The titanium dioxide TiO₂ (Degussa P25 Evonik) nanopowder with a particle size of 21 nm and measured specific surface area of 56 m² g⁻¹ (BET) was purchased from Sigma-Aldrich and used as a photocatalyst. The 1.5 wt% TiO₂ suspension was prepared by adding the nanopowder into deionized water. The suspension was sonicated using an ultrasound bath (RK 255 H, Bandelin Sonorex) for 30 min and immobilized onto a glass fiber filter purchased from LydAir® with a basic weight of 68 g m⁻² and 22% penetration (0.3 μm DEHS @5.33 cm s⁻¹). The loading of TiO₂ was set to 12 g m⁻². The final mass loading of TiO₂ was determined by weighing the dry filter before and after the immobilization process.

Three aerosol deposition methods were employed to immobilize the TiO₂ catalyst onto the glass fiber substrates. The suspension of TiO₂ nanoparticles (NPs) was atomized using an atomizer from TSI (Model 3079A) and a spray gun (3.5 bar Aerotec 1.8 mm). First, the aerosol containing colloidal TiO₂ droplets was generated by the TSI atomizer and exposed to different conditions. The relative humidity conditions of 30, 55, and 85%RH were controlled by a diffusion dryer and dry compressed air (Figure 1A), which is referred to as the spray-drying method (SD). The relative humidity and temperature during the deposition process were monitored using a Sensirion SHT7x sensor. Second, the atomized TiO₂ droplets were directly deposited on the filter media corresponding to the spray atomization (SA) method (Figure 1B). In both cases, the depositions continued until the loadings of 12 g m⁻² were obtained. The exact loadings of TiO₂ were determined by the subtraction of the pristine filter mass from the mass of the filter after the coating and drying steps. The drying step was conducted under the ambient temperature, pressure, and relative humidity. The velocity of deposition, that is, the airflow velocity in the deposition chamber (Figures 1A and B), was 5 cm s⁻¹ for both SD and SA methods. The size distributions of particles generated under the SD and SA conditions were measured upstream and downstream using a scanning mobility particle sizer and an aerodynamic particle sizer (Supplementary Figure S1). Third, the spray gun (SG) was connected to compressed air, and the sprayed TiO₂ droplets were deposited with 50 L min⁻¹ flow rate at a distance of 10 cm from the filter media. After the filter was covered with a layer of suspension, it was dried, and the process was repeated until the loadings of 12 g m⁻² were obtained. In addition, a dip-coating method was used as a reference; a filter was dipped into the TiO₂ suspension for 10 min, dried, and dipped again until the required loading was obtained.

The final crystalline phase of the produced samples was obtained with X-ray diffraction (XRD) on a PANalytical (X’Pert PRO) powder diffractometer equipped with a copper anode (Cu Kα radiation) and with X’Celerator detector. The time of a scan was set to 30 min, and the scan range was from 10 to 90°. The microstructures of as-prepared TiO₂ nanoparticle clusters were characterized by transmission electron microscopy (TEM, JEOL2200FS), operating at the voltage of 200 kV. TiO₂ nanoparticles were collected in situ after the exposure to different operational conditions on the ultrathin carbon film-supported Cu TEM grid (300 mesh, Ted Pella). To characterize TiO₂ agglomerates deposited on the substrates, scanning electron microscopy (SEM) (Generation five Phenom ProX and FEI Quanta 650 FEG ESEM) was used. In order to prevent surface charging during imaging, the samples were sputtered with a 10-nm platinum layer. The size distribution and aspect ratio were estimated from the SEM images of particles deposited onto Nuclepore filters using ImageJ software.

Gas adsorption (BET) was used to measure the specific surface area, pore size, and volume distribution of materials through nitrogen adsorption (Micrometrics), and the software (MicroActive for 3Flex, Version 4.04, Micromeritics) was used for the computation of results. Nitrogen adsorption and desorption isotherms were measured at liquid nitrogen temperature (−196°C) on a Micromeritics 3Flex instrument after degassing the samples at 2.5 × 10^–2 mmHg and 100°C for 20 h. The specific surface areas were calculated from seven data points in the linear range of P/P₀ between 0.05 and 0.3 using the Brunauer–Emmett–Teller (BET) method. The average pore width and total pore volume were estimated from the Barrett–Joyner–Halenda (BJH) model assuming cylindrical pores and from the non-local density function theory (NLDFT) assuming slit pore dimensions.

Ultra-small-angle X-ray scattering (USAXS) and small-angle X-ray scattering (SAXS) measurements were carried out using the 9ID USAXS instrument at Advanced Photon Source, Argonne National Laboratory (Ilavsky and Jemian, 2009; Ilavsky et al., 2013). The combined q range was between 1.10⁻⁴ Å⁻¹ and 1.3 Å⁻¹, where q was equal to 4π/λ sin(θ) (λ was the wavelength, and θ was ½ of the scattering angle). The X-ray energy was 21 keV (λ = 0.5895 Å). The X-ray photon flux was ≈5∙10¹² mm⁻²s⁻¹. The samples were prepared by the deposition of powders in the center of a steel washer (outer diameter 8.55 mm and inner diameter 1.5 mm), which was closed on both sides with a scotch tape. The mass of powders was measured using a microbalance. Depending on the sample, the mass was between 31 and 50 mg.

Thermogravimetric analysis (TGA) was performed on a Perkin Elmer TGA 8000 under a constant flow of nitrogen of 40 ml/min. First, the samples were heated from 30 to 120°C at the rate of 20°C/min and held for 10 min to remove the physically adsorbed water. After that, the temperature was elevated from 120 to 500°C (at 20°C/min) to determine the amount of adsorbed hydroxyl groups. A Thermo Scientific ESCALab 250Xi XPS (Thermo Fisher Scientific) with 200-W monochromated Al Kα radiation was used to perform X-ray photoelectron spectroscopy (XPS) analysis. All binding energy values were calibrated by fixing the C1s pattern to 285 eV. Gaussian–Lorentzian components were used for peak deconvolutions after a Shirley background subtraction. The relative area ratio of XPS peaks was used for sample comparison.

Diffuse reflectance spectroscopy was used to characterize the optical properties of the coated filters. The UV-Vis-NIR Spectrophotometer (UV-3600, Shimadzu) was used to measure the transmittance and reflectance of light using three detectors (photomultiplier tube (PMT), In-GaAs, and cooled PbS) that enabled the measurement over the ultraviolet, visible, and near-infrared regions (300–1,500 nm). A high-performance double monochromator was used for the measurements. A high-speed USB pressure sensor (PX409 Series Pressure Transducer, Model No. PX409-10WDWUUSBH, Omega) with a micro-machined silicon design and the viewing software (Digital Transducer Application, Omega) were used to measure the pressure drop of coated and uncoated glass fiber filter to determine the change in performance. The flow velocity applied to the samples was 5 cm s⁻¹. The flow conditions in the system provided laminar flow allowing using Darcy’s law for the determination of permeability of the filter media before and after immobilization of TiO₂ particles. The stability of TiO₂ NPs on the coated filter was tested under high-stress conditions. The procedure was adapted from Wohlleben et al. (2017). A square piece of 2 cm × 2 cm of each sample was immersed in 6 ml of the solution prepared by adding 5.7 ml glacial acetic acid in 500 ml of reagent water and 64.3 ml of 1 M NaOH, which was diluted to a volume of 1 L. The solution with the sample was held in a 10 ml cylindrical glass container and sonicated in an ultrasonic bath (RK 255 H, Bandelin Sonorex) for an hour. The resulting specimens were sonicated for an extra half hour after the samples were removed in order to eliminate the agglomeration and sedimentation of particles before absorbance analysis using a UV-Vis spectrometer.

The X-ray scattering curves obtained from ultra-small scattering measurements were analyzed using the Unified fit in the “Irena” software package (Ilavsky and Jemian, 2009; Ilavsky, 2018). The Unified fit uses the equation for the scattered intensity as derived by Beaucage (1995); Beaucage (1996); Beaucage et al. (1997): I i ( q ) =   G i ⋅ e − q 2 ⋅ R g 2 3 +   e − q 2 ⋅ R g _ c o 2 3   B i   { ( e r f ( q ⋅ R g 6 ) ) 3 q } P i . (1)

The relevant fit parameters are the radii of gyration R _g , the exponent of decay P _i , and the two constants G and B. The Unified fit splits the scattering curve into different structural levels with a Guinier regime reflecting the structure’s radius of gyration and a power law regime, which displays the type of the structure (Londoño et al., 2018). The first structural level describes the primary particle radius of gyration R _g1 , which is the mean square distance from the center of electron density of a particle (in analogy to the momentum of inertia in mechanics) and is converted to the diameter of spherical primary particles as: D p p = 2 ⋅ 5 3 ⋅ R g 1 . (2)

From the slope of scattering intensity decay, fractal dimension D_f can be assessed as the power law regime following I(q) ∼ q^Df, which allows investigation of the structure of the aggregates. The radius of gyration (R_g2) of the second structural level was used for the determination of the radius and diameter of agglomerates using Eq. 2 (Braun et al., 2004). The degree of aggregation is characterized by a parameter z that is directly obtained from a constant G for primary particles (G _pp ) and aggregates (G _agg ) as z = G a g g / G p p (Hyeon-Lee et al., 1998). The radius of gyration for aggregates is determined from the diameter of primary particles, aggregation number z, and fractal dimension as: R g a g g =   ( D p p 2 ⋅ z 2 D f ( 1 + 2 D f ) ⋅ ( 2 + 2 D f ) ) 1 2 . (3)

The evaluation of catalytic degradation activity was performed in a stainless steel reactor of the filter holder type, which was designed and assembled according to a previous study (Zhao et al., 2020). The filter coated with the catalyst was installed between the upper and bottom parts of the photoreactor and supported by a stainless steel mesh. The upper part consisted of a boron silicate glass cover allowing the filter illumination using a lamp placed horizontally above the reactor. Toluene mixed with air was introduced into the reactor chamber through two opposite inlets facilitating better flow distribution. The concentration of toluene was monitored by gas chromatography GC-2014 (Shimadzu) equipped with an Rtx-Wax separation column and an FID-2014 (Shimadzu) detector (Figure 2). The initial toluene concentrations were 5, 20, and 40 ppm. After the satisfactory introduction of the toluene–air mixture into the reactor, the system was closed, and the gas was circulated until reaching a steady-state condition. The temperature and relative humidity inside the reactor were approximately 22°C and 20% RH, respectively. Several control tests, such as adsorption or photolysis evaluation, were performed (Supplementary Figures S2A–C). The samples were irradiated using six UV BLB lamp bulbs (368-nm) emitting an intensity of 0.6 mW cm⁻² (Supplementary Figure S2D).

The photodegradation rate of toluene was evaluated following the Langmuir–Hinshelwood (L-H) kinetic model (Coronado et al., 2013). The initial degradation rate (r₀) is observed to be a function of the initial concentration (C₀). A linear plot of r₀ ⁻¹ versus C₀ ⁻¹ allowed us to obtain the L-H rate constant k and Langmuir adsorption constant K _L-H as follows: r 0 = k L − H K L − H C 0 1 + K L − H C 0   o r   1 r 0 = 1 k L − H K L − H 1 C 0 + 1 k L − H . (4)

For diluted systems, where the concentration is lower than 10⁻³ M (1,000 mmol m⁻³), KC ₀ becomes much lower than 1, and the reaction follows an apparent first-order kinetic model (Herrmann, 1999). Therefore, a simplified pseudo-first-order kinetic model was applied for the degradation of toluene as (Liu et al., 2005) follows: l n C 0 C = k 1 ⋅ t . (5) where C ₀ is the initial concentration of toluene, C is the concentration of toluene at the specific reaction time t in min, and k ₁ is the rate constant of the pseudo-first-order model in min⁻¹. The reaction constant can be obtained from the slope of lnC ₀ /C versus t plot. When R² is larger than 0.95, the experimental data are considered to be in good agreement with this model (Fox and Dulay, 1993).

The photocatalytic activity of TiO₂-loaded filters prepared by different aerosol-assisted methods was evaluated using the pseudo-first-order kinetic and Langmuir–Hinshelwood (L-H) models (Liu et al., 2005; Debono et al., 2011; Kim et al., 2012). Toluene has been selected as a model pollutant for the evaluation as its photocatalytic degradation mechanism has been investigated extensively (Marcı̀ et al., 2003; Sleiman et al., 2009; Bianchi et al., 2014; Chen et al., 2021). The results from the pseudo-first-order kinetic analysis showed that the photocatalytic activity depended on the preparation conditions of the photocatalytic filters, and the highest activity was achieved with the samples prepared by SA and SG methods (Figures 10A and B). The SG method showed the best photocatalytic activity followed by the SA method, whereas SD 30RH% revealed the worst activity. These methods were further evaluated using the L-H model.

The L-H model analysis revealed information about the adsorption properties and degradation rate of toluene. It was shown that the adsorption constant and site accessibility of coatings toward the toluene pollutant decreased in the order of SD 30%RH > SA > SG conditions (Supplementary Table S1). The decreased adsorption was also represented by the coverage Θ (Figure 10C). Although the adsorption of toluene on the TiO₂ surface was greater for the SD method, the reaction rate was lower for SA and SG methods. This behavior was attributed to a strong competition between adsorbed water and pollutant (e.g., toluene) (Zhang et al., 2020). Therefore, the adsorbed water that remained on the sample surface after the preparation and drying process was analyzed using a TGA technique. The TGA analysis revealed that the SA and SG methods led to greater hydration remaining on the TiO₂ surface (Figure 8B). As reported in several studies, adsorbed water can occupy the active sites and limit the accessibility of pollutant molecules when the amount of adsorbed water is high (Coronado et al., 2003; Jeong et al., 2013). On the other hand, the presence of water and hydroxyl groups promotes the formation of hydroxyl radicals, which are generally considered to initiate the degradation of adsorbed pollutants upon irradiation (Coronado et al., 2003; Park et al., 2012). Moreover, the adsorbed water on the TiO₂ surface stabilizes photogenerated charge carriers and delays the electron–hole recombination compared to the dehydrated TiO_2, as shown by Litke et al. (2017). Furthermore, it was reported that the increasing amount of adsorbed water/hydroxyl groups enhanced the complete oxidation of toluene molecules (Jeong et al., 2013). The photocatalytic reaction is a complex process involving many steps and is affected by multiple factors. Based on our results, the enhanced degradation performance of toluene using SA and SG samples was attributed to the above-discussed hydration along with the structural features of TiO₂ coating, for example, larger pore size and lower fractal dimension of agglomerates. The hydration provided the enrichment of the reactive species formation (i.e., hydroxyl radicals) during the irradiation of TiO_2, while the structural features enabled efficient irradiation (Brosillon et al., 2008; He et al., 2014; Sun et al., 2018). In our study, irradiation was not the limiting factor for the photocatalytic activity of TiO₂ deposited by aerosol-assisted methods. As shown by the optical properties of prepared samples (Figure 9), the sharp change in the UV absorption between SD and SA/SG methods does not reflect the continuous increase in the reaction rate (Figure 10B). Therefore, the superior reaction rate of SA and SG samples is mainly attributed to the synergistic effect of the agglomerate and pore properties together with the amount of remaining water molecules on the surface/in the pores of TiO₂ agglomerates. Specifically, for SD 30%RH, the amount of water and pore size and volume was small, and thus, the amount of generated radicals was the most limited. As the water remained on the surface and/or into pores due to the SA and SG preparation conditions, the formation of reactive species and slower electron–hole recombination enhanced the degradation efficiency of toluene. These findings are consistent with the study of He et al. (2015), who highlighted the importance of pore size and wide pore size distribution for the photocatalytic activity of mesoporous TiO₂. The large pore size distribution enhances the accessibility of small and otherwise separated pores and therefore facilitates the adsorption and competition of molecules, efficient light activation, and reactive species generation (He et al., 2015). On the other hand, the dip-coating method showed similar toluene adsorption behavior and coverage as SD 30%RH but slightly higher photocatalytic performance. This is assigned to the formation of a thick layer of TiO₂ NPs during the dip-coating process. Similarly, as for the SD method, the thick agglomerates were located on the top of filter fibers, which allowed higher adsorption of the pollutant but limited the photoabsorption and efficient activation. Although the photoabsorption was relatively high due to the low reflection, the photocatalytic activity was reduced in comparison to SA and SG methods owing to the loss of the incident light due to the transmission through the uncoated parts of the filter (Supplementary Figure S10).

For the practical implementation of photocatalytically active filters, the choice of the photocatalyst immobilization procedure would affect not only the removal performance of toluene but also the operational and mechanical properties, such as permeability of the filter (and pressure drop) and mechanical stability of TiO₂ coating. Therefore, we propose a simple evaluation of the overall PCO filter performance using a parameter entitled “coating quality factor”, which is based on the application of a broadly used concept to assess the filtration efficiency (Wang et al., 2008). This evaluation combines parameters, such as the reaction constant k (min⁻¹), the permeability of filter K (m²), and mass loss of coatings (g m⁻²) as follows: " C o a t i n g   Q u a l i t y   F a c t o r " = k . K m l o s s ( m 4 . g − 1 . m i n − 1 ) . (6)

It provides a quantitative comparison of PCO filters including reactivity and operational properties. The evaluation of pressure drop (Figure 11A) was performed at a standard value of 5 cm s⁻¹ for filtration studies, which exceeded significantly the filtration velocities used for the PCO tests. It, nevertheless, provides comparable information about the permeability K of the airflow through the filters tested at the same velocity. The stability of coating (Figure 11B) (i.e., loss of coating under applied stress m_loss) provides another important information for the practical implementation, which is the adhesion of photocatalyst particles onto the filter tested under ultrasonic stress. The SD methods demonstrated both large pressure drop and poor stability of TiO₂ coating. This was caused by the formation of “a cake” between and on the top of filter fibers. On the other hand, SA and SG methods displayed low-pressure drop and high stability of coating (Figures 11B and C) as TiO₂ nanoparticles formed a continuous layer wrapping around each individual fiber. The thinner coating layer provided better adhesion strength than the SD method. The van der Waals attraction forces between the TiO₂ agglomerates and GF substrate showed better adhesion strength and facilitated the stability of SA coating, while these forces were weaker in the presence of the large and round agglomerates produced by the SD method (Walsh et al., 2012; Gardon and Guilemany, 2014). As a result, the obtained “coating quality factor” values differ significantly for SA and SG conditions in comparison to the SD method (Figure 11C). The microstructure and macrostructure formed during the deposition using the dispersion of nanoparticle suspension through an atomization process (SA and SG) demonstrated both enhanced PCO reactivity and improved operational performance. The best overall performance was obtained for the SG method due to the larger droplet size and higher deposition speed. The benchmark dip-coating method demonstrated comparable performance with SA and SG methods. Even though the SA method displayed slightly lower overall performance than SG and dip-coating methods, this method has a potential for further optimization and improvement due to its advantages. First, the deposition conditions can be controlled efficiently (resulting in lower deviations in the performances). Second, it can be operated as a continuous deposition process resulting in better control of mass loading, that is, eliminating the need for the multiple immobilization and drying steps. Despite the high photocatalytic performance of TiO₂ photocatalyst, its efficiency has been reduced in the cycle tests due to the photocatalyst poisoning effect (Supplementary Figure S11) (Weon et al., 2018). Nevertheless, several strategies have been employed to minimize the damage, for example, catalyst doping and the control of catalyst properties, process, and regeneration conditions (Wu et al., 2021). Hence, these strategies could be integrated with SA and/or SG deposition methods in future studies. Additionally, these deposition methods provide many possibilities for the adjustment of deposition conditions, such as the concentration and pH of TiO₂ suspension (Wang et al., 2005; Sen et al., 2009; Lee et al., 2010) and/or the addition of surfactants (Veronovski et al., 2010). These adjustments would enable further investigation of the agglomeration and deposition properties and their effect on the photocatalytic performance. Furthermore, different types of nozzles could be selected and evaluated (Lee et al., 2011).