You are on page 1of 13

Chemical Engineering Journal 404 (2021) 126519

Contents lists available at ScienceDirect

Chemical Engineering Journal


journal homepage: www.elsevier.com/locate/cej

Mass Transfer Mechanism and Equilibrium Modelling of Hydroxytyrosol T


Adsorption on Olive Pit–Derived Activated Carbon
Severin Edera, Katrin Müllera, Paride Azzarib, Andrea Arcifac, Mohammad Peydayeshb,

Laura Nyströma,
a
ETH Zurich, Department of Health Science and Technology, Institute of Food, Nutrition and Health, Laboratory of Food Biochemistry, Schmelzbergstrasse 9, 8092 Zurich,
Switzerland
b
ETH Zurich, Department of Health Science and Technology, Institute of Food, Nutrition and Health, Laboratory of Food and Soft Materials, Schmelzbergstrasse 9, 8092
Zurich, Switzerland
c
EMPA, Swiss Federal Institute for Material Science and Technology, Switzerland. Überland Str. 129, 8600 Dübendorf, Switzerland

H I GH L IG H T S G R A P H I C A L A B S T R A C T

• Preparation of olive pit-derived acti-


vated carbon (OPAC) for hydro-
xytyrosol adsorption.
• OPAC complies to adsorption perfor-
mance of commercial activated
carbon.
• OPAC exhibits high adsorption capa-
cities over wide pH and temperature
range.
• Adsorption isotherm model selection
based on Akaike Information
Criterion.
• Investigation of adsorption kinetics by
pore volume and surface diffusion
model.

A R T I C LE I N FO A B S T R A C T

Keywords: The rise in olive oil production coincides with an increase in olive oil side-streams, such as olive mill wastewater
Hydroxytyrosol and olive pits, enabling the exploitation of higher value-added products through proper processes. Adsorption
Olive pit-derived activated carbon techniques, common in wastewater treatment, facilitate the recovery of valuable compounds found in these by-
Olive oil by-products products, such as the strong natural antioxidant hydroxytyrosol present in olive mill wastewater. Owing to their
Akaike Information Criterion
chemical composition, olive pits are a promising precursor for activated carbon production intended for ad-
Adsorption
sorption purposes. However, a complementary side-stream strategy requires an in-depth analysis of the equili-
Diffusion mechanism
brium state and the governing mass transfer mechanism based on sophisticated mathematical models for a
reliable investigation of the processes occurring during hydroxytyrosol adsorption on olive pit–derived activated
carbon (OPAC). Here we show the technological suitability of OPAC for adsorption of hydroxytyrosol in com-
parison to an existing commercial activated carbon (CAC). We found that the removal efficiency of OPAC was
superior to CAC with increasing initial concentrations of hydroxytyrosol. The Redlich-Peterson isotherm pro-
vided the best fit to the adsorption equilibrium data based on the Akaike Information Criterion (AIC).
Elucidation of the adsorption kinetics by means of the pore volume and surface diffusion model (PVSDM)
showed that intraparticle diffusion resistance dominated hydroxytyrosol adsorption on OPAC. In addition, in-
vestigations over a broad temperature and pH range revealed the versatile applicability of OPAC and indicated a
physisorption governed interaction. Our results demonstrate the fundamental parameters associated with


Corresponding author.
E-mail address: laura.nystroem@hest.ethz.ch (L. Nyström).

https://doi.org/10.1016/j.cej.2020.126519
Received 6 June 2020; Received in revised form 29 July 2020; Accepted 30 July 2020
Available online 03 August 2020
1385-8947/ © 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

hydroxytyrosol adsorption on OPAC, crucial for the development of adsorption systems to convert by-products
into opportunities.

1. Introduction implying thereby that the adsorption process follows a rate as a che-
mical reaction [35]. These models fail to provide insight into diffusional
The international council for olive oil production recorded an all- mass transfer parameters, typically associated with porous adsorbents.
time production rate record of over 3.3 million tons in 2018 [1]. Moreover, they do not enable conclusions on the underlying mechan-
However, the rising demand for olive oil as a health-promoting in- isms of the mass transfer studied. On the contrary, diffusional mass
gredient has consequently led to a concomitant increase in olive oil by- transfer models incorporate three consecutive steps: external mass
products. During the three-phase olive oil production, only 20% of the transfer, intraparticle diffusion (controlled by pore volume or surface
initial olive mass converts to the desired olive oil, leaving 50% to end diffusion or a combination of both), and adsorption on active sides. The
up as olive mill wastewater (OMWW) and 30% as solid waste (pulp and pore volume and surface diffusion model (PVSDM) is one of the most
pits) [2]. comprehensive approaches to describe the adsorption kinetics using
OMWW is a recalcitrant effluent with acidic pH (4.5 to 5.2) [3], diffusional mass transfer models [36]. In addition, the interpretation of
high organic content (chemical oxygen demand up to 220 g/L), and the adsorption equilibrium state was derived from statistically in-
elevated amounts of phenols (up to 80 g/L) [4], owing to the enhanced adequate model selection measures [37]. Yet, detailed knowledge on
water-solubility of phenolic substances [5]. The ecologically hazardous the appropriate equilibrium correlation and the governing mass transfer
OMWW impedes its biodegradability through conventional micro- parameters are indispensable for a reliable adsorption process in-
biological treatments and imposes detrimental effects on the environ- vestigation and the establishment of an effective adsorption operation
ment. For instance, the discharge of OMWW inhibits seed germination system [28].
[6], plant growth [7] and contaminates aquifer [8]. The adverse long- The present work seeks to describe a novel combination of an
term consequences are inadequately considered in current low-cost adequate mathematical equilibrium modelling and comprehensive dif-
disposal practices such as open evaporation ponds in rural areas of the fusional mass transfer model to evaluate the technological suitability of
Mediterranean region [9]. For the utilization of solid by-products i.e. an olive pit derived activated carbon (OPAC) for the recovery of HT in a
olive pits, a comprehensive concept for the utilization of olive pits is complementary approach. For this purpose, we focused on the char-
likewise unestablished, besides the use for combustion to produce heat acterization of the equilibrium, thermodynamics and major parameters
and energy [10]. affecting HT recovery by OPAC, besides an in-depth elucidation of the
Olive oil production side-streams offer the opportunity to generate governing mechanism of the adsorption process using a batch adsorp-
higher value-added products through proper exploitation processes. A tion system. Particular attention was paid to the numerical fitting of the
wide variety of phenolic compounds found in OMWW exhibit health- kinetic decay data with the PVSDM model for the determination of the
promoting effects, such as antioxidant, anti-inflammatory, and anti- diffusional parameters and the investigation of the limiting adsorption
microbial properties [11]. Among them, hydroxytyrosol (HT) is one of mechanism. Furthermore, this study conducted the interpretation of
the most abundant phenolic substance in OMWW (up to 1.4 g/mL) adsorption equilibrium based on the Akaike Information Criterion (AIC)
[12], and is one of the strongest antioxidants known in nature [13]. [38] as appropriate measure for isotherm model selection. Moreover, to
Recently, the European Food Safety Authority (EFSA) endorsed the evaluate the technological performance required for an integrated side-
preventive effects of HT on the oxidative damage of blood lipids [14] stream concept, all analysis were performed for an available commer-
and the European commission authorized its implementation as food cial active carbon (CAC) in parallel.
additive to fish and vegetable oil and margarines [15]. Owing to these
beneficial properties, HT aroused considerable interest in cosmetic
2. Materials & methods
[16], food [17], and medical industry [18].
Over recent years, numerous studies proposed methods for the re-
2.1. Chemicals
covery of HT from OMWW, including solvent extraction [19], selective
nano- and ultrafiltration [20], integrated membrane filtration systems
Methanol (> 99.9%), formic acid (98–100%), sodium hydroxide
[21], adsorption onto resin and coated activated carbon [22–24], solid
(≥98%), and graphite flakes were purchased from Sigma-Aldrich (St.
phase extraction [25], solar distillation [26], and coagulation techni-
Louis, United States). Hydrochloric acid (> 37%) was obtained from
ques [27]. However, most of these methods require significant resource
VWR International (Radnor, United States). Hydroxytyrosol (≥98%)
and operating input, thus hindering a cost-effective exploitation. Ad-
was purchased from Extrasynthese (Genay, France). All solutions were
sorption is considered a promising technique for HT recovery from
prepared with purified water using a Millipore MilliQ-system (Billerica,
OMWW, owing to its low investment cost, simplicity in design, high
United States).
removal efficiency, and ease of regeneration [28]. Activated carbon
represents the most important adsorbent applied in various water
treatment and purification processes [29]. The olive pits, mainly com- 2.2. Preparation of olive pit–derived activated carbon (OPAC)
posed of lignocellulosic biomass, provide excellent characteristics as
versatile raw material for the preparation of adsorbent material [30]. Olive pits were provided washed and ground by a Spanish olive mill
Several works showed the successful application of activated carbon (Alcubilla 2000, Castro del Río, Spain). The olive pits were carbonized
derived from olive pits for the removal of metals [31,32], minerals in a MTF tubular furnace (Carbolite, Neuhausen, Germany) equipped
[33], and organic substances [34]. Nevertheless, a complementary ap- with a temperature programmer (Eurotherm, Worthing, United
proach using activated carbon produced from olive pits for the ex- Kingdom). Pyrolysis was carried out at 873 K for 1 h under a constant
ploitation of valuable phenols from OMWW, such as HT, is still lacking. flowing nitrogen atmosphere. Subsequently, the physical activation of
Furthermore, although adsorption is a well-known phenomenon, the char obtained was performed at 1173 K for 1 h using water vapor.
many models and assumptions, made in the previous studies, are not The activated carbon prepared was washed according to the procedure
valid and far from the reality of the process. Those investigations based of Kula et al. [32] to remove residual organic and mineral impurities. In
the involved adsorption kinetics on adsorption reaction models, brief, the activated carbon was washed with 0.5 M HCl, hot water, and
finally with cold water until the effluent showed neutral pH.

2
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

Afterwards, the sample was dried in a hot air oven at 105 °C overnight. temperature with an adsorbent dose of 0.01 g, solution volume of
The dried activated carbon was ground and sieved to obtain a particle 10 mL, and initial adsorbate concentration ranging from 50 to 750 mg/
size < 250 µm. The adsorption performance of OPAC was evaluated in L. The solution was kept in contact for 180 min under vigorous agita-
comparison to Pulsorb C, a commercial activated carbon (CAC), re- tion in a tumbler mixer (Bioengineering, Wald, Switzerland). The effect
commended for adsorption of phenolic substances. Pulsorb C was ob- of pH on the adsorption capacity was studied over a pH range from 2 to
tained from Chemviron (Feluy, Belgium). OPAC and CAC were stored in 8. The pH of the solution was adjusted with HCl or NaOH. The effect of
a desiccator under N2–atmosphere until use to prevent oxidation. temperature on HT uptake was studied over a temperature range from 4
to 60 °C. The kinetic study was conducted using six initial concentra-
2.3. Characterization of OPAC and CAC tions of HT for each adsorbent. The kinetic studies were conducted with
adjusted experimental parameter, namely an adsorbent dose of 50 mg,
Micrographs of the activated carbons were obtained using a Hitachi and a solution volume of 50 mL, to avoid volume distortion due to
SU5000 high-resolution field emission scanning electron microscope aliquots taken at preset time points.
(SEM) equipped with a secondary electron (SE) detector (Tokyo, The mixtures were centrifuged at 4000 rpm for 15 min after in-
Japan). Samples were sputter coated with 4.5 nm platin-paladium prior cubation. Residual HT concentration in the supernatant was measured
to SEM investigation. using reversed-phase ultra-performance liquid chromatography with
The surface chemical characterization of the carbonaceous samples diode array detection (RP-UHPLC-DAD) (see section 2.5). Using the mass
by X-ray photoelectron spectroscopy (XPS) was performed with a balance equation, the adsorbed amount of HT per mass unit of ad-
Quantum 2000 X-ray photoelectron spectrometer (Physical Electronics, sorbent was derived. The mass balance of any given point of the ad-
Minnesota, United States), equipped with an Al Kα monochromatic sorption process is given by
source (1486.6 eV), a hemispherical capacitor electron-energy ana- (c0 − ct ) V
qt = , (3)
lyzer, and a 16–channel plate detector. All the spectra were acquired W

with an emission angle of 45°, in fixed analyzer transmission mode,


where c0 and ct are the initial HT concentration in solution (mg/L) and
using a nominal X-ray beam-spot size of 150 μm and at a pressure lower
at any given time point (mg/L), respectively. qt is the adsorbate con-
than 4 × 10–7 Pa. Survey and high-resolution spectra were acquired at
centration on the adsorbent at given time point (mg/g). V is the volume
a pass energy of 117.4 eV (energy step of 1 eV) and 29.35 eV (energy
of the solution (L), and W the weight of the adsorbent (g) used in the
step of 0.125 eV), respectively. The samples were prepared by me-
batch adsorption experiment. All experiments were performed in tri-
chanically embedding the powder into thin indium foil. The atomic
plicates unless otherwise described.
fraction of the detectable elements was estimated from the survey
spectra, according to the following equation [39]:
2.5. Determination of hydroxytyrosol concentration
Ia / RSFa
xa = ∑in Ii / RSFi
,
(1) The HT concentration was analyzed by RP-UPLC-DAD after batch
where xi is the atomic concentration of the element “i”, and Ii is the adsorption experiments. An Agilent 1290 Infinity II System (Agilent
intensity (area) of the selected peak of the element “i”. RSFi is the as- Technologies, California, United States) equipped with an Acquity
sociated relative sensitivity factor. The RSFs used in this work were UPLC BEH C18 (2.1 × 100 mm) column and an Acquity UPLC BEH C18
derived from the analysis software of the XP-spectrometer. Eq. (1) is VanGuard (2.1 × 5 mm) (Waters, Massachusetts, United States) guard
strictly valid for materials that are homogeneous to a depth greater than column operating at 35 °C was used for analysis. Injection volume of the
the information depth of the technique. As this assumption might not sample was 10 μl and eluted using a combination of the two mobile
hold for the case of activated carbon, the concentrations obtained by phases: (A) water/formic acid (88/12 v/v) adjusted to pH 1.7 and (B)
Eq. (1) are referred to as “apparent” concentrations in the following methanol/water/formic acid (90/8/2 v/v) at a flow rate of 0.4 mL/min.
[40]. The applied gradient was adapted from the method described by Öztürk
The specific surface area (SBET) and pore volume of the activated et al. [41] with slight modifications. The resulting gradient program
carbon were derived from N2 adsorption/desorption experiments ac- was 0–5 min, isocratic 100% A; 5–10 min, linear to 80% A and 20% B;
cording to the Brunauer, Emmett, and Teller (BET) method using a 11–18 min, linear to 50% A and 50% B; 18–22 min, linear to 100% B;
TriStar II Plus Analyzer (Micromeritics, Georgia, United States). The 22–24 min, linear to 100% A. Detection was performed at a wavelength
average pore sizes were obtained from the desorption branch of the N2 of 280 nm. Quantification was based on external calibration using a HT
experiments using the Barrett, Joyner, and Halenda (BJH) method. standard at eight concentration levels (0.5 to 50 mg/L). Data was
Particle size distributions of activated carbons were measured with a processed using the Chromeleon Chromatography Data System (CDS)
Partica LA–950 Laser Diffraction Particle Size Distribution Analyzer Version 7 (Thermo Fisher Scientific, Massachusetts, United States).
(Horiba, Kyoto, Japan). He-Pycnometry (AccuPyc 1330 Pycnometer,
Micromeritics) was utilized to determine the solid density (ρs) of the 2.6. Adsorption kinetics
adsorbents. The apparent density (ρp) of the adsorbents was analyzed
using a Pascal 140/440 Hg-Porosimeter (Thermo Fisher Scientific, 2.6.1. Pore volume and surface diffusion model (PVSDM)
Massachusetts, United States). The equation used to calculate the par- Detailed adsorption kinetic studies were carried out to reveal the
ticle porosity (εp) for both activated carbons is given by underlying mechanisms and characteristics of the adsorption systems
ρp investigated. The PVSDM developed by Leyva-Ramos and Geankoplis
εp = 1 − . [42] was applied to describe the adsorption kinetics. This model
ρs (2)
characterizes the progress of the adsorption process by three con-
secutive steps, namely external mass transfer, intraparticle diffusion,
2.4. Batch adsorption experiments and active site adsorption. The following assumptions were considered
for the model: the batch adsorption system is operated at constant
The capacity of OPAC and CAC to adsorb HT was evaluated by the temperature; the adsorbent particles are spherical; the intraparticle
effect of initial adsorbent concentration, contact time, pH, and tem- diffusion can occur by pore diffusion and/or surface diffusion; the pore
perature. The study was conducted as a batch adsorption process while volume diffusion coefficient (DP) and the surface diffusion coefficient
varying the parameter under investigation and maintaining all others (DS) are constant; the mass transfer by convection inside the pores is
constant. Experiments were carried out at neutral pH and room negligible, the adsorption rate on an active site is instantaneous. The

3
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

PVSDM model and the initial and boundary conditions are defined as to the method described by Furusawa and Smith [44]. Evaluation of Eq.
dc
(4) at initial conditions (t = 0, c = c0, and cr = 0) can be expressed as
V dt
= −AS kF (c − cr |r = R ) (4)
d (c / c0 ) AS kF
=− .
t = 0, c = c0 (5)
dt t=0 V (15)

Integration of Eq. (15) allowed the determination of kF based on the


εp
∂cr
∂t
+ ρp ∂t =
∂q 1 ∂
r 2 ∂r
⎡r 2 DP
⎣ ( ∂cr
∂r
+ ρp DS ∂r ⎤

∂q
) (6) slope of the initial linear part of the curve [45], and was given by
c AS
t = 0, 0 ≤ r ≤ R, cr = 0 (7) ln c = −kF t
0 V (16)

∂cr Hence, a plot of ln c/co versus time, at t = 0, was utilized to derive


=0
∂r r = 0 (8) kF considering the first two data points of each kinetic curve. The ex-
ternal surface area of the adsorbent particles was calculated according
∂cr ∂q
DP + ρp DS ∂r = kF (c − cr |r = R ) to
∂r r = R r=R (9)
6M
In these equations, c is the adsorbate concentration in aqueous so- AS = dp ρp (1 − εp )
,
(17)
lution (mg/L), cr is the adsorbate concentration within the particle at
distance r (mg/L), AS is the total external surface for mass transfer (1/ where M is the adsorbent mass concentration in solution (g/mL) and
cm), εp is the adsorbent particle porosity, ρp is the apparent density of dp is the adsorbent particle diameter (cm) [44]. DP was obtained ac-
the adsorbent (g/cm3), kf is the external mass transfer coefficient cording to the model based on molecular diffusivity (DH) and tortuosity
through the boundary layer (cm/s), DP is the pore volume diffusion (τ) [46] and is given by
coefficient (cm2/s), and DS is the surface diffusion coefficient (cm2/s). DH εP
The assumption that the adsorption on an active site is instantaneous DP = τ
, (18)
implies the existence of a local equilibrium between the pore solution
where
concentration, cr, and the solid-phase concentration of HT, q. This
equilibrium relationship is described by the adsorption isotherm (2 − εp )2
τ= .
εp (19)
q = f (cr ). (10)
The molecular diffusivity of HT in water was derived from the Wilke
The model equations were solved numerically using MatLab®, fol-
and Chang [47] correlation, which is defined as
lowing the work of Červeňanský et al. [43]. For this purpose, the partial
differential equation Eq. (6) was transformed into a set of ordinary (φMSolv )0.5T
DH = 7.4x10−8 ⎡ ⎤,
differential equations substituting the first and second spatial deriva- ⎣
0.6
ηSolv V H
⎦ (20)
tives with a second-order central difference approximation. The spatial
domain 0 ≤ r ≤ R was discretized into N = 200 segments of length where φ is the association factor for the solvent. For water φ = 2.6.
R
h = N , creating N + 1 nodes, indexed with an integer i = 1, ⋯, N + 1, MSolv is the molecular weight of water (18.02 g/mol). ηSolv is the dy-
where i = 1 corresponds to r = 0 and i = N + 1 to r = R . The temporal namic viscosity of water (0.89 cP) and T is temperature (298.15 K). VH
derivative of the adsorbed amount was obtained using the chain rule is the molar volume at normal boiling point of HT (162.6 cm3/mol)
estimated by the Le Bas additive method. The molecular diffusivity of
∂q ∂c
= q' ∂t , (11) HT in water obtained with Eq. (20) was 7.99 × 10–6 cm2/s.
∂t
The initial estimates of the coefficients kF and DP, for the solution of
' dq
where q = can be easily calculated. The time evolution of the
dc
, the PVSDM model, were calculated according to Eq. (16) and (18),
concentration at each node can be calculated from respectively. At first, we optimized the value of kF to fit the asymptote,
obtained as the average of the last three experimental points. Based on
(∊p + ρp qi' ) dti = Dp
dc
( ci + 1 − ci − 1
hri
+
ci + 1 − 2ci + ci − 1
h2 ) these preliminary values, the surface diffusion coefficients (DS) and the
qi + 1 − qi − 1 qi + 1 − 2qi + qi − 1 final kF and DP were determined using the non-linear least squares
+ ρp Ds ( hri
+
h2 ), (12) optimization. The non–linear least squares objective function is defined
as the sum of the squared differences between the predicted data (ccal)
where ci is the value of the concentration at the i -node and
and experimental data (cexp) of HT concentration and was calculated
qi = q (ci ) , while ri is the value of the radius at that node. Discretizing
according to
the value of the derivatives of q allows overriding the problem of di-
verging second derivatives of the Redlich-Peterson when the con- N
SSE = ∑i = 1 (ccal − cexp )i2 . (21)
centration is zero. The boundary condition Eq. (8) was rewritten using a
forward second-order difference This objective function was minimized using MatLab®.
4c 2 − c3
c1 = 3
. (13) 2.6.3. Relative contribution of pore and surface diffusion to intraparticle
A fictitious node i = N + 2 was created to fulfill the last boundary diffusion
condition Eq. (9), the value of the concentration at this node is The relative contributions of pore volume diffusion (NP) and surface
diffusion (NS) to the overall intraparticle diffusion of HT were evaluated
2h
cN + 2 = cN + 1 + '
(c − cN + 1). based on the respective mass flux of each mechanism [36]. The relative
(Dp + ρp Ds qN + 1) kF (14) contribution of each intraparticle diffusion mechanism is given by
Together with Eq. (4) describing the concentration of the adsorption DP ∂cr
system, we obtain a system of N + 1 ordinary diffential equations with NP ∂r
NP + NS
= DS ρp ∂q
DP ∂cr
algebraic constraints. The equations were solved numerically using the ∂r
+
∂r (22)
solver ode15s of MatLab®.
DS ρp ∂q
NS ∂r
2.6.2. Estimation of parameters NP + NS
= DS ρp ∂q
.
DP ∂cr
+
The external mass transfer coefficient (kF) was estimated according ∂r ∂r (23)

4
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

2.6.4. Rate-limiting mass transfer resistance plausibility that fitted model i is the best approximating model within
The dimensionless Biot number (Bi) reflects the ratio of external and the set of models decreases the larger Δi. For the probability assessment
internal mass transfer resistances of the adsorption process. The Biot of a model being the best among the models considered, an estimate of
number in case of a combined pore volume and surface diffusion model the relative likelihood, given our data L (model/data), can be obtained.
1
is defined as The transformation L (model/data) ∝ exp(- 2 Δi) provides a relative
1
kF rp c0 likelihood, where exp°(- 2 Δi) is the relative likelihood of model i.
Bi = ,
DP c0 + DS ρp qe (24) Normalization of the relative likelihoods results in Akaike weights ,
given by
where rp is the adsorbent particle radius (cm) and qeq is the equili-
brium concentration on the adsorbent (g/g) with the initial con- e(− 12 Δi)
wi = ,
centration c0 (g/mL) [45]. The rate of film diffusion in comparison to −1Δ
∑iR= 1 e( 2 i ) (29)
intraparticle diffusion becomes faster as the Bi value increases. For
th
Bi < 0.5, the film diffusion resistance controls the adsorption rate where wi is the Akaike weight of the i model [50]. The weight wi
while the influence of intraparticle diffusion on the overall adsorption reflects the “weight of evidence” in favor of a model to describe given
process is negligible. A reasonably dominance of intraparticle resistance data with the most statistical confidence. The ratio of Akaike weights,
exists at Bi > 30 [48]. termed evidence ratio (ER), represents an additional measure to judge
the evidence and is defined as
2.7. Adsorption isotherm models wmin
ERi = wi
, (30)
The description of the adsorption equilibrium is a crucial step in the where wmin is the Akaike weight of the best model [50]. This ratio
assessment of a particular adsorption process. The equilibrium data was provides evidence of how much more likely the best model is than
analyzed using the Langmuir, Freundlich, Redlich-Peterson, Sips, and model i.
Tóth isotherm models. A plot of ceq versus qeq yielded the experimental
isotherm curves. The experimental results were fitted by nonlinear re-
2.9. Adsorption thermodynamics
gression to the isotherm models described to elucidate the adsorption
capacity and the respective isotherm parameters. The parameters of the
The thermodynamic parameters standard free Gibbs energy (ΔG°),
isotherm models were derived by minimizing the sum of squared errors
standard enthalpy (ΔH°), and standard entropy (ΔS°) were evaluated to
analogous to Eq. (21) between experimental and predicted data. For
describe the effect of temperature on the adsorption process. The re-
more information and details on the isotherm models, see
lationship of ΔG° to ΔH° and ΔS° can be defines as [51]
Supplementary information.
ΔG ° = ΔH ° − T ΔS ° (31)
2.8. Akaike information Criterion The values of these parameters were calculated from experimental
data using following equations [52]:
The AIC developed by Akaike [38] provides a statistical criterion to
compare the adequacy of multiple models to predict the experimental ΔG ° = −RTln (KD ) (32)
data. AIC represents a comparison between the achievable goodness-of-
ΔS ° ΔH °
fit and the simplicity of the model to avoid over-fitting among the ln (KD ) = R
− RT (33)
models considered. The general formula to calculate the AIC value is
qeq
given by KD = ,
ceq (34)
AIC = −2ln (L) + 2K , (25)
where KD is the distribution coefficient, T is the temperature (K) and R
where L is the maximized likelihood function for the estimated model represents the gas constant (8.3145 J/mol K). The parameters ΔH° and
and K is the number of parameters in the model. Applying the least ΔS° were obtained from the slope and the intercept of the Van‘t Hoff
squares estimation and assuming normally and independently dis- plot of lnKD versus 1/T, respectively.
tributed model errors, AIC can be described as
3. Results and discussion
AIC = Nln ( ) + 2K ,
SSE
N (26)
3.1. Characterization of OPAC and CAC
where N is the number of observations and SSE is the sum of squared
errors, which is calculated analogous to Eq. (21) for the difference N2 sorption experiments and BET analysis revealed a considerably
between the predicted (qeq,cal) and experimental adsorbent loading greater internal surface area of OPAC (1040 m2/g) than CAC (740 m2/
(qeq,exp) at equilibrium. In case of small sample sizes (N/K < 40), g) (Table 1, Fig. S1). A comprehensive overview of the physical prop-
Sugiura [49] proposed a bias corrected version of AIC containing an erties of both adsorbents is given in Table 1.
additional penalty term, called the corrected Akaike information cri- SEM investigations revealed significant differences in surface
terion (AICc). The AICc is defined as
Table 1
AICc = Nln ( ) + 2K +
SSE
N
2K (K + 1)
N−K−1
.
(27) Physical characteristics of OPAC and CAC.

The AIC and AICc scores are ordinal and not interpretable on their Adsorbent OPAC CAC
own. A meaningful strength-of-evidence comparison requires rescaling 2
Specific surface area (m /g) 1040 740
the AICc values to Pore volume (cm3/g) 0.69 0.50
Average pore size (Å) 44 52
Δi = AICci − AICcmin , (28) Particle size (μm) 124 42
Apparent density (g/cm3) 0.64 0.54
where Δi is the information loss resulting if model i is chosen instead of
Solid density (g/cm3) 2.2 2.3
the best approximating model, imin, for inferences. AICci is the AIC value Particle porosity (-) 0.71 0.76
for model i. AICcmin is the minimum AICc value of all models tested. The

5
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

Fig. 1. SEM micrographs of (A) OPAC and (B) CAC at a series of magnifications. (C) XPS survey scan of OPAC ( ) and CAC ( ). (D) Apparent oxygen elemental
concentration on surface of both adsorbent materials. (E) C1s and (F) O1s chemical state spectra of OPAC and CAC in comparison to graphite ( ) and tentative
functional group assignments Different letters denote significant differences between apparent oxygen concentrations (p < 0.05, n = 3).

6
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

morphology and pore distribution of both activated carbons OPAC and similar shape. This observation suggests a high degree of similarity in
CAC (Fig. 1A, B). The surface of OPAC showed a great variety of distinct terms of surface chemistry of both materials. However, the complex line
cavities resembling a sponge-like structure. Furthermore, OPAC ex- shape of the C1s peak of graphite-like materials significantly compli-
hibited a heterogeneous surface with well-developed pores of randomly cates the curve fitting of the signal, such that obtaining a detailed de-
distributed size. In contrast, the micrographs of CAC revealed a smooth scription of the type and relative amount of C-O surface functionalities
surface with sharp edges. The surface of CAC possessed a uniform is challenging [53]. The contribution of the secondary structure of the
porosity with sporadic macropores. OPAC had a considerably more C1s signal is particularly relevant if the amount of oxygen at the surface
developed macroporosity on the surface in comparison to CAC. The is small, which is the case for the present study (Fig. S3). The spectra of
particular surface morphology of OPAC is consistent with observations the activated carbon is comparable with that of commercial graphite,
made in previous studies on olive pit activated carbon [31,33]. whose spectra are presented in Fig. S4. As expected, also the C1s
XPS was used to analyze the surface chemistry of the activated spectrum of this reference material exhibited an asymmetric line shape,
carbon under investigation. The survey spectra of the two materials and satellites peaks, which are ascribed to the contribution of defective
(Fig. 1C) showed the expected signals of carbon (C1s and C KKL). sites and energy losses resulting from shake-up excitations [53,54].
Oxygen signals (O1s and O KKL) were also clearly detectable, indicating For both OPAC and CAC, the O1s signals showed a maximum at
the presence of oxygen-containing functional groups at the surface of around 533.0 eV (Fig. 1F). This suggests that alcohol and ether groups,
both activated carbons. The apparent concentration of oxygen was rather than carbonyl moieties, dominate the surface chemistry of both
found to be 4.36 ± 0.04% and 6.9 ± 0.6%, for OPAC and CAC, re- materials (Table S1) [55]. On the other hand, given the rather low
spectively (Fig. 1D). These values are comparable with those reported apparent concentration of oxygen in these samples, the presence of
by Ruiz et al. [40]. For both samples, a careful inspection of the survey minor amount of oxides has an impact on the shape of the O1s signals.
spectra (Fig. S2) revealed the presence of additional elements on the This is particularly relevant for CAC, as the amount of aluminium and
surface of the active carbons. silicon was found to be higher in the commercial sample (Table S2). The
The C1s spectra of OPAC and CAC (Fig. 1E) exhibited remarkably contribution of aluminium and silicon oxides to the O1s signal is

Fig. 2. Experimental data and PVSDM prediction for HT decay curve during adsorption on (A) OPAC ( ) and (B) CAC ( ) at six initial concentrations. Relative
contribution of surface and pore volume diffusion to the total intraparticle diffusion at different radial position during HT adsorption on (C) OPAC (500 mg/L) and
(D) CAC (450 mg/L).

7
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

expected to be significant between 531.0 and 533.0 eV [56].This is in OPAC showed kF values about an order of magnitude higher than those
line with the observation that the O1s signal of CAC was broader than observed for CAC. The smaller particles of CAC had a larger total ex-
that of OPAC (the full-with-at-half maximum are 3.0 and 2.6 eV, re- ternal surface with a higher number of available binding sites. This
spectively), and slightly shifted towards lower binding energies factor eventually decreased the overall rate of external mass transfer
(Fig. 1F). The higher apparent concentration of aluminium and silicon until surface coverage in comparison to OPAC. The correlation between
oxides in CAC than in OPAC might also explain the higher apparent decreasing kF values with increasing initial HT concentration observed
concentration of oxygen in the commercial product (Fig. 1D). On the is in accordance with adsorption equilibrium theory. The slope of the
other hand, the different morphology and possible surface chemical operating lines of the kinetic equilibrium isotherms, the lines joining c0
inhomogeneity of the two materials may also play a role in affecting the to ceq, determine the influence of the initial HT concentration on the
values of apparent concentrations. In summary, the XPS analysis re- external mass transfer rate. Higher c0 values result in a lower slope of
vealed that both the activated carbons exhibited similar surface the operating line and consequently in decreasing kF values [57]. This
chemistry, the major difference being related to the different amount observation might be related to higher concentration gradients at
and type of contaminants, that are present in rather small amounts higher c0 values, whereas the number of available surface active sites is
(< 1%). limited for a considered adsorption system [58]. Therefore, higher in-
itial concentrations ultimately result in higher external mass transfer
3.2. Adsorption kinetics resistances, as reflected in decreased kF values. The effect of initial
concentration on kF described is in agreement with previous work using
The investigation of adsorption kinetics is essential to evaluate the Lantana camara forest waste for phenol removal [58].
characteristic mass transfer parameters and reveal the governing me- The surface diffusion coefficients (DS) calculated, ranging from 0.8
chanisms in the adsorption process. The pore volume and surface dif- to 8.1 × 10–9cm2/s and 0.4 to 15 × 10–9cm2/s for OPAC and CAC,
fusion model (PVSDM) applied in this study assumes that surface and respectively, showed increased values at higher initial concentration for
pore diffusion mechanisms contribute to the intraparticle diffusion. The both activated carbons studied (Table 2). The initial increase of DS for
optimal values of the parameters kF, DP, and DS were determined by OPAC up to a HT concentration of 300 mg/L was followed by a gradual
fitting the numerical solution of the PVSDM model to the experimental decrease towards lower values. The fact that surface diffusion is con-
kinetic data. trolled by the most loosely bound molecules provides an explanation for
The optimal kF, DP, and DS values obtained for each data set were the concentration dependency observed of DS [59]. Lower initial con-
used for the PVSDM prediction of the HT decay curves for six initial centrations result in a low internal surface coverage with adsorption
concentrations with each activated carbon (Fig. 2A, B). The time until occurring on high-energy adsorption sites with little molecular mobility
attainment of equilibrium was 65 min and 15 min for HT adsorption on after adsorption. With higher initial concentration and consequently
OPAC and CAC, respectively (Fig. 2A, B). The consistency of the PVSDM increasing surface coverage, the adsorbed HT molecules will occupy
model prediction with the experimental decay curve for OPAC, de- more and more low-energy actives sites. Those active sites exhibit a
monstrated that the PVSDM fitted the experimental data satisfactorily. lower binding force that causes molecules to desorb faster and diffuse to
The simulation of the kinetic process of HT adsorption on CAC showed another site [59]. In this context, the pattern observed for OPAC at
an increased deviation from the experimental data with higher initial higher initial concentration might indicate a higher overall availability
concentration. The mean particle diameter of CAC incorporated in the of high-energy binding sites. Ocampo-Perez et al. [60] reported a si-
PVSDM provided a rough reflection of the actual size proportions due to milar dependency of DS values and initial adsorbate concentration for
a bimodal particle size distribution (Fig. S2). Owing to these circum- phenol adsorption on organobentonite. Difference in DS values ob-
stances, the optimal numerical solution still showed inconsistencies served between OPAC and CAC are attributable to the respective par-
with the experimental data. Table 2 summarizes the mass transfer ticle size of the adsorbent. A decrease in particle sizes was suggested to
parameters, derived from the PVSDM model, and experimental para- results in an increased DS coefficient owing to shorter intraparticle
meters describing the adsorption kinetic process of HT on OPAC and diffusion pathways [45].
CAC. The pore volume diffusion coefficients (DP) derived showed com-
The film diffusion coefficient kF showed decreasing values with in- parable values for HT adsorption on both activated carbons (Table 2).
creasing initial concentration and ranged from 13 to 2.7 × 10–3 cm/s As anticipated, the DP values were smaller than the diffusion coefficient
for OPAC and 2.9 to 0.3 × 10–3 cm/s for CAC. The HT adsorption on in free liquid (DH), since additional restrictions hinder the adsorbate

Table 2
Experimental conditions and mass transfer parameters derived from the PVSDM for the HT adsorption on OPAC and CAC.
OPAC c0 ceq qeq kF × 10–3 DS × 10–9 DP × 10–7 NS
[%]
NP
[%] Bi
[mg/l] [mg/L] [mg/g] [cm/s] [cm2/s] [cm2/s] NP + NS NP + NS

50 0.1 49.9 13 0.8 4.8 99.3 0.7 164


100 1.3 98.7 13 3.2 4.8 99.4 0.6 166
200 17 183 9.2 8.1 4.8 94.4 5.4 118
300 34 266 5.8 4.6 4.8 80 20 75
500 157 343 3.5 4.5 4.8 54 46 45
700 325 375 2.7 3.2 4.8 30 70 35

CAC c0 ceq qeq kF × 10–3 DS × 10–9 DP × 10–6 NS


[%]
NP
[%] Bi
[mg/l] [mg/L] [mg/g] [cm/s] [cm2/s] [cm2/s] NP + NS NP + NS

150 0.4 149.6 2.9 0.4 3.0 72 28 2


250 30 220 1.5 0.7 3.0 4.5 95.6 1
300 72 228 1.2 7 3.0 16 84 0.8
450 183 267 0.6 10 3.0 11 89 0.4
600 319 281 0.5 12 3.0 8 92 0.3
750 442 308 0.3 15 3.0 7 93 0.2

8
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

transport within the pore. The values obtained displayed strong corre- Overall, the observation of intraparticle diffusion as rate-controlling
lation with the expected range of DP values according to literature (10–4 mechanism together with the DS values discussed above suggest that a
to 10–6 cm2/s) [45]. decrease in particle size might ultimately reduce the equilibration time
The relative contributions of pore volume diffusion (NP) and surface for a considered adsorption system [45]. Sze and McKay [62] reported
diffusion (NS) to the overall intraparticle diffusion process of HT were an accelerated attainment of equilibrium with decreasing adsorbent
determined with Eqs. (22) and (23). The relative contribution of both size when studying phenol adsorption on activated carbon with three
mechanisms varied with time and radial position within the particle distinct particle sizes, which agrees with the equilibration time differ-
until equilibrium was reached (Fig. 2C, D). Furthermore, the relative ences of OPAC and CAC observed. The contribution of film, surface, and
contribution was dependent on the initial concentration of HT pore volume diffusion to the adsorption of HT on OPAC and CAC, re-
(Table 2). With rising initial concentration, the contribution of surface spectively, supported the adequacy of the PVSDM model for the de-
diffusion diminished for both adsorption systems investigated and in- tailed investigation of adsorption kinetics.
traparticle diffusion was dominated progressively by pore diffusion.
The relative contributions at 250 mg/L initial HT concentration showed 3.3. Adsorption isotherm
an inversion of the prevailing intraparticle diffusion mechanism within
the trend of CAC. This observation might be attributed to the impedi- The adsorption equilibrium isotherms showed the adsorbed amount
ments encountered during the numerical modelling as discussed above of HT on OPAC and CAC in relation to the equilibrium concentration of
and limit its significance. The differences in the relative contributions to HT in solution (Fig. 3A, B). The shape of the adsorption isotherm de-
intraparticle diffusion between the activated carbons were consistent scribes the affinity of the adsorbate towards the adsorbent and provides
with the observations and trends of the DS values for the respective insight into the possible mechanism of interaction. Both isotherms ob-
adsorbent (Table 2). The involvement of both internal diffusion re- served can be classified as L curves according to the classification for
sistances during HT adsorption on OPAC and CAC illustrated the re- liquid-solid adsorption systems [63]. This isotherm type is character-
quired application of the PVSDM. ized by a decrease in vacant sites on which the adsorbate can be ad-
The Biot number (Bi) represents a ratio of external and internal mass sorbed with increasing concentration of HT adsorbed on the solid. The
transfer rates to deduce the rate-controlling diffusion mechanism Eq. concave curves reflect the progressive saturation of the activated car-
(24). The intraparticle diffusion was the rate-limiting mechanism for bons studied. The L curve isotherms indicate that the intermolecular
HT adsorption on OPAC independent of the initial HT concentration attractions between HT and both activated carbons are driven by van
(Table 2). In case of CAC, the adsorption occurring with the three der Waals interactions [51]. In this work, we applied the Langmuir,
highest initial HT concentration showed a film diffusion controlled Freundlich, Sips, Tóth, and Redlich-Peterson isotherm models to the
process. Generally, the influence of intraparticle diffusion resistance equilibrium data in order to elucidate the characteristics of the ad-
decreased with increasing initial concentration for both adsorption sorption process. The statistical comparison of the models considered
systems under consideration. The Bi values in the range of 35 to 166 for selecting the best approximation to predict the experimental data
suggested that intraparticle diffusion resistance exclusively dominated was based on the Akaike Information Criterion (AIC) methodology. The
HT adsorption on OPAC. The Bi values obtained for CAC, ranging from model possessing the highest wi and lowest Δi value was estimated to be
0.2 to 2, indicated a complete dominance of film diffusion or a si- the best fitting model. The Redlich-Peterson isotherm model provided
multaneous resistance of film and intraparticle diffusion to the overall the best fit to HT adsorption on OPAC and CAC (Table 3). Using the
adsorption process [48]. The stronger contribution of external mass AICc values and associated statistics, the isotherm models ranked in the
transfer to overall diffusion restriction in the adsorption on CAC as following order to predict the empirical data: Redlich-Peterson >
compared to OPAC is in agreement with the previously discussed in- Tóth > Sips > Freundlich > Langmuir. This relative ranking was
fluences on kF. The observation towards higher contribution of external valid for HT adsorption on both activated carbons studied. Generally,
mass transfer to the overall diffusion resistance with increasing initial the three-parameter isotherm models provided a better fit than the two-
concentration is in line with previous work. Franco et al. [61] and parameter models. The relative adequacies of the models in the set can
Girish and Murty [58] reported a similar trend in the ratio of external to be evaluated based on rough guidelines. A Δi ≤ 2 suggests substantial
internal mass transfer resistances studying the adsorption of Indium on evidence for the model. Models having 4 ≤ Δi ≤ 7 provide con-
chitosan and chitin and phenol on Lantana camara, respectively. siderably less support, whereas a Δi > 10 indicates basically no

Fig. 3. Experimental data (■) and Redlich-Peterson isotherm model (- - -) of adsorption isotherm for HT adsorption on (A) OPAC ( ) and (B) CAC ( ). Different
letters denote significant differences between isotherm points (p < 0.05, n = 3).

9
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

Table 3 incorporates Langmuir and Freundlich model features to represent ad-


AIC results of five isotherm models for adsorption equilibrium modelling and sorption equilibrium over a wide concentration range. The model
Redlich-Peterson parameters derived for HT adsorption of OPAC and CAC. parameters derived permit inferences of the underlying adsorption
Freundlich Langmuir Sips Tóth Redlich-Peterson mechanism during adsorption onto OPAC and CAC. The values of the
heterogeneity factor g were close to 1 for both adsorption systems,
KR [L/g] aR [L/mg] g suggesting that the isotherms tended towards the Langmuir form
(Table 3). Hence, the g values indicated that adsorption approached
OPAC 255 1.56 0.89
monolayer adsorption behavior on homogenous active sites [65]. The
K 2 2 3 3 3 large KR values demonstrate the high adsorption capacity of OPAC and
AIC 178 181 154 153 151 CAC for hydroxtyrosol adsorption [66]. The rapid uptake rate of HT at
AICc 179 182 155 154 152
low concentrations might be reflected in the particular high KR con-
Δi 26 29 2.8 1.4 0.0
wi 0.01 × 10-4 0.03 × 10-5 0.14 0.29 0.57 stants obtained for CAC. Generally, the KR values are not useful as a
ER 489,690 2,241,026 4 2 1 measure to derive maximum achievable adsorption loading, as the
KR [L/g] aR [L/mg] g
Redlich-Peterson isotherm shows no saturation at higher concentra-
tions. This limits the comparability with previously studied adsorption
CAC 2627 12.9 0.95 systems that were described by different isotherm models. For a
meaningful interpretation, the KR value observed for OPAC was com-
K 2 2 3 3 3
pared with adsorption isotherms of phenolic adsorbats predicted with
AIC 131 146 124 124 123
AICc 131 146 125 125 124 the Redlich-Peterson isotherm. The KR value of OPAC was higher than
Δi 7.5 22 1.5 1.2 0.0 the other values reported for the removal of phenolic substances using
wi 0.01 0.08 × 10-4 0.23 0.26 0.49 various adsorbents (Table 4). Furthermore, OPAC and CAC exhibited
ER 42 65,020 2 2 1
removal efficiencies above 95% up to an initial HT concentration of
200 mg/L (Fig. 3A, B). At higher initial HT concentrations, such as
600 mg/L, OPAC showed a remarkable removal efficiency above 50%
evidence for the model [64]. According to this classification, the
and exceeded the performance of CAC.
Langmuir (Δi = 29) and Freundlich (Δi = 26) models showed no
The high internal surface area of OPAC might be the driving factor
support for being a suitable approximation with respect to the best
for the overall adsorption strength observed. The step of film diffusion
model for HT adsorption onto OPAC. However, there was substantial
occurred orders of magnitude faster than the intraparticle diffusion, as
strength of evidence for the Tóth (Δi = 1.4) and Sips (Δi = 2.8) iso-
described in detail above. Thus, the high external surface area of CAC
therm models. For adsorption on CAC, the Langmuir model (Δi = 22)
may contribute to the strong affinity for adsorption observed, which
likewise had no evidence as being a plausible model fit. However, the
was reflected in the KR value. The adsorption behavior of OPAC mat-
Freundlich model (Δi = 7.5) represented moderate evidence for being a
ched the performance characteristics of a commercial activated carbon,
suitable prediction of the data obtained compared to the Redlich-Pe-
showing its promising suitability for HT adsorption. Smaller particle
terson model. Both other three-parameter models, Sips (Δi = 1.5) and
size with a narrower size distribution might further enhance OPAC
Tóth (Δi = 1.2), showed considerable support as adequate prediction of
performance for future research and applications using it as powdered
the equilibrium data. The wi values obtained can be interpreted as
activated carbon. However, smaller adsorbent particles in the adsorp-
probability that respective model is actually the best model for the data
tion system possibly imply altered adsorption parameters [45]. In this
given. The Redlich-Peterson model showed wi values of 0.57 and 0.42,
context, it is pivotal to adjust the particle size to the intended adsorber
interpretable as a 57% and 42% chance that this model was the best
design and concomitant constraints to achieve an optimal compromise
approximating model for the adsorption process onto OPAC and CAC,
for the desired application.
respectively. The relatively high wi values excluded a potential model
selection uncertainty and showed clear support for the Redlich-Peterson
model regarding the data sets of both activated carbons. The evidence 3.4. Effect of solution temperature on adsorption capacity
ratio is an additional measure derived from Δi to assess the relative
merits of each candidate model. Considering the evidence ratios cal- The adsorption equilibrium in a considered system is extensively
culated for OPAC, it could be assumed that the best-fitting Redlich- affected by solution temperature. The effect of temperature on the ad-
Peterson model was 2 times more likely than the Tóth model, 4 times sorption capacity was examined at 300 mg/L initial HT concentration
more likely than the Sips, 489,690 times more likely than the Freun- and neutral pH. Adsorption capacities of HT on OPAC and CAC in-
dlich model, and 2,241,026 times more likely than the Langmuir model. creased with increasing temperature from 277 to 333 K, with an in-
Similarly, for adsorption onto CAC, it was observed that the Redlich- crease from 235 ± 5 to 261 ± 9 mg/L and 216.5 ± 2.3 to
Peterson model was 2, 2, 42, and 65,020 times more likely than the 262 ± 6 mg/L, respectively, indicating an endothermic nature of the
Tóth, Sips, Freundlich, and Langmuir models, respectively. Based on the adsorption process (Fig. 4A, B). The adsorption capacities obtained at
evidence ratios obtained, the conventional two-parameter isotherm 277 and 298 K differed significantly from 333 K for OPAC. Uptake
models (Langmuir and Freundlich) presented poor model fits (ER > amounts observed at 313 and 323 K deviated neither from 298 K nor
150) for given data [64]. from 333 K significantly. An increase in temperature from 277 to 313 K
The Redlich-Peterson isotherm is a hybrid isotherm model that showed a significant higher adsorption capacity of HT on CAC. Values
observed above 313 K showed no deviation.

Table 4
Comparison of Redlich-Peterson isotherm parameter for adsorption of phenolic substances in various adsorption systems.
Adsorbent Adsorbate Isotherm Model KR [L/g] Reference

Graphene oxide nanosheets Hydroxytyrosol Redlich-Peterson 7 [67]


Activated carbon fibers Phenol Redlich-Peterson 71 [68]
Commercial active carbon Phenol Redlich-Peterson 54 [69]
Granular active carbon Chlorinated phenol Redlich-Peterson 23–26 [66]
OPAC Hydroxytyrosol Redlich-Peterson 255 This study

10
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

Fig. 4. Effect of temperature on HT adsorption capacity of (A) OPAC ( ) and (B) CAC ( ). Van‘t Hoff plot and regression line (- - -) for HT adsorption on (C) OPAC
and (D) CAC. Different letters denote significant differences between isotherm points (p < 0.05, n = 3).

The thermodynamic parameters used to assess the adsorption pro- adsorption loadings for OPAC and CAC, with an increase from 204 ± 6
cess, namely the standard free Gibbs energy (ΔG°), the standard en- to 247 ± 5 mg/L and 180 ± 11 to 250 ± 14 mg/L, respectively.
thalpy (ΔH°), and the standard entropy (ΔS°) (Table 5), are derived OPAC was particularly effective at low pH compared to CAC. The ad-
from the Van‘t Hoff plot (Fig. 4C, D). Negative values of ΔG° for OPAC sorption capacities achieved with OPAC at pH 2 deviated significantly
and CAC showed the spontaneity and energetic favorability of the ad- from the remaining pH range, however no significant difference was
sorption process. The observation to more negative values of ΔG°, for observed between pH 4 to 8. Similarly, CAC showed a significant lower
both activated carbons, indicated the enhanced feasibility of adsorption adsorption loading at pH 2. Furthermore, the adsorption capacity dif-
at higher temperature. Positive ΔH° confirmed the endothermic nature fered significantly at pH 4 and 8, but did not deviate between pH 5 and
of the adsorption onto OPAC and CAC. Values of ΔH° < 20 KJ indicated 8. The pH range at highest adsorption capacity corresponded to the pH
a physisorption governed interaction between adsorbent and adsorbate, of OMWW, ranging from 4.5 to 5.2, for both activated carbons [3]. The
caused by van der Waals forces. This observation corroborates the as- increase of adsorption capacities with higher pH was in accordance
sumption made above based on the L curve isotherms. The positive ΔS° with previous work on phenol adsorption on olive pit-based activated
reflected the increased disorder and randomness at the solid/solution carbon, considering the pH range under investigation [70]. The ad-
interface during adsorption. Ultimately, HT adsorption on OPAC and sorption behavior observed might be attributed to a beneficial change
CAC was an entropy controlled process, as TΔS° contributed more than
ΔH° to the negative ΔG° values obtained [51].
Table 5
Thermodynamic parameters of the adsorption process of HT on OPAC and CAC.
3.5. Effect of solution pH on adsorption capacity ΔH° [kJ/ ΔS° [J/ ΔG° [kJ/mol]
mol] (mol*K]
The pH of the solution is considered a governing factor in the ad-
sorption behavior. The effect of pH on the adsorption capacity was 277 K 298 K 313 K 323 K 333 K

examined at 300 mg/L initial HT concentration and 25 °C. The results


OPAC 7.3 37 −2.9 −3.6 −4.4 −4.6 −5.3
showed an effect of initial pH on the adsorption capacity of both acti- CAC 14 59 −2.2 −3.4 −4.7 −4.9 −5.4
vated carbons (Fig. 5). An increase from pH 2 to 8 resulted in higher

11
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

Fig. 5. Effect of temperature on HT adsorption capacity of (A) OPAC ( ) and (B) CAC ( ).

in surface functionality regarding the affinity towards the adsorbate at Energy Science and Engineering, ETH Zürich and Claudio Reinhard
higher pH. The surface functionalities proposed by means of XPS ana- from the Laboratory of Food Biochemistry, ETH Zürich for their assis-
lysis partially dissociate at higher pH. As a result, the stronger disper- tance in the preparation of olive pit-derived activated carbon. The au-
sion forces present eventually lead to enhanced interactions and result thors acknowledge Dr. Anne Greet Bittermann of Scope M, ETH Zürich,
in higher HT adsorption capacities. Moreover, HT is a weak acid for supporting the SEM analysis and sharing her expertise. The authors
(pka = 9.45) and is present in its undissociated form over the studied are thankful to Dr. Michael Plötze from the Claylab, ETH Zürich and
pH range. Higher pH leads to irreversibly conversion into the corre- Natalia Smatsi from Particle Technology Laboratory, ETH Zürich for
sponding chinon form with changed chemical properties. Therefore, their help in the physical characterization of the adsorbent materials.
higher pH values were excluded in this study. This study was financed by ETH Zürich.

4. Conclusion Appendix A. Supplementary data

Olive pit–derived activated carbon (OPAC) exhibited promising Supplementary data to this article can be found online at https://
merits as technological suitable and sustainable adsorbent material for doi.org/10.1016/j.cej.2020.126519.
the recovery of hydroxytyrosol (HT) found in olive mill wastewater.
The application of a pore volume and surface diffusion model and an References
AIC based isotherm model selection demonstrated conclusively that the
adsorption performance of OPAC was equivalent to CAC. The high re- [1] I.O.O. Council, World olive oil figures, 2019, https://www.internationaloliveoil.
org/wp-content/uploads/2020/04/HO-W901-29-11-2019-P.pdf, (02.06.2020).
moval efficiency towards higher initial HT concentrations, exceeding
[2] M. Niaounakis, C.P. Halvadakis, Introduction, in: M. Halvadakis, C.P. Niaounakis
CAC, corroborated the remarkable properties of OPAC. Intraparticle (Eds.), Waste Management Series, Elsevier, 2006, pp. 3–22.
diffusion as governing mass transfer mechanism, reflected by the Biot [3] I.E. Kapellakis, K.P. Tsagarakis, J.C. Crowther, Olive oil history, production and by-
ratio represented the decisive role of the high internal surface area in product management, Rev. Environ. Sci. Bio/Technol. 7 (2008) 1–26.
[4] N. Azbar, A. Bayram, A. Filibeli, A. Muezzinoglu, F. Sengul, A. Ozer, A review of
HT adsorption on OPAC. The PVSDM revealed a shift of the relative waste management options in olive oil production, Crit. Rev. Environ. Sci. Technol.
contribution of surface to pore diffusion with increasing initial HT 34 (2004) 209–247.
concentration. The undiminished adsorption capacity within the pH [5] P.S. Rodis, V.T. Karathanos, A. Mantzavinou, Partitioning of olive oil antioxidants
between oil and water phases, J. Agric. Food. Chem. 50 (2002) 596–601.
range of OMWW offers optimal conditions for application of OPAC in [6] C. Amaral, M.S. Lucas, A. Sampaio, J.A. Peres, A.A. Dias, F. Peixoto, M.D.R. Anjos,
the treatment of olive oil production effluents. The binding mechanism C. Pais, Biodegradation of olive mill wastewaters by a wild isolate of Candida
between OPAC and HT proposed, based on weak intermolecular bonds, oleophila, Int. Biodeterior. Biodegrad. 68 (2012) 45–50.
[7] I. Aviani, M. Raviv, Y. Hadar, I. Saadi, Y. Laor, Original and Residual Phytotoxicity
will facilitate future work on the regeneration of the adsorbent and the of Olive Mill Wastewater Revealed by Fractionations before and after Incubation
efficient desorption of HT. This work provides the scientific frame for with Pleurotus ostreatus, J. Agric. Food. Chem. 57 (2009) 11254–11260.
the development of a suitable adsorber design based on OPAC aiming at [8] V. Kavvadias, M.K. Doula, K. Komnitsas, N. Liakopoulou, Disposal of olive oil mill
wastes in evaporation ponds: Effects on soil properties, J. Hazard. Mater. 182
a complementary exploitation concept of olive oil by-products. (2010) 144–155.
Additionally, the rigorous analytical data evaluation of the adsorption [9] R. Jarboui, F. Sellami, C. Azri, N. Gharsallah, E. Ammar, Olive mill wastewater
process studied provides a methodical design for future adsorption evaporation management using PCA method: Case study of natural degradation in
stabilization ponds (Sfax, Tunisia), J. Hazard. Mater. 176 (2010) 992–1005.
studies.
[10] G. Rodríguez, A. Lama, R. Rodríguez, A. Jiménez, R. Guillén, J. Fernández-Bolaños,
Olive stone an attractive source of bioactive and valuable compounds, Bioresour.
Declaration of Competing Interest Technol. 99 (2008) 5261–5269.
[11] K.L. Tuck, P.J. Hayball, Major phenolic compounds in olive oil: metabolism and
health effects, J. Nutr. Biochem. 13 (2002) 636–644.
The authors declare that they have no known competing financial [12] N. Allouche, I. Fki, S. Sayadi, Toward a High Yield Recovery of Antioxidants and
Purified Hydroxytyrosol from Olive Mill Wastewaters, J. Agric. Food. Chem. 52
interests or personal relationships that could have appeared to influ-
(2004) 267–273.
ence the work reported in this paper. [13] Y. Achmon, A. Fishman, The antioxidant hydroxytyrosol: biotechnological pro-
duction challenges and opportunities, Appl. Microbiol. Biotechnol. 99 (2015)
1119–1130.
Acknowledgments [14] N. Efsa Panel on Dietetic Products, Allergies, Scientific Opinion on the substantia-
tion of health claims related to polyphenols in olive and protection of LDL particles
The authors gratefully thank Dr. Felix Donat from the Laboratory of from oxidative damage (ID 1333, 1638, 1639, 1696, 2865), maintenance of normal

12
S. Eder, et al. Chemical Engineering Journal 404 (2021) 126519

blood HDL cholesterol concentrations (ID 1639), maintenance of normal blood in supercapacitors, Electrochim. Acta 52 (2007) 4969–4973.
pressure (ID 3781), “anti-inflammatory properties” (ID 1882), “contributes to the [41] N. Öztürk, M. Tunçel, N.B. Tunçel, Determination of phenolic acids by a modified
upper respiratory tract health” (ID 3468), “can help to maintain a normal function HPLC: its application to various plant materials, J. Liquid Chromatogr. Related
of gastrointestinal tract” (3779), and “contributes to body defences against external Technol. 30 (2007) 587–596.
agents” (ID 3467) pursuant to Article 13(1) of Regulation (EC) No 1924/2006, [42] R. Leyva-Ramos, C.J. Geankoplis, Model simulation and analysis of surface diffusion
EFSA J. 9 (2011) 2033. of liquids in porous solids, Chem. Eng. Sci. 40 (1985) 799–807.
[15] N. Efsa Panel on Dietetic Products, Allergies, D. Turck, J.-L. Bresson, B. Burlingame, [43] I. Červeňanský, M. Mihaľ, J. Markoš, Modeling of 2-phenylethanol adsorption onto
T. Dean, S. Fairweather-Tait, M. Heinonen, K.I. Hirsch-Ernst, I. Mangelsdorf, polymeric resin from aqueous solution: Intraparticle diffusion evaluation and dy-
H.J. McArdle, A. Naska, M. Neuhäuser-Berthold, G. Nowicka, K. Pentieva, Y. Sanz, namic fixed bed adsorption, Chem. Eng. Res. Des. 147 (2019) 292–304.
A. Siani, A. Sjödin, M. Stern, D. Tomé, M. Vinceti, P. Willatts, K.H. Engel, [44] T. Furusawa, J.M. Smith, Fluid-Particle and Intraparticle Mass Transport Rates in
R. Marchelli, A. Pöting, M. Poulsen, J. Schlatter, E. Turla, H. van Loveren, Safety of Slurries, Ind. Eng. Chem. Fundam. 12 (1973) 197–203.
hydroxytyrosol as a novel food pursuant to Regulation (EC) No 258/97, EFSA J. 15 [45] E. Worch, Adsorption kinetics, Adsorption Technology in Water Treatment,
(2017) e04728. Fundamentals, Processes, and Modeling, De Gruyter, Berlin, Boston, 2012, pp.
[16] S. D'Angelo, D. Ingrosso, V. Migliardi, A. Sorrentino, G. Donnarumma, A. Baroni, 123–168.
L. Masella, M. Antonietta Tufano, M. Zappia, P. Galletti, Hydroxytyrosol, a natural [46] C. Valderrama, X. Gamisans, X. de las Heras, A. Farrán, J.L. Cortina, Sorption ki-
antioxidant from olive oil, prevents protein damage induced by long-wave ultra- netics of polycyclic aromatic hydrocarbons removal using granular activated
violet radiation in melanoma cells, Free Radical Biol. Med. 38 (2005) 908–919. carbon: Intraparticle diffusion coefficients, J. Hazard. Mater. 157 (2008) 386–396.
[17] L. Martinez, G. Ros, G. Nieto, Hydroxytyrosol: Health Benefits and Use as [47] C.R. Wilke, P. Chang, Correlation of diffusion coefficients in dilute solutions, AIChE
Functional Ingredient in Meat, Medicines (Basel) 5 (2018). J. 1 (1955) 264–270.
[18] S. Bulotta, M. Celano, S.M. Lepore, T. Montalcini, A. Pujia, D. Russo, Beneficial [48] D.O. Cooney, Comparison of simple adsorber breakthrough curve method with
effects of the olive oil phenolic components oleuropein and hydroxytyrosol: focus exact solution, AIChE J. 39 (1993) 355–358.
on protection against cardiovascular and metabolic diseases, J. Transl. Med. 12 [49] N. Sugiura, Further analysts of the data by akaike' s information criterion and the
(2014) 219. finite corrections, Commun. Stat. – Theor. Meth. 7 (1978) 13–26.
[19] N. Kalogerakis, M. Politi, S. Foteinis, E. Chatzisymeon, D. Mantzavinos, Recovery of [50] K.P. Burnham, D.R. Anderson, Information and Likelihood Theory: A Basis for
antioxidants from olive mill wastewaters: A viable solution that promotes their Model Selection and Inference, in: K.P. Burnham, D.R. Anderson (Eds.), Model
overall sustainable management, J. Environ. Manage. 128 (2013) 749–758. Selection and Multimodel Inference: A Practical Information-Theoretic Approach,
[20] C.M. Galanakis, E. Tornberg, V. Gekas, Clarification of high-added value products Springer, New York, New York, NY, 2002, pp. 49–97.
from olive mill wastewater, J. Food Eng. 99 (2010) 190–197. [51] J.S. Piccin, T.R.S.A. Cadaval, L.A.A. de Pinto, G.L. Dotto, Adsorption Isotherms in
[21] D.P. Zagklis, A.I. Vavouraki, M.E. Kornaros, C.A. Paraskeva, Purification of olive Liquid Phase: Experimental, Modeling, and Interpretations, in: A. Bonilla-
mill wastewater phenols through membrane filtration and resin adsorption/deso- Petriciolet, D.I. Mendoza-Castillo, H.E. Reynel-Ávila (Eds.), Adsorption Processes
rption, J. Hazard. Mater. 285 (2015) 69–76. for Water Treatment and Purification, Springer International Publishing, Cham,
[22] D. Frascari, G. Rubertelli, F. Arous, A. Ragini, L. Bresciani, A. Arzu, D. Pinelli, 2017, pp. 19–51.
Valorisation of olive mill wastewater by phenolic compounds adsorption: [52] P. Rani Agrawal, N. Singh, S. Kumari, S.R. Dhakate, The removal of pentavalent
Development and application of a procedure for adsorbent selection, Chem. Eng. J. arsenic by graphite intercalation compound functionalized carbon foam from con-
360 (2019) 124–138. taminated water, J. Hazard. Mater. 377 (2019) 274–283.
[23] A. Yangui, M. Abderrabba, Towards a high yield recovery of polyphenols from olive [53] D.-Q. Yang, E. Sacher, Carbon 1s X-ray Photoemission Line Shape Analysis of Highly
mill wastewater on activated carbon coated with milk proteins: Experimental de- Oriented Pyrolytic Graphite: The Influence of Structural Damage on Peak
sign and antioxidant activity, Food Chem. 262 (2018) 102–109. Asymmetry, Langmuir 22 (2006) 860–862.
[24] G. Fava, M.D. Di Mauro, M. Spampinato, D. Biondi, G. Gambera, G. Centonze, [54] M. Smith, L. Scudiero, J. Espinal, J.-S. McEwen, M. Garcia-Perez, Improving the
R. Maggiore, N. D'Antona, Hydroxytyrosol Recovery From Olive Mill Wastewater: deconvolution and interpretation of XPS spectra from chars by ab initio calcula-
Process Optimization and Development of a Pilot Plant, CLEAN – Soil, Air, Water 45 tions, Carbon 110 (2016) 155–171.
(2017) 1600042. [55] G. Beamson, D. Briggs, High Resolution XPS of Organic Polymers: The Scienta
[25] L. Bertin, F. Ferri, A. Scoma, L. Marchetti, F. Fava, Recovery of high added value ESCA300 Database, Wiley, 1992.
natural polyphenols from actual olive mill wastewater through solid phase ex- [56] D. Briggs, Handbook of X-ray Photoelectron Spectroscopy C. D. Wanger, W. M.
traction, Chem. Eng. J. 171 (2011) 1287–1293. Riggs, L. E. Davis, J. F. Moulder and G. E.Muilenberg Perkin-Elmer Corp., Physical
[26] S. Sklavos, G. Gatidou, A.S. Stasinakis, D. Haralambopoulos, Use of solar distillation Electronics Division, Eden Prairie, Minnesota, USA, 1979. 190 pp. $195, Surface
for olive mill wastewater drying and recovery of polyphenolic compounds, J. and Interface Analysis 3 (1981) v-v.
Environ. Manage. 162 (2015) 46–52. [57] G.M. Walker, L. Hansen, J.A. Hanna, S.J. Allen, Kinetics of a reactive dye adsorption
[27] P. García-García, A. López-López, J.M. Moreno-Baquero, A. Garrido-Fernández, onto dolomitic sorbents, Water Res. 37 (2003) 2081–2089.
Treatment of wastewaters from the green table olive packaging industry using [58] C.R. Girish, V.R. Murty, Mass Transfer Studies on Adsorption of Phenol from
electro-coagulation, Chem. Eng. J. 170 (2011) 59–66. Wastewater Using Lantana camara, Forest Waste, Int. J. Chem. Eng. 2016
[28] K.Y. Foo, B.H. Hameed, Insights into the modeling of adsorption isotherm systems, (2016) 11.
Chem. Eng. J. 156 (2010) 2–10. [59] D.C.K. Ko, J.F. Porter, G. McKay, Effect of Concentration-Dependent Surface
[29] E. Worch, Introduction, Adsorption Technology in Water Treatment, Fundamentals, Diffusivity on Simulation of Fixed Bed Sorption Systems, Chem. Eng. Res. Des. 81
Processes, and Modeling, De Gruyter, Berlin, Boston, 2012, pp. 1–10. (2003) 1323–1332.
[30] J. Saleem, U.B. Shahid, M. Hijab, H. Mackey, G. McKay, Production and applica- [60] R. Ocampo-Perez, R. Leyva-Ramos, J. Mendoza-Barron, R.M. Guerrero-Coronado,
tions of activated carbons as adsorbents from olive stones, Biomass Convers. Adsorption rate of phenol from aqueous solution onto organobentonite: Surface
Biorefin. 9 (2019) 775–802. diffusion and kinetic models, J. Colloid Interface Sci. 364 (2011) 195–204.
[31] T.M. Alslaibi, I. Abustan, M.A. Ahmad, A. Abu Foul, Preparation of activated carbon [61] D.S.P. Franco, J. Vieillard, N.P.G. Salau, G.L. Dotto, Interpretations on the me-
from olive stone waste: optimization study on the removal of Cu2+, Cd2+, Ni2+, chanism of In(III) adsorption onto chitosan and chitin: A mass transfer model ap-
Pb2+, Fe2+, and Zn2+ from aqueous solution using response surface metho- proach, J. Mol. Liq. 304 (2020) 112758.
dology, J. Dispersion Sci. Technol. 35 (2014) 913–925. [62] M.F. Sze, G. McKay, An adsorption diffusion model for removal of para-chlor-
[32] I. Kula, M. Uğurlu, H. Karaoğlu, A. Çelik, Adsorption of Cd(II) ions from aqueous ophenol by activated carbon derived from bituminous coal, Environ. Pollut. 158
solutions using activated carbon prepared from olive stone by ZnCl2 activation, (2010) 1669–1674.
Bioresour. Technol. 99 (2008) 492–501. [63] C.H. Giles, T.H. MacEwan, S.N. Nakhwa, D. Smith, 786. Studies in adsorption. Part
[33] M.L. Martínez, M.M. Torres, C.A. Guzmán, D.M. Maestri, Preparation and char- XI. A system of classification of solution adsorption isotherms, and its use in di-
acteristics of activated carbon from olive stones and walnut shells, Ind. Crops Prod. agnosis of adsorption mechanisms and in measurement of specific surface areas of
23 (2006) 23–28. solids, J. Chem. Soc. (Resumed) (1960) 3973–3993.
[34] R. Ubago-Pérez, F. Carrasco-Marín, D. Fairén-Jiménez, C. Moreno-Castilla, Granular [64] K.P. Burnham, D.R. Anderson, Multimodel inference: understanding AIC and BIC in
and monolithic activated carbons from KOH-activation of olive stones, Microporous model selection, Sociol. Meth. Res. 33 (2004) 261–304.
Mesoporous Mater. 92 (2006) 64–70. [65] M.A. Al-Ghouti, D.A. Da'ana, Guidelines for the use and interpretation of adsorption
[35] H. Qiu, L. Lv, B.-C. Pan, Q.-J. Zhang, W.-M. Zhang, Q.-X. Zhang, Critical review in isotherm models: A review, J. Hazard. Mater. 122383 (2020).
adsorption kinetic models, J. Zhejiang Univ. Sci. A 10 (2009) 716–724. [66] Z. Aksu, J. Yener, A comparative adsorption/biosorption study of mono-chlorinated
[36] R. Ocampo-Perez, R. Leyva-Ramos, P. Alonso-Davila, J. Rivera-Utrilla, M. Sanchez- phenols onto various sorbents, Waste Manage. 21 (2001) 695–702.
Polo, Modeling adsorption rate of pyridine onto granular activated carbon, Chem. [67] S. Şahin, Z. Ciğeroğlu, O.K. Özdemir, M. Bilgin, E. Elhussein, Ö. Gülmez, Recovery
Eng. J. 165 (2010) 133–141. of hydroxytyrosol onto graphene oxide nanosheets: Equilibrium and kinetic models,
[37] A.-N. Spiess, N. Neumeyer, An evaluation of R2 as an inadequate measure for J. Mol. Liq. 285 (2019) 213–222.
nonlinear models in pharmacological and biochemical research: a Monte Carlo [68] Q.-S. Liu, T. Zheng, P. Wang, J.-P. Jiang, N. Li, Adsorption isotherm, kinetic and
approach, BMC Pharmacol. 10 (2010) 6. mechanism studies of some substituted phenols on activated carbon fibers, Chem.
[38] H. Akaike, A new look at the statistical model identification, IEEE Trans. Autom. Eng. J. 157 (2010) 348–356.
Control 19 (1974) 716–723. [69] M. Ahmaruzzaman, D.K. Sharma, Adsorption of phenols from wastewater, J. Colloid
[39] M.P. Seah, I.S. Gilmore, S.J. Spencer, Quantitative XPS: I analysis of X-ray photo- Interface Sci. 287 (2005) 14–24.
electron intensities from elemental data in a digital photoelectron database, J. [70] M. Termoul, B. Bestani, N. Benderdouche, M. Belhakem, E. Naffrechoux, Removal
Electr. Spectrosc. Related Phenomena 120 (2001) 93–111. of phenol and 4-chlorophenol from aqueous solutions by olive stone-based activated
[40] V. Ruiz, C. Blanco, E. Raymundo-Piñero, V. Khomenko, F. Béguin, R. Santamaría, carbon, Adsorpt. Sci. Technol. 24 (2006) 375–388.
Effects of thermal treatment of activated carbon on the electrochemical behaviour

13

You might also like