Modified Fly Ash as an Adsorbent for the Removal of Pharmaceutical Residues from Water

Abstract

In this work, different methods for fly ash modification were applied to obtain an adsorbent for the efficient removal of selected pharmaceuticals from a multiclass aqueous solution. Morphological and surface properties of the modified fly ash samples were analyzed by scanning electron microscopy, X-ray fluorescence, X-ray diffraction, Fourier transform infrared spectroscopy, and point of zero charge, and the influence of the applied modifications was determined by comparison with the results obtained for unmodified fly ash. Experimental parameters of the adsorption of the pharmaceutical onto the modified fly ash were optimized, and special attention was paid to the influence of different parameters on the adsorption capacities. Multivariate methods of analysis, such as artificial neural networks, applied to the obtained results showed that the contact time, the initial concentration of the pharmaceutical solution, and the pH value had the strongest influence on the adsorption process. Fly ash modified with chitosan and magnetic iron oxide showed the best adsorption properties (removal efficiency above 80% for the majority of the selected pharmaceuticals), and artificial neural networks confirmed its susceptibility to the modeling process.

1. Introduction

Despite the tendency to use renewable energy sources for energy production, coal combustion still represents one of the most common ways to generate electric energy. In Serbia, the share of coal in electricity production is 65%. During coal combustion in thermal power plants, large amounts of fly and bottom ash are generated as byproducts. This waste ends up in landfills, and its disposal and transport represent a growing environmental problem due to destructive effects on ecosystems and living organisms. Fly ash is a pozzolanic material, mainly consisting of oxides (SiO₂, Al₂O₃, Fe₂O₃, CaO, MgO, K₂O, Na₂O, and TiO₂) as major components, as well as different amounts of trace elements, such as As, Se, Cd, and Cr [1], that make it potentially toxic. Therefore, it is of major concern to reduce this type of waste by converting it into a higher-level product. Diverse applications for fly ash have been explored, including its reuse for concrete production in the construction industry [2], as well as in road construction and masonry [3]. Additionally, a lot of research has been conducted on the transformation of fly ash as a source of aluminosilicate into zeolites [4,5] and the modification of fly ash to obtain efficient adsorbents for the removal of organic and inorganic pollutants from water [6,7,8]. Different modification methods (alkaline and hydrothermal treatments, chemical and alkaline activation, polymerization, and coating techniques, etc.) were used to obtain adsorbents with negatively charged surfaces with increased ability for the adsorption of metal ions from aqueous solutions [1]. These modification methods also proved to be successful in obtaining adsorbents efficient in the removal of organic pollutants, such as dyes [7,9], pesticides [10,11,12], and pharmaceuticals [13,14]. According to the literature, most research is focused on the removal of a single pollutant from water using fly-ash-based adsorbents. Therefore, this work aims to examine the adsorption properties of differently modified fly ash for the removal of multiclass pharmaceuticals from water.

Pharmaceuticals, classified as emerging pollutants, persist in aquatic environments due to the limitations of traditional water purification systems. These substances are extensively used in human and veterinary medicine, leading to their widespread presence in aquatic ecosystems [15]. Waste from drug manufacturing, improper disposal of expired medications, and incomplete metabolism further exacerbate water pollution. Detectable levels of pharmaceuticals have been found in both surface and groundwater, underscoring their designation as environmental contaminants [16]. Moreover, these compounds can pose significant risks to human health even at low concentrations. Addressing this issue requires effective removal techniques, as conventional methods have proven inadequate. In this work, different methods for fly ash modification were applied to obtain an adsorbent for the efficient removal of selected pharmaceuticals from a multiclass aqueous solution. Findings from the literature show that alkali-activated fly ash shows good adsorption properties in terms of pharmaceutical removal [17,18], and therefore, in this work, alkali-activated fly ash was obtained with NaOH using reflux. Chitosan was also chosen as a modifier due to its high affinity for pharmaceuticals [19,20], which is a consequence of the presence of functional groups that may act as active sites for pharmaceutical adsorption. The obtained chitosan-modified fly ash was additionally modified with magnetic iron oxide to enable the easier separation of the adsorbent and improve the adsorption properties [20,21]. Experimental parameters of pharmaceutical adsorption of the modified fly ash were optimized, and special attention was paid to the influence of different parameters (ratio of adsorbent mass and volume, pH value, initial adsorbent concentration, and contact time) on the adsorption capacities. Multivariate analysis was applied to the obtained results to establish a correlation between the examined adsorption parameters, reduce the number of variables, and reveal which of the parameters had the strongest influence on the adsorption efficiency. Also, artificial neural networks were used to predict the adsorption characteristics and reduce the number of experiments, obtaining a good basis for the commercial application of fly ash in the field of wastewater treatment.

2. Materials and Methods

2.1. Material Preparation

The fly ash (FA) utilized in this investigation was sourced from the Nikola Tesla B coal-fired thermal power plant situated in Belgrade, Serbia. FA was used as a starting material for obtaining adsorbents for the removal of selected pharmaceuticals from water. Following sieving, the material underwent thorough washing with distilled water and was subsequently dried at a temperature of 105 °C in an oven. The first modification was performed as follows: 1.0 g FA and 1.0 g chitosan (Ch) were mixed with 50.0 cm³ of 5% acetic acid. The solution was lightly mixed for 24 h at ambient temperature to ensure the full dissolution of the chitosan and the effective incorporation of FA particles into the chitosan molecular structure. Following this, 3.90 g of FeCl₃∙6H₂O and 2.70 g of FeCl₂∙4H₂O were dissolved in 10.0 cm³ of distilled water and introduced into the chitosan and fly ash suspension and mixed for one hour. The resulting solution was injected into 1000 cm³ of sodium hydroxide solution (2 mol/dm³) using a syringe and a needle with constant mixing for 6 h, after which, sample FA/Ch_mag with magnetic properties was formed [22]. The obtained magnetic beads were washed with distilled water until the sodium hydroxide was removed and dried at 105 °C in an oven.

The second modification was performed by mixing 9.0 g FA and 100 cm³ sodium hydroxide (2 mol/dm³). The obtained suspension was transferred into a Teflon-lined stainless-steel autoclave (150 cm³) and hydrothermally treated at a temperature of 96 °C and self-generated pressure for 8 h. The resulting suspension was centrifuged (5000 rpm for 5 min), and the obtained precipitate was washed with distilled water until a neutral pH value was obtained, dried at 105 °C overnight, and crushed in an avan with a pestle. The obtained material (5.0 g) was added to chitosan solution (1.0 g of chitosan dissolved in 100 cm³ acetic puffer (pH = 4.5)) and constantly shaken (150 rpm) for 24 h at room temperature. Thereafter, the obtained suspension was centrifuged (5000 rpm for 5 min) and washed with distilled water until the acetate was removed and finally dried at 105 °C overnight. In this way, sample FA_ac/Ch was obtained. The third modification of FA was carried out by mixing 2.0 g of FA with 20 cm³ NaOH solution (2 mol/dm³). The resulting suspension was transferred to a flask and heated in a water bath (up to 100 °C) under reflux for 5 h. After cooling, the obtained material was washed with distilled water and dried at 105 °C overnight. This sample was marked as FA_ref.

2.2. Material Characterization

Scanning electron microscopy—SEM (SEM JEOL JSM-6610LV, JEOL Ltd., Hertfordshire, UK)—was used to determine the morphology and surface structure of obtained materials.

X-ray fluorescence (XRF) was used for the chemical analysis of the unmodified and modified fly ash surfaces. For this purpose, a hand-held XRF spectrometer, Thermo Scientific Niton™ XL3t GOLDD (Thermo Fisher Scientific Inc., Waltham, MA, USA), was used.

X-ray diffraction (XRD) patterns for all the materials under investigation were obtained using an Ultima IV Rigaku (Rigaku Holdings Corporation, Tokyo, Japan) diffractometer, which employed CuKα1,2 radiation. The XRD data were collected over a 2θ range of 5–60°, utilizing a continuous scanning mode. The scanning was conducted with a step size of 0.02° and at a rate of 5° per minute.

Fourier transform infrared spectroscopy (FTIR, Bomem MB-Series, Hartmann Braun, Bomem Inc./Hartmann & Braun, Quebec, QC, Canada) was used to determine the type of surface functional groups on the examined materials. KBr pellets with the examined materials were prepared, and FT-IR spectra were recorded from 400 to 4000 cm⁻¹.

The point of zero charge (pH_PZC) of the examined materials was determined by immerging 0.06 g of the material into cuvettes containing 20 cm³ of 0.01 M NaCl solutions with the initial pH (pH_i) of the solution adjusted to 2, 4, 6, 8, 10, and 12. The initial pH of the solution was adjusted by adding 0.1 M HCl or 0.1 M NaOH. Before closing, nitrogen was introduced into the cuvettes, which were then shaken for 48 h at room temperature. The final pH (pH_f) of the solutions was measured, and the dependencies pH_f − pH_i were plotted to determine pH_PZC [23].

2.3. Adsorption Experiments

The obtained materials were used as adsorbents for the removal of selected pharmaceuticals belonging to different classes of pharmaceuticals: lorazepam, bromazepam, and diazepam (sedatives); erythromycin and doxycycline (antibiotics); and atorvastatin, clopidogrel, simvastatin, and cilazapril (cardiovascular). A stock solution of a mixture of the pharmaceuticals (25 mg/dm³ per component) was prepared using methanol due to the low water solubility of erythromycin, atorvastatin, and simvastatin (Table S1 in Supplementary Materials), and this solution was used for the preparation of all pharmaceutical water solutions. The adsorption of the selected pharmaceuticals from the multiclass aqueous solution was carried out in a batch system in glass beakers with constant shaking (150 rpm) at room temperature. The influence of the applied modification method on the adsorption efficiency was studied through the adsorption of the selected pharmaceuticals from the aqueous solution (25 cm³, initial concentration 500 µg/dm³ per compound) onto 0.05 g of fly ash (FA), hydrothermally treated fly ash (FA_ac), FA/Ch_mag, FA_ac/Ch, and FA_ref. In order to achieve the maximum adsorption capacity, optimization of the adsorption process was conducted, involving assessments of the adsorbent mass, solution volume, initial pH, pharmaceutical solution concentration, and contact time to determine their respective influences. All adsorption experiments were performed at an initial concentration of 500 μg/dm³ for each pharmaceutical. The effect of the initial pH of the adsorbate solution on the adsorption was investigated by adjusting the starting pH to 2, 4, 6, and 8 with an accuracy of ±0.01 using a diluted HCl and NaOH solutions. Additionally, an adsorption experiment was performed from the pharmaceutical solution without pH adjustment (measured initial pH was 5.5). Based on the obtained results of the adsorption efficiency, the optimal pH value was selected and used in further experiments. The influence of the ratio of the adsorbent mass to the adsorbate volume on the pharmaceutical adsorption was examined using different ratios of the adsorbent mass to the adsorbate volume: 1, 2, 3, 4, 5, and 6 g/dm³. For assessing the influence of the contact time, adsorption experiments were conducted using 0.1 g of adsorbent and 50 cm³ of pharmaceutical solution. Samples were taken at certain time intervals of 5, 15, 30, 60, 120, 180, 480, and 1440 min, and the concentration of the selected pharmaceuticals in all the adsorption experiments was measured by liquid chromatography with tandem mass spectrometry (HPLC-MS/MS). The conditions of the HPLC-MS/MS analysis, along with the mass chromatogram of the selected pharmaceuticals, are given in the Supplementary Materials. Also, the influence of the initial pharmaceutical concentration (250, 500, 750, 1000, 1250, 1500, and 1750 μg/dm³) on the adsorption capacity of the tested materials was examined. Along with the adsorption experiment, a control experiment (solution without the adsorbent), which was carried out to ensure that the only interaction during the experiments was adsorption to the material used, showed that there was no interaction between the pharmaceutical components and no decrease in concentration during the experiment. Although the initial concentrations in the selected range were relatively high compared to typical concentrations in real water systems, the initial concentrations were chosen to examine the adsorption capacities of the obtained materials and ensure that the concentrations following adsorption were above the limit of detection. Potential leaching of toxic constituents from the adsorbent was examined by immersing 0.05 g of modified fly ash samples in 50 cm³ of water with the initial pH adjusted to 4 (for FA_ref) and 5.5 (for FA_ac/Ch and FA/Ch_mag). Leached element concentrations were measured using inductively coupled plasma optical emission spectroscopy (ICP-OES). Experimental data were analyzed using theoretical models: pseudo-first-order (Equation (1)) [24] and pseudo-second-order (Equation (2)) [25] order kinetic models for the adsorption kinetics, and the Langmuir (Equation (3)) [26] and Freundlich (Equation (4)) [27] isotherm models for equilibrium data interpretation.

$$q_t=q_e·\left(1-e^{-k_{1}t}\right)$$

(1)

$$q_t=q_e-{\left({\displaystyle \frac{1}{q_e}}+k_2·t\right)}^{-1}$$

(2)

$$q_e={\displaystyle \frac{Q_0·b·C_e}{1+b·C_e}}$$

(3)

$$q_e=K_f·C_e^{1/n}$$

(4)

In Equations (1)–(4), q_e and q_t (μg/g) are the amounts of dyes adsorbed at equilibrium and at time t (min), respectively; C_e is the equilibrium dye concentration (μg/dm³); Q₀ is the amount of solute adsorbed per unit mass of adsorbent required for monolayer coverage of the surface (μg/g); b is a constant related to the heat of adsorption (dm³/μg); K_f (μg g⁻¹(μg dm⁻³)^−1/n) is the Freundlich constant, related to the adsorption capacity; 1/n is the heterogeneity factor; and k₁ (1/min) and k₂ (g/(μg min)) are the pseudo-first-order and pseudo-second-order rate constants.

The possibility of using modified fly ash for water treatment in a flow system was examined on a laboratory scale using a column experiment. A polypropylene column was filled with modified fly ash (0.4 g) between two polyethylene frits. The water-based solution of the pharmaceuticals (50 cm³, initial concentration 10 µg/dm³ per compound) was passed through the column with a flow rate of 1 cm³/min. The pharmaceutical concentration in the effluent was determined using solid-phase extraction (SPE) as a sample preparation method followed by HPLC-MS/MS analysis. Preconcentration of the pharmaceuticals was performed using Oasis HLB cartridges, and the SPE procedure was as follows: The HLB cartridges were preconditioned with 5 cm³ of a mixture of methanol/dichloromethane (1/1) followed by 5 cm³ of deionized water. Effluent water samples were loaded on the HLB cartridges at a flow rate of 1 cm³/min and then dried under vacuum for 10 min. The pharmaceuticals were eluted with methanol/dichloromethane to extract a volume of 10 cm³. The extracts were evaporated to dryness under N₂, reconstituted with 1 cm³ of methanol, and filtered through 0.45 µm polyvinylidene fluoride filters before analysis.

2.4. Computational Modeling by Artificial Neural Networks

Due to obtaining a large amount of experimental data, it was necessary to properly process and analyze these data. In order to determine the correlation between the investigated parameters, identify those parameters that had the most significant influence on the adsorption process, and simplify and reduce the data set, artificial neural networks (ANNs) were used. All analyses were conducted using IBM SPSS Statistics 20. Unlike many statistical models that assume a linear relationship between response and predictor variables and their normal distribution, ANNs are able to map non-linear relationships between system characteristics. Artificial neural networks are non-linear and non-parametric classification methods. Artificial neural networks are defined as structures composed of densely connected adaptive simple processing elements. Those processing elements are called neurons or nodes that are able to perform massively parallel calculations for data processing and knowledge representation. ANNs have several advantages over traditional phenomenological or semi-empirical models, as they require known input data entered without assumptions. An ANN develops a mapping of input and output variables, which can later be used to predict the desired output as a function of the corresponding inputs. A neural network is characterized by its architecture (geometry), which represents the pattern of connections between neurons [28]. An ANN is particularly suitable when the underlying mathematical model is unknown or uncertain. Application of artificial neural networks is suitable, for example, in multivariate calibration when there is strong interference between analytes. Artificial neural networks, as their name suggests, use models of neural network structures, which is a very powerful computational technique for modeling complex non-linear connections, especially in situations where the explicit form of the relationship between the involved variables is not known [29,30].

Artificial neural networks are a widely accepted model for identification, analysis, and prediction and for design recognition and optimization. The network will tend to make the sum of the squared error (the sum of squared difference between the target network and the actual output result for a given input vector or set of vectors) as small as reported. Three different criteria were used to determine the efficiency of each selected network model: root mean square error (RMSE), bias (representing neurons that produce a constant signal), and coefficient of determination [31]. The root mean square error (RMSE) best describes the average measure of the error in predicting the dependent variable. Bias is the mean of all individual errors and indicates whether the model overestimates or underestimates the dependent variables. The coefficient of determination (R²) represents the percentage of the variability that can be explained by the model.

3. Results and Discussion

3.1. Material Characterization

Scanning electron microscopy was used to determine the modified fly ash samples’ morphology and surface structure (Figure 1). The applied treatments induced changes in the surface morphology, increasing the roughness and porosity compared to the unmodified fly ash. An increase in surface roughness was especially noticeable for the samples obtained by alkaline activation (FA_ac/Ch and FA_ref).

The results of the XRF screening analysis of the unmodified and modified samples are shown in Figure 2. Peaks corresponding to Cl, Ca, Fe, Ni, Zn, As, Rb, Sr, Zr, Nb, and Mo were observed on the XRF spectra of all the tested samples but had different intensities depending on the sample. Sample FA/Ch_mag showed the highest intensity of Cl, Ca, and especially Fe peaks due to the applied modification. Additionally, peaks for As, Rb, and Zr were not observed on the spectra of this sample. The peaks of Ni, Zn, Sr, Zr, Nb, and Mo were of a similar intensity for all the examined samples.

The results of the XRD analysis are presented in Figure 3. The XRD spectra of FA_ac/Ch and FA_ref showed the presence of similar mineral phases to those observed on the XRD spectrum of FA. These samples were characterized by quartz (Q), mullite (M), feldspate (f), and hematite (H) as the main phases of the mineralogical composition. The observed increase in peak intensity on 2θ of 26.6, 27.4, 27.8, and 50.1° for FA_ac/Ch and FA_ref compared with FA may be the consequence of the different grain stacking during the sample preparation [32,33]. On the other hand, the XRD spectrum of sample FA/Ch_mag did not display any distinguish peak, most likely due to the surface coverage by chitosan.

Fourier transform infrared spectroscopy (FTIR) was used to identify surface functional groups in the synthesized materials (Figure 4a).

The FTIR spectrum of the unmodified fly ash sample showed sharp peaks at around 1070 cm⁻¹ and 790 cm⁻¹, originating from asymmetric and symmetric stretching of the Si-O-Si bond, respectively. For the modified samples, the peak corresponding to the asymmetric stretching of the Si-O-Si bond was shifted to the lower wavenumbers: 1060 cm⁻¹ for FA/Ch_mag, 1015 cm⁻¹ for FA_ac/Ch, and 960 cm⁻¹ for FA_ref, which confirms that alkaline activation of the fly ash occurred for samples FA_ac/Ch and FA_ref. The broad band around 3300 cm⁻¹ observed for samples FA_ref and FA/Ch_mag can be assigned to the stretching vibration of the O-H bond in hydroxyl groups, while the two peaks of a low intensity on the FTIR spectra of FA/Ch_mag at 2850 cm⁻¹ and 2920 cm⁻¹ can be assigned to the symmetrical and asymmetrical vibrations of the C-H bond in methyl and methylene groups, respectively. A peak at about 1640 cm⁻¹ on the FTIR spectra of the modified samples may indicate absorbed water or the bending vibration of the N-H bond of the amino group in the samples modified with chitosan [34]. In the case of the FA/Ch_mag material, the presence of chitosan was confirmed, with bands at around 1030 cm⁻¹ and 1070 cm⁻¹ due to the stretching vibration of the C-O bond in the chitosan molecule [35]. In the FA_ac/Ch sample, the bands mentioned above were most likely superimposed with an intense and broad band of the Si-O-Si bond at about 1000 cm⁻¹. The bands around 567, 690, 773, and 789 cm⁻¹ indicate the formation of zeolitic material after the alkali (FA_ref) and hydrothermal (FA_ac/Ch) treatment of the fly ash [36].

The influence of applied modification on the surface charge of the tested samples was examined by the determination of the point of zero charge (PZC) (Figure 4b). The PZC is the pH value of the solution at which the observed charge of the material surface is zero, and the obtained pH_PZC values were 9.08 for FA/Ch_mag, 5.64 for FA_ac/Ch, and 6.20 for FA_ref. In solutions with a pH value less than pH_PZC, the protonation of functional groups occurs and the adsorbent behaves as positively charged, while in solutions with a pH value higher than pH_PZC, functional groups are deprotonated and the material behaves as negatively charged.

3.2. Adsorption Experiments

The alkali activation of the fly ash, along with the modification with chitosan and magnetic iron particles, increased the adsorption efficiency due to the increased surface roughness of the modified samples and the presence of surface functional groups introduced by the chitosan modification. The efficiency of the unmodified and modified fly ash samples to adsorb the selected pharmaceuticals is presented in Figure 5.

The adsorption process is influenced by different experimental parameters that need to be optimized to achieve the highest adsorption efficiency. Optimization of the adsorption parameters was performed through the adsorption of the selected pharmaceuticals onto the modified fly ash samples, FA/Ch_mag, FA_ref, and FA_ac/Ch. The initial pH of the pharmaceutical solution is a key factor influencing the adsorption efficiency. The influence of initial pH on the adsorption capacities of the examined samples is given in Figure 6a. In the case of the FA/Ch_mag sample, the adsorption experiment was not performed at an initial pH of 2 due to the dissolution of iron from the material’s surface. The pH value of the aqueous solution affects the ionization of the adsorbent’s surface groups, which serve as active sites for adsorption, as well as the solubility and ionic states of the pharmaceuticals. Depending on the pKa value and pH of the solution, the pharmaceuticals can be found as neutral, in a cationic form, when the functional groups are protonated, or in an anionic form due to the deprotonation of the functional groups. Due to the high value of pKa (Table S1), in the examined pH range, erythromycin was in a cationic form, while lorazepam and simvastatin were neutral. According to the pKa values (Table S1) in the solution with pH 2, all the other pharmaceuticals were in a cationic form except atorvastatin, which was neutral. In the solution with an initial pH value of 4, cilazapril, bromazepam, diazepam, and atorvastatin were neutral, doxycycline was a zwitterion [37], while clopidogrel was in a cationic form. With the further increase in the initial pH from 5.5 to 8, bromazepam, diazepam, and clopidogrel were neutral, cilazapril and atorvastatin were in an anionic form, while doxycycline retained its zwitterion structure and was in an anionic form above pH 8. Additionally, the pH value of the solution affected the surface charge of the modified fly ash samples.

According to the obtained pH_PZC values, FA/Ch_mag was positively charged in the examined pH range, FA_ac/Ch was positively charged below pH 5.64 and negatively charged at higher pH, while FA_ref was positively charged in the solutions with a pH less than 6.20. It is expected that the positively charged surface of materials will attract negatively charged pharmaceutical compounds and vice versa. According to the obtained results (Figure 6a), doxycycline, erythromycin, atorvastatin, and clopidogrel were highly adsorbed on all the examined samples from the solutions with an initial pH higher than 2. Negatively charged atorvastatin (pH > 5.5) was highly adsorbed on the positively charged surface of FA/Ch_mag, while for samples FA_ac/Ch and FA_ref, an increase in the pH value led to a slight decrease in the adsorption capacities due to the deprotonation of the surface functional groups of the material. The high adsorption capacities of the examined samples for doxycycline in the pH range of 4–8 was most likely related to the presence of doxycycline in zwitterion and anionic forms and its electrostatic attraction to the material surface. However, the materials’ affinity towards cationic erythromycin and the cationic/neutral form of clopidogrel indicates that electrostatic attraction was not the key factor influencing the adsorption process. The good adsorption of erythromycin and clopidogrel may be a consequence of hydrogen bonds between hydrogen atoms from the material surface (-OH groups) and oxygen or nitrogen atoms from the pharmaceutical structure. Also, the high adsorption capacities obtained for the mentioned pharmaceuticals may be a consequence of the relatively higher pK_ow values (Table S1) compared to the other investigated pharmaceuticals. Generally, all the examined samples showed good adsorption efficiency in the solution without pH adjustment. However, the highest adsorption capacity of sample FA_ref was achieved for the initial pH value of 4, while samples FA/Ch_mag and FA_ac/Ch showed the highest adsorption capacities at an initial pH of 5.5. Therefore, for the further adsorption experiments on FA_ref, pH = 4 was chosen as the optimal condition, while the adsorption experiments on FA/Ch_mag and FA_ac/Ch were performed without pH adjustment.

The following step in the optimization of the adsorption process was to select the most appropriate ratio of the adsorbent mass to the adsorbate volume (Figure 6b). As can be seen, the capacities of all the examined samples for the adsorption of the selected pharmaceuticals decreased with an increase in the mass-to-volume ratio. Therefore, a ratio of the adsorbent mass to the adsorbate volume of 1 g/dm³ (adsorbent mass of 0.05 g and adsorbate volume of 50 cm³) was chosen as optimal for the further adsorption experiments.

The influence of the contact time on the adsorption capacities of the modified fly ash samples is given in Figure 7.

Most pharmaceuticals did not reach adsorption equilibrium in the examined period. Depending on the adsorbent, only doxycycline, simvastatin, clopidogrel, and erythromycin reached equilibrium in the first 60 min of adsorption, with a removal efficiency of approximately 80%.

The experimental data were examined by pseudo-first- and pseudo-second-order kinetic models, and the calculated kinetic parameters are presented in

Table 1. Kinetic parameters for pharmaceutical adsorption onto modified fly ash samples.

Sample	Analyte *	Pseudo-First-Order			Pseudo-Second-Order			qe,exp, µgg−1
		qe,cal, µgg−1	k1, min−1	R2	qe,cal, µgg−1	k2 × 105, g(µgmin)−1	R2
FA/Chmag	Do	221	0.0783	0.73163	235	52.0	0.92902	246
	Ci	184	0.0108	0.98985	207	6.34	0.97181	189
	E	238	0.0112	0.98502	266	5.32	0.95894	237
	B	186	0.0159	0.97777	203	11.1	0.97634	198
	L	144	0.0147	0.89495	157	13.5	0.9496	156
	Di	239	0.0177	0.93875	261	9.71	0.92521	247
	A	235	0.0289	0.91834	254	16.8	0.96762	245
	Clo	218	0.0790	0.84053	233	50.8	0.96665	246
	S	211	0.2913	0.61521	218	258	0.88395	229
FAac/Ch	Do	194	0.1518	0.50562	205	116	0.83764	215.5
	Ci	202	0.0126	0.99147	224	7.23	0.98453	207
	E	204	0.0124	0.98337	226	7.19	0.98536	212
	B	104	0.0126	0.98946	116	13.9	0.98173	107
	L	132	0.0096	0.98691	149	7.74	0.96804	134
	Di	144	0.0135	0.98271	158	11.4	0.97858	146.5
	A	233	0.0187	0.98382	256	9.80	0.98397	239.8
	Clo	214	0.0371	0.81057	233	22.0	0.93871	239.1
	S	223	0.2105	0.57548	234	138	0.8603	243.5
FAref	Do	134	0.0129	0.98278	148	11.5	0.98897	140
	Ci	129	0.0174	0.90487	141	17.9	0.95044	138
	E	191	0.0933	0.59572	201	76.7	0.86784	205
	B	150	0.0122	0.97364	167	9.67	0.99225	159
	L	142	0.0106	0.99036	159	8.39	0.97847	144
	Di	135	0.0155	0.97614	149	14.1	0.99239	142
	A	202	0.0142	0.9703	224	8.21	0.97661	212.5
	Clo	230	0.0282	0.90504	250	16.3	0.96641	245
	S	220	0.0839	0.79322	233	58.0	0.95531	239

* Do: doxycycline, Ci: cilazapril, E: erythromycin, B: bromazepam, L: lorazepam, Di: diazepam, A: atorvastatin, Clo: clopidogrel, S: simvastatin.

.

Additionally, the values of the experimental equilibrium adsorption capacities (q_e,exp), obtained after 24 h when the adsorption reached equilibrium, are also summarized in Table 1. According to the correlation coefficients, the adsorption process of the majority of the examined pharmaceuticals onto the modified fly ash samples followed pseudo-second-order kinetics. Nevertheless, the adsorption of cilazapril, erythromycin, bromazepam, and diazepam onto FA/Ch_mag, cilazapril, bromazepam, lorazepam, and diazepam onto FA_ac/Ch, and lorazepam onto FA_ref were better described by pseudo-first-order kinetics. A similar conclusion can be drawn by comparing the adsorption capacity values obtained experimentally and those calculated using the applied kinetic models, except for the adsorption of bromazepam onto FA/Ch_mag, the adsorption of erythromycin and atorvastatin onto FA_ac/Ch, and the adsorption of doxycycline and atorvastatin onto FA_ref. The highest values of the pseudo-first- and pseudo-second-order constant rates were obtained for simvastatin, clopidogrel, doxycycline, and erythromycin, thus confirming the fast adsorption of these pharmaceuticals.

The influence of the initial concentration on the efficiency of the examined materials to adsorb the pharmaceuticals is given in Figure 8a. It was observed that with an increase in the initial pharmaceutical concentration, the adsorption efficiency decreased. To investigate the adsorption process at equilibrium and examine how the initial concentration of the pharmaceuticals impacted the adsorption, the equilibrium adsorption data were analyzed using the Langmuir and Freundlich isotherm models (Figure 8b).

Across the range of initial concentrations tested, the adsorption capacity of all the samples consistently increased with the initial pharmaceutical concentrations, which was especially pronounced for the adsorption of atorvastatin onto all the samples, doxycycline onto FA/Ch_mag and FA_ref, simvastatin onto FA/Ch_mag and FA_ac/Ch, clopidogrel onto FA_ac/Ch and FA_ref, as well as the adsorption of erythromycin and cilazapril onto FA_ac/Ch. This trend suggests the presence of unoccupied active sites on the surface of the modified fly ash samples, which were available for further adsorption. The adsorption equilibrium data were fitted with the Langmuir and Freundlich isotherm models, and the obtained isotherm parameters are summarized in

Table 2. Isotherm parameters for pharmaceutical adsorption onto modified fly ash samples.

Sample	Analyte *	Freundlich’s Isotherm			Langmuir’s Isotherm
		Kf, µg1−1/nL 1/n g−1	1/n	R2	Q0, µgg−1	b, dm3/μg	R2
FA/Chmag	Do	186.86	0.18	0.78523	519.47	0.021	0.92283
	Ci	44.07	0.30	0.9034	463.45	0.029	0.91625
	E	113.41	0.20	0.87565	428.89	0.050	0.98225
	B	63.89	0.19	0.69111	227.50	5.62·10−8	0.90992
	L	26.63	0.33	0.93055	293.72	4.14·10−3	0.99787
	Di	160.93	0.08	0.57722	274.06	2.10·10−4	0.94106
	A	186.78	0.18	0.70056	517.54	5.21·10−3	0.98139
	Clo	137.55	0.20	0.87801	595.55	0.153	0.90217
	S	101.87	0.23	0.60934	427.57	1.29·10−13	0.92858
FAac/Ch	Do	78.39	0.31	0.64927	542.79	2.75·10−4	0.77459
	Ci	43.71	0.38	0.80849	498.75	2.19·10−4	0.98921
	E	71.65	0.32	0.83242	631.85	0.01263	0.87541
	B	8.01	0.52	0.86147	393.94	5.45·10−4	0.87533
	L	6.76	0.57	0.91695	443.68	3.11·10−4	0.95261
	Di	10.60	0.54	0.8374	546.95	7.09·10−4	0.84036
	A	149.39	0.25	0.74729	625.52	6.74·10−3	0.97016
	Clo	96.94	0.28	0.8133	594.46	0.01766	0.92075
	S	90.03	0.26	0.86299	482.22	4.28·10−11	0.04654
FAref	Do	5.01	0.75	0.68722	557.21	7.81·10−13	0.95545
	Ci	11.21	0.53	0.89257	531.13	8.56·10−4	0.84577
	E	23.69	0.41	0.88847	2089.77	9.30·10−3	0.90764
	B	23.24	0.36	0.99691	9681.34	2.34·10−3	0.96305
	L	12.83	0.42	0.96800	646.98	0.01039	0.72683
	Di	7.38	0.60	0.87469	802.32	1.49·10−3	0.73685
	A	39.11	0.46	0.78965	627.25	2.01·10−5	0.94944
	Clo	99.70	0.28	0.95135	2706.44	0.03502	0.85879
	S	77.40	0.25	0.78019	80161.93	9.66·10−4	0.78745

* Do: doxycycline, Ci: cilazapril, E: erythromycin, B: bromazepam, L: lorazepam, Di: diazepam, A: atorvastatin, Clo: clopidogrel, S: simvastatin.

.

According to the correlation coefficient values (R²), the adsorption of the selected pharmaceuticals onto the modified fly ash samples can mainly be described by the Langmuir isotherm. However, the adsorption of simvastatin onto FA_ac/Ch and the adsorption of cilazapril, bromazepam, lorazepam, diazepam, and clopidogrel onto FA_ref fit better with the Freundlich isotherm model. The highest values of the Freundlich constant, K_f, which is related to the adsorption capacities, were obtained for the adsorption of six pharmaceuticals (doxycycline, erythromycin, diazepam, atorvastatin, clopidogrel, and simvastatin) onto FA/Ch_mag, indicating that this material showed high affinity for the selected pharmaceuticals. These observations were confirmed by the experimentally obtained equilibrium adsorption capacities (values for q_e,exp in Table 1). The maximum adsorption capacities (Q₀) obtained by the Langmuir model were very high, especially for the adsorption onto FA_ref which can be attributed to the lack of equilibrium surface saturation in the examined concentration range. According to the FTIR and XRD analyses, the surface of sample FA/Ch_mag was covered with chitosan, the functional groups of which acted as active sites for pharmaceutical adsorption, leading to the good adsorption properties of FA/Ch_mag. The heterogeneity factor (1/n) values were less than unit (Table 2), especially for sample FA/Ch_mag, indicating the homogeneous surface of the adsorbent. Considering the magnetic properties of this sample, which enabled the easy separation from the adsorption solution using a magnetic field, FA/Ch_mag can be regarded as a good adsorbent for removing pharmaceutical residues from water. A comparison of the adsorption capacities of the examined fly ash samples with the adsorption capacities for the tested compounds by different adsorbents reported in the literature is shown in

Table 3. Adsorption capacities of different adsorbents used for pharmaceutical adsorption.

Adsorbent	Pharmaceutical	Initial Concentration, mg/dm3	Adsorption Capacity, mg/g	Removal Efficiency, %	Ref.
Rice husk ash	Doxycycline	10–90	17.74	98.85	[38]
Powdered activated carbon from pumpkin seed	Doxycycline	10–100	23.6	85.82	[39]
Rice straw biochar	Doxycycline	5–60	8.93–108.42		[40]
Fly-ash-based synthetic zeolite Carbon–zeolite composite	Erythromycin	0.1–200	314.7 363.0	94–99	[41]
Pyrolyzed industrial waste oil sludge and sewage sludge	Erythromycin	0.1–200	0–2	0–3	[42]
Biochar from cotton gin waste and guayule bagasse	Erythromycin	2–50	17.123	50–70	[43]
Activated charcoal Micelle–clay complex	Diazepam	0.01–100	28.9 ± 1.9 31.2 ± 1.7		[44]
Activated charcoal Micelle–clay complex	Atorvastatin/ Simvastatin	0.01–100	9.1/11.9 23.2/24.4		[45]
Natural zeolite	Diazepam	1–25	8.051–8.259		[46]
MIL100(Fe) MIL100(Fe)@CMC	Lorazepam	80	150 811	80 95	[47]
Carboxymethylcellulose–iron	Atorvastatin	10–60	(10–32) × 10−3	100–53	[48]

. Generally, the adsorption capacities reported in the literature were higher than those obtained in this study (Table 1 and Table 2) due to the higher initial concentrations of the pharmaceuticals used and the different experimental conditions. Also, when comparing the adsorption capacities reported in the literature with those obtained in this study, it is important to consider that the adsorption experiments were mainly conducted using single-component solutions for pharmaceutical adsorption. Nevertheless, the removal efficiencies (Figure 8a) of the examined pharmaceuticals onto the modified fly ash samples presented in this study were comparable to or higher than those shown in the literature (Table 3).

Due to its good adsorption properties, sample FA/Ch_mag was used as an adsorbent for pharmaceutical removal in the flow system study. The high efficiency (Figure 9) of FA/Ch_mag for the adsorption of the pharmaceuticals from the solution with a low initial concentration, simulating real water conditions, indicates its potential for water treatment in a flow system.

The results of leaching experiments, which were performed to examine the presence of potentially toxic elements leached from the modified fly ash during the adsorption experiments, are presented in

Table 4. Concentration of elements leached from modified fly ash samples during adsorption experiments.

Sample	Concentration of Leached Elements, mg/dm3
	Ag	Cd	Co	Cr	Cu	Fe	Mn	Ni	Pb	Zn	Al	Ca	K	Mg	Na
FA/Chmag	n.d.	n.d.	n.d.	n.d.	0.02	0.07	n.d.	n.d.	n.d.	0.01	0.45	7.8	0.6	5.25	32.2
FAac/Ch	n.d.	n.d.	n.d.	n.d.	0.02	0.08	0.01	0.01	n.d.	0.01	0.40	2.2	0.5	0.8	12.1
FAref	n.d.	n.d.	n.d.	n.d.	0.01	0.12	n.d.	n.d.	n.d.	n.d.	1.2	0.4	0.4	n.d.	32.3
WHO guideline value *	0.1	0.003	-	0.05	2.00	0.009	0.08	0.07	0.01	-	0.9	-	-	-	-

* Guideline value given by the World Health Organization [49]; n.d.—not detected.

. The obtained results show that the concentration of Fe leached from all the samples and the concentration of Al leached from FA_ref were higher than the WHO guideline values [49]. However, according to the Environmental Protection Agency National Secondary Drinking Water Regulations [50], only the concentration of Al exceeded the permitted concentrations.

Additionally, modified fly ash samples loaded with pharmaceuticals pose a potential environmental threat and should not be disposed of in the environment. However, after adsorbing pharmaceuticals, these materials can be repurposed as raw materials for geopolymer production. In this process, the pharmaceuticals will be fully degraded at the high temperatures involved, while the resulting geopolymer can be utilized in the construction industry, particularly for concrete production.

3.3. Computational Modeling by Artificial Neural Networks

A computational modeling of the data was performed in order to determine the correlation between the examined parameters in order to discover which of the parameters had the highest impact on the adsorption process and to reduce the amount of data. The goal was to establish a correlation between the investigated parameters of adsorption, the materials used, and the pharmaceuticals in order to determine which parameters had the most influence on the adsorption in order to reduce the number of experiments and the number of variables that would be required in further research. Depending on the type of data, taking into account the assumptions of the application of certain tests and the research design, artificial neural networks were used for this purpose. Artificial neural networks represent a complex computational model, which, unlike all others, assumes a non-linear correlation between parameters. Similar to the human brain, it has the ability to learn. The type of network used for data processing is feed-forward, while the type of learning used is backpropagation. The drug concentrations after adsorption, under different conditions, were in the input layer, whereas the materials that were used as the adsorbents, as well as the parameters of adsorption (m/V, pH, Co, and t), were in the output layer (Figure 10a).

The artificial neural network automatically selected two units in the hidden layer after several iterations (training and testing). Figure 10b shows the results of the modeling using an artificial neural network—they refer to the materials used in the adsorption process in order to determine which modified material had the best adsorption characteristics and performance. The results obtained on the scatter diagram (Figure 10b), based on the coefficient of determination, show that the most effective adsorption process was demonstrated by the material FA_ac/Ch, followed by FA/Ch_mag, while the material FA_ref showed the poorest characteristics in terms of the adsorption of the selected pharmaceuticals. The modeling and optimization of the adsorption parameters can accelerate this process the most, predominantly with the materials FA_ac/Ch and FA/Ch_mag. Figure 10c shows the results of the modeling using an artificial neural network, which relates to the parameters of the adsorption process. The results obtained and shown in Figure 10c, referring to the impact of the adsorption parameters, show that the initial concentration of the pharmaceuticals present and the pH value of the solution had the strongest influence on this process, followed by the contact time, while the parameter representing the ratio of the mass and volume of the adsorbate showed a lesser influence and lower statistical significance.

Figure 11 shows the results of the modeling using an artificial neural network, which refers to the tested pharmaceuticals, in order to determine which of them had the highest adsorption capacity, i.e., the best removal efficiency from the water solutions.

The highest level of significance in terms of the adsorption onto the modified materials was shown by clopidogrel, doxycycline, erythromycin, cilazapril, and lorazepam (over 50%). Taking all the adsorption parameters into account, the network came to the conclusion that these pharmaceuticals will be adsorbed onto the modified fly ash the most, and this prediction will help further research into the application of new modified materials as adsorbents and help to test their performance in removing other drug residues from water.

4. Conclusions

This study explored the use of modified fly ash as an effective adsorbent for removing pharmaceutical residues from water, addressing the growing issue of fly ash accumulation in landfills. The fly ash was treated through alkali activation with NaOH and reflux, followed by modification with chitosan, which introduced functional groups that enhanced the pharmaceutical adsorption. Additionally, magnetic iron oxide was incorporated to facilitate the easier separation of the adsorbent after use. The alkali activation and hydrothermal treatment resulted in increased surface roughness, while the characterization of the materials revealed consistent mineral phases and elemental compositions across all the samples. Notably, the presence of chitosan on the surface of the fly ash modified with chitosan and magnetic iron oxide appeared to enhance the adsorption properties. The experimental parameters of the pharmaceutical adsorption onto the modified fly ash were optimized, and multivariate chemometric methods indicated that the pH value, initial concentration of the pharmaceuticals, and contact time significantly influenced the adsorption process. Data analysis through kinetic and isotherm models revealed that the adsorption followed pseudo-second-order kinetics and that the equilibrium could be described by the Freundlich isotherm model. The fly ash modified with chitosan and magnetic iron oxide exhibited the highest adsorption efficiency, achieving over 80% removal for most of the pharmaceuticals tested. Furthermore, artificial neural networks effectively modeled the adsorption characteristics, allowing for predictions that can reduce the need for extensive experimental trials.