN: 978-1-84339-165-4.
Voutchkov, Nikolay (2005). Settling Tank Design – Let’s settle the matter. en. 10.1002/047147844X.mw506.
URL : https://www.researchgate.net/publication/284774604_The_activated_
sludge_process_Part_1_-_Steady_state_behaviour (visited on 07/23/2019).
Weiss, Michael et al. (2007). “Suction-lift sludge removal and non-Newtonian flow
behaviour in circular secondary clarifiers: Numerical modelling and measurements”.
In: Chemical Engineering Journal 132.1–3, pp. 241–255. ISSN: 1385-8947. DOI: 10 .
1016/j.cej.2007.01.004. URL: https://www.sciencedirect.com/science/
article/pii/S1385894707000101 (visited on 01/27/2017).
Wells, Scott A. and David M. LaLiberte (1998). “Winter temperature gradients in
circular clarifiers”. en. In: Water Environment Research 70.7, pp. 1274–1279. ISSN:
10614303. DOI: 10.2175/106143098X123642. URL: http://doi.wiley.com/10.
2175/106143098X123642 (visited on 08/01/2019).
Wicklein, Edward et al. (2015). “Good modelling practice in applying computational
fluid dynamics for WWTP modelling”. en. In: Water Science and Technology, wst2015565.
ISSN: 0273-1223, 1996-9732. DOI : 10.2166/wst.2015.565. URL : https://iwaponline.
com/wst/article/73/5/969-982/18977 (visited on 10/09/2019).
Wicklein, Edward A. and Randal W. Samstag (2009). “Comparing Commercial and
Transport CFD Models for Secondary Sedimentation”. en. In: Proceedings of the Wa-
ter Environment Federation 2009.10, pp. 6066–6081. ISSN: 1938-6478. DOI: 10.2175/
193864709793952765. URL: http://www.ingentaconnect.com/content/10.2175/
193864709793952765 (visited on 10/09/2019).
Wilson, T.E. (1996). “A new approach to interpreting settling data.” In: Proceeding
of the 69th Annual WEF Conference and Exposition. WEF Alexandria, Virginia 1,
pp. 491–497.
Xanthos, S. et al. (2013). “Implementation of CFD modeling in the performance as-
sessment and optimization of secondary clarifiers: the PVSC case study”. en. In:
Water Science & Technology 68.9, p. 1901. ISSN: 0273-1223. DOI: 10.2166/wst.2013.
280. URL: http://wst.iwaponline.com/cgi/doi/10.2166/wst.2013.280 (visited
on 07/25/2017).
 181

Xanthos, Savvas et al. (2011). “Performance Assessment of Secondary Settling Tanks

Using CFD Modeling”. en. In: Water Resources Management 25.4, pp. 1169–1182.
ISSN : 0920-4741, 1573-1650. DOI: 10 . 1007 / s11269 - 010 - 9620 - 1. URL: http :
/ / link . springer . com / article / 10 . 1007 / s11269 - 010 - 9620 - 1 (visited on
01/27/2017).
Xie, Hao et al. (2014). “Simulation of flow field and sludge settling in a full-scale
oxidation ditch by using a two-phase flow CFD model”. In: Chemical Engineering
Science 109, pp. 296–305.
Xu, Shuang et al. (2018). “Analysis of Bacterial Community Structure of Activated
Sludge from Wastewater Treatment Plants in Winter”. In: BioMed Research Inter-
national 2018. ISSN: 2314-6133. DOI: 10.1155/2018/8278970. URL: https://www.
ncbi.nlm.nih.gov/pmc/articles/PMC5863335/ (visited on 07/24/2019).
Yang, Fei et al. (2009). “Yield stress and rheological characteristics of activated sludge
in an airlift membrane bioreactor”. English. In: Journal of Membrane Science 334.1-2,
pp. 83–90. DOI: 10.1016/j.memsci.2009.02.022.
Zhang, Daijun et al. (2006). “A method for characterizing the complete settling pro-
cess of activated sludge”. In: Water Research 40.14, pp. 2637–2644. ISSN: 0043-1354.
DOI : 10.1016/j.watres.2006.05.017. URL: https://www.sciencedirect.com/
science/article/pii/S0043135406003034 (visited on 01/27/2017).
Zhiyin, Yang (2015). “Large-eddy simulation: Past, present and the future”. In: Chi-
nese Journal of Aeronautics 28.1, pp. 11–24. ISSN: 1000-9361. DOI: 10.1016/j.cja.
2014.12.007. URL: https://www.sciencedirect.com/science/article/pii/
S1000936114002064 (visited on 01/27/2017).
Zhou, Siping, J. A. McCorquodale, and A. M. Godo (1994). “Short Circuiting and
Density Interface in Primary Clarifiers”. en. In: Journal of Hydraulic Engineering
120.9, pp. 1060–1080. ISSN: 0733-9429, 1943-7900. DOI: 10 . 1061 / (ASCE ) 0733 -
9429(1994)120:9(1060). URL: http://ascelibrary.org/doi/10.1061/%28ASCE%
290733-9429%281994%29120%3A9%281060%29 (visited on 08/01/2019).
Zhou, Siping, J. A. McCorquodale, and Z. Vitasovic (1992). “Influences of Density on
Circular Clarifiers with Baffles”. en. In: Journal of Environmental Engineering 118.6,
pp. 829–847. ISSN: 0733-9372, 1943-7870. DOI: 10.1061/(ASCE)0733- 9372(1992)
118:6(829). URL: http://ascelibrary.org/doi/10.1061/%28ASCE%290733-
9372%281992%29118%3A6%28829%29 (visited on 10/09/2019).
Zhou, Siping and John A. McCorquodale (1992). “Influence of skirt radius on per-
formance of circular clarifier with density stratification”. en. In: International Jour-
nal for Numerical Methods in Fluids 14.8, pp. 919–934. ISSN: 0271-2091, 1097-0363.
DOI : 10 . 1002 / fld . 1650140804. URL: http : / / doi . wiley . com / 10 . 1002 / fld .
1650140804 (visited on 10/09/2019).
Resumen
El proceso de lodos activados convencional para el tratamiento del agua residual es el más usado para
remover los contaminantes del agua residual urbana. La biomasa (lodo activado) crece y forma flocs
biológicos que deben ser separados del agua tratada. Normalmente esta acción se realiza por medio de la
gravedad en un sedimentador. Las partículas de lodo activado pueden someterse a diferentes
comportamientos de sedimentación dependiendo de sus propiedades. La simulación de los clarificadores
secundarios es probablemente el área más desarrollada para la aplicación de la mecánica de fluidos
computacional en el tratamiento del agua residual. Sin embargo, no todos los mecanismos de sedimentación
están siempre representados en un modelo. Este trabajo presenta la adición de la compresión como un
término de segundo orden en una ecuación diferencial parcial que describe la sedimentación del lodo. La
identificación de los parámetros del modelo se realizó a través de experimentos realizados en un sistema
cerrado. Luego, simulaciones en un clarificador a escala real permitieron validar el modelo basándose en la
medición de la altura del manto de lodos y los perfiles de velocidad de las partículas. Estaciones depurados
de pequeña capacidad se caracterizan por una dinámica discontinua de las condiciones de entrada
(variaciones en el caudal, ciclos de encendido/apagado). Así el modelo validado se usó para simular estas
condiciones operacionales.

PALABRAS CLAVE: Sedimentadores secundarios, Mecánica de fluidos computacional, Compresión,

Comportamiento dinámico de sedimentadores, velocidad de sedimentación de partículas, OpenFOAM,
DAKOTA.