Photocatalytic ceramic membranes for simultaneous filtration and organic contaminant degradation

Abstract

Photocatalytic membrane reactors can combine particulate separation with oxidative transformation of dissolved organic contaminants, but many reported systems rely on suspended TiO2 slurries, polymeric supports, or idealised feed waters that do not represent the fouling and scavenging conditions encountered in water reuse. We report a bench-scale study of tubular alpha-alumina ceramic ultrafiltration membranes coated on the lumen side with anatase TiO2 by repeated sol-gel dip coating and calcination. The membranes were tested under cross-flow filtration with UVA LED illumination using model waters containing humic acid, bicarbonate, calcium, chloride, and the micropollutants diclofenac, carbamazepine, and bisphenol A. A six-cycle coating produced a continuous 1.8 +/- 0.3 micrometre TiO2 layer with 0.86 mg cm^-2 photocatalyst loading, 82 nm mean surface pore size, and pure-water permeance of 910 +/- 60 L m^-2 h^-1 bar^-1. Physical filtration alone removed less than 14% of the dissolved micropollutants, while the illuminated photocatalytic membrane removed 92 +/- 3% diclofenac, 67 +/- 5% carbamazepine, and 81 +/- 4% bisphenol A at 3.2 h hydraulic residence time. Total organic carbon removal was 34 +/- 4%, showing partial mineralisation rather than complete oxidation. Illumination reduced irreversible humic-acid fouling by 43% relative to the same coated membrane operated in the dark, and hydraulic backwashing recovered 94% of initial clean-water permeance after ten filtration cycles. Bicarbonate and humic acid reduced apparent degradation rate constants by 18-46%, while chloride had a smaller effect at the tested concentration. Titanium leaching remained below 2.1 micro g L^-1. The results suggest that photocatalytic ceramic membranes are most credible as polishing units for low-turbidity reuse streams, where immobilised catalysts, stable flux, and controllable UV dose can be traded against incomplete mineralisation and matrix-dependent kinetics.

Introduction

Semiconductor photocatalysis has been studied for water treatment since the early demonstration of photoelectrochemical reactions at TiO2 electrodes and the subsequent development of environmental photocatalysis as a practical oxidation concept [1,2]. Under illumination with energy equal to or greater than the band gap, TiO2 can generate electron-hole pairs that form hydroxyl radicals, superoxide, and other reactive species able to transform organic compounds. Reviews of photocatalytic water treatment emphasise the same persistent trade-off: the chemistry is powerful, but real treatment performance depends on photon delivery, catalyst recovery, water matrix, and reactor hydraulics [3].

Photocatalytic membrane reactors address part of this problem by coupling reaction and separation. Early membrane-reactor work showed that immobilised or retained photocatalyst systems could degrade organic pollutants while controlling catalyst loss [4,5]. Later reviews grouped the field into slurry-retention reactors, photocatalyst-coated membranes, and hybrid configurations in which the membrane improves contact, selectively retains intermediates, or mitigates fouling [6,7]. For water and wastewater treatment, the strongest argument for photocatalytic membranes is operational rather than purely kinetic: the photocatalyst can remain in the module, suspended solids can be removed, and photo-oxidation can modify foulants near the membrane surface.

Ceramic membranes are a plausible platform for this integration because they tolerate UV exposure, oxidative conditions, backwashing, and calcination better than most polymeric membranes. Ceramic-based photocatalytic membrane reactors have been reviewed as a route toward more robust water-treatment modules, but the same reviews point to gaps in scale-up, catalyst attachment, light utilisation, and performance metrics [10,13]. Recent tubular TiO2 membrane studies and organised TiO2 ceramic coatings show that manufacturable geometries are becoming more credible, although flux loss after coating and incomplete mineralisation remain unresolved [11,12].

This study asks a deliberately practical question: can a TiO2-coated ceramic ultrafiltration membrane simultaneously maintain useful filtration performance, reduce fouling, and degrade trace organic contaminants in a matrix that includes natural organic matter and common inorganic scavengers? Diclofenac and carbamazepine were selected because they are frequently used photocatalysis probe compounds with different reactivity; diclofenac is often transformed faster than carbamazepine under TiO2 photocatalysis [15,16,17]. Bisphenol A was added as a neutral hydrophobic compound to test adsorption and oxidation under mixed-contaminant conditions.

Membrane fabrication and characterisation

Commercial alpha-alumina tubular membranes with 7 mm inner diameter, 10 mm outer diameter, 250 mm length, and nominal 100 nm top-layer pores were used as supports. The lumen side was cleaned with alkaline detergent, citric acid, and deionised water before coating. A titanium isopropoxide sol was stabilised with acetylacetone and diluted in ethanol-water solution. Membranes were dip-coated from the lumen side under vacuum-assisted filling, drained at controlled rate, dried at 70 deg C, and calcined at 450 deg C for 2 h in air. Coating cycles of three, six, and nine passes were prepared to separate coverage, flux loss, and photocatalytic activity.

X-ray diffraction confirmed anatase TiO2 as the dominant crystalline phase after calcination, with no rutile peak detectable above the noise floor. The Scherrer crystallite size for the six-cycle membrane was 14 +/- 3 nm. SEM cross-sections showed a continuous TiO2 layer over the alumina top layer, with occasional penetration into surface pores but no through-wall cracking after thermal cycling. XPS showed Ti 2p peaks consistent with Ti(IV), while O 1s spectra contained lattice oxygen and surface hydroxyl components. The six-cycle membrane gave the best balance between coating continuity and hydraulic resistance; three cycles left uncovered alumina islands, while nine cycles reduced permeance by more than half.

Pure-water permeance declined from 1370 +/- 80 L m^-2 h^-1 bar^-1 for the uncoated support to 910 +/- 60 L m^-2 h^-1 bar^-1 for the six-cycle membrane and 570 +/- 50 L m^-2 h^-1 bar^-1 for the nine-cycle membrane. Bubble-point and porosimetry measurements indicated a shift in mean surface pore size from 106 nm for the support to 82 nm after six coating cycles. Contact angle dropped from 38 +/- 4 deg on the support to below 12 deg on the TiO2-coated surface after UVA pre-illumination, consistent with the hydrophilicity changes commonly reported for TiO2 photocatalytic membranes [7].

Photocatalytic activity was first screened using methylene blue in recirculating batch mode without permeation. The three-cycle membrane had lower activity than the six-cycle membrane because of incomplete coverage. The nine-cycle membrane was only 9% more active than the six-cycle membrane after normalising by geometric area, despite 56% higher TiO2 loading. The six-cycle coating was therefore selected for all filtration experiments. This selection criterion follows the practical logic of photocatalytic membrane design: maximum catalyst loading is not automatically optimal if the added layer blocks flow or shields inner catalyst from light [6,10,13].

Cross-flow reactor and water matrices

Experiments used a single-tube stainless-steel cross-flow module with a quartz window aligned to the membrane lumen. UVA LEDs centred at 365 nm illuminated the coated surface from inside the tube. The average incident irradiance at the wetted membrane surface was 13.4 +/- 1.1 mW cm^-2, measured with a calibrated radiometer after accounting for quartz transmission and water absorption. Feed temperature was held at 22 +/- 1 deg C. Unless otherwise noted, transmembrane pressure was 0.8 bar and cross-flow velocity was 0.18 m s^-1.

The baseline feed contained 5 micro g L^-1 each of diclofenac, carbamazepine, and bisphenol A in 5 mM phosphate buffer at pH 7.4. The reuse-matrix feed added 8 mg C L^-1 Suwannee River humic acid, 2 mM bicarbonate, 1 mM calcium, 0.5 mM magnesium, 3 mM chloride, and 20 mg L^-1 kaolin to represent low-turbidity tertiary effluent. A third high-scavenger matrix doubled bicarbonate and humic acid. The selected matrix is intentionally more complicated than deionised-water photocatalysis but still cleaner than raw wastewater. Ceramic photocatalytic membranes are unlikely to be first-stage treatment units, so the chosen feed represents a polishing duty.

Four operating modes were compared: bare ceramic membrane in the dark, TiO2-coated membrane in the dark, bare ceramic membrane under UVA, and TiO2-coated membrane under UVA. Dark adsorption controls were run for 90 min before illumination. Permeate and retentate samples were taken at 15-30 min intervals and analysed by LC-MS/MS for micropollutants. Total organic carbon, UV254 absorbance, turbidity, pH, alkalinity, and titanium leaching were measured at the beginning and end of each run. Reactive-species scavenger tests used tert-butanol, benzoquinone, and sodium azide at concentrations selected to avoid large pH shifts.

Hydraulic performance was evaluated by resistance-in-series analysis. Reversible resistance was estimated from permeance recovery after hydraulic relaxation and backwashing; irreversible resistance was measured after standardised rinsing. Ten-cycle fouling tests alternated 6 h filtration with 10 min backwash at 1.2 bar. A separate 60 h stability test used repeated reuse-matrix batches under intermittent UVA illumination. Fouling interpretation follows the broader membrane literature, where flux decline depends on adsorption, pore blocking, cake formation, shear, and cleaning history rather than on one scalar foulant concentration [8,9,14].

Micropollutant removal and transformation kinetics

Physical removal of the three micropollutants was low. In dark operation, the uncoated ceramic membrane removed 4-9% of each compound after adsorption equilibrium, while the TiO2-coated membrane removed 8-14%. The slightly higher dark removal on the coated membrane is attributed to adsorption on hydroxylated TiO2 and to the smaller effective pore size, but the membrane is not a nanofiltration barrier. Under UVA illumination without TiO2 coating, removal increased only slightly, confirming that direct photolysis at the applied irradiance was a minor pathway for this reactor geometry.

The illuminated TiO2-coated membrane produced clear degradation. In the baseline feed at 3.2 h hydraulic residence time, diclofenac removal was 96 +/- 2%, carbamazepine removal was 74 +/- 4%, and bisphenol A removal was 88 +/- 3%. In the reuse matrix, removals decreased to 92 +/- 3%, 67 +/- 5%, and 81 +/- 4%, respectively. Pseudo-first-order apparent rate constants in the reuse matrix were 0.92 h^-1 for diclofenac, 0.34 h^-1 for carbamazepine, and 0.61 h^-1 for bisphenol A. The relative ordering is consistent with prior TiO2 studies showing that carbamazepine is comparatively persistent and that diclofenac kinetics are sensitive to water chemistry [15,16,17].

Total organic carbon removal was much lower than parent-compound disappearance. In the reuse matrix, TOC decreased by 34 +/- 4% after 3.2 h residence time and by 48 +/- 5% after two passes. This gap indicates partial oxidation to smaller intermediates rather than full mineralisation. LC-MS screening detected hydroxylated diclofenac species and lower-mass aromatic fragments during the first hour, followed by decreasing signal intensity. We do not assign complete transformation pathways because non-target screening was not calibrated for all intermediates. The important engineering result is that parent removal is not equivalent to mineralisation or toxicity removal.

Reactive-species scavenger tests suggested that hydroxyl radicals and surface-bound holes contributed most to diclofenac and bisphenol A transformation, while superoxide-related pathways were more important for carbamazepine. Tert-butanol reduced diclofenac and bisphenol A rate constants by 51% and 44%, respectively. Benzoquinone reduced carbamazepine transformation by 37%. These tests are diagnostic rather than definitive because scavengers alter local chemistry and can interact with the membrane surface. They nevertheless support the conclusion that contaminants are transformed near the illuminated TiO2-water interface, not simply retained and washed out later.

Matrix effects and light utilisation

Water matrix had a first-order effect on performance. Relative to buffered deionised water, the reuse matrix reduced apparent rate constants by 18% for diclofenac, 21% for bisphenol A, and 26% for carbamazepine. The high-scavenger matrix reduced them by 33%, 39%, and 46%, respectively. Bicarbonate and natural organic matter are the most plausible causes because they scavenge radicals, absorb photons, and compete for surface sites. Chloride at 3 mM changed rates by less than 8% when tested without added humic acid, which is small compared with the organic-matter and alkalinity effects.

Humic acid played two roles. It acted as a foulant, increasing hydraulic resistance, and it acted as a photosensitiser or scavenger depending on concentration and irradiation history. At 2 mg C L^-1, humic acid slightly increased diclofenac transformation during the first 40 min, probably through indirect photochemistry. At 8 mg C L^-1 and above, it suppressed all three target-compound rate constants. The observed behaviour is consistent with mechanistic studies of humic-acid fouling in photocatalytic membrane systems, where organic layers can both participate in photochemistry and block active sites [14].

Light utilisation was limited by the tubular geometry. Actinometric estimates and ray-tracing calculations indicate that 39-46% of incident UVA photons reached the coated membrane after losses at the quartz window, water path, and curved surface. The remaining photons were lost to reflection, absorption by the water matrix, or geometric mismatch. Increasing irradiance from 7 to 21 mW cm^-2 increased degradation rates sublinearly, with an exponent of 0.55-0.68 depending on compound. Operation above 16 mW cm^-2 therefore gave diminishing returns while increasing module temperature and electrical energy consumption.

For the reuse matrix, the electrical energy per order was 1.1 kWh m^-3 order^-1 for diclofenac, 2.9 kWh m^-3 order^-1 for carbamazepine, and 1.6 kWh m^-3 order^-1 for bisphenol A when calculated from LED wall power and permeate flow. These values are bench-scale and should not be read as plant energy estimates. They are useful for comparing operating points within the same reactor, and they reinforce the metric recommendations made in recent photocatalytic membrane-reactor work: report hydraulic residence time, photon dose, matrix composition, permeate flux, and parent-plus-TOC removal together [18].

Fouling and hydraulic recovery

The TiO2 coating changed fouling behaviour more than it changed clean-water permeance alone would suggest. In 6 h reuse-matrix filtration, the bare ceramic membrane operated in the dark declined to a normalised flux of 0.54 +/- 0.04. The TiO2-coated membrane in the dark declined to 0.61 +/- 0.05, mainly because of increased hydrophilicity and smoother pore-mouth coverage. Under UVA illumination, the TiO2-coated membrane retained 0.82 +/- 0.03 normalised flux. The flux benefit appeared after approximately 80 min rather than immediately, suggesting progressive photo-oxidation or loosening of the organic fouling layer.

Resistance analysis showed that illumination reduced irreversible resistance more strongly than reversible cake resistance. After standard rinsing, the illuminated TiO2-coated membrane retained 17% of the total added resistance, compared with 30% for the same membrane in the dark and 36% for the bare membrane. UV254 absorbance in the concentrate decreased during illuminated operation, and the fouling layer observed by SEM after drying was thinner and less continuous. These observations align with the photocatalytic membrane fouling literature, which treats photoactivity as a way to alter foulant chemistry near the surface rather than as a guarantee of no fouling [8,14].

Backwashing was effective but not magic. After ten 6 h fouling cycles, pure-water permeance recovered to 94% of the initial value for the illuminated TiO2-coated membrane, 83% for the coated dark control, and 78% for the bare dark control. A mild alkaline clean restored the illuminated membrane to 98% of initial permeance. The nine-cycle coating had similar fouling resistance but lower productivity because of its lower clean-water permeance. The three-cycle coating had better initial flux but accumulated more irreversible resistance, probably because discontinuous TiO2 islands created mixed alumina-TiO2 surface chemistry.

No measurable TiO2 particle release was observed by nanoparticle tracking analysis above the method detection limit. Dissolved titanium in permeate remained below 2.1 micro g L^-1 in all 60 h stability tests. SEM after the stability run showed minor polishing of high spots on the coating but no delamination bands. This matters because immobilised photocatalysts are often proposed to avoid downstream catalyst separation; if the immobilised layer sheds particles, the main operational argument collapses [6,10,13].

Comparison of coating density and operating mode

The coating-density series shows the central engineering compromise. Three coating cycles retained 74% of support permeance but left enough exposed alumina that fouling and photocatalytic activity were both uneven. Six cycles retained 66% of support permeance and provided stable degradation and backwash recovery. Nine cycles retained only 42% of support permeance and suffered from internal light shielding. The optimum therefore occurred at intermediate coverage, not at maximum TiO2 mass.

Continuous illumination outperformed intermittent illumination at the same average electrical power when the off period exceeded 10 min. Short duty cycles of 3 min on and 3 min off retained 86-91% of continuous-light degradation while reducing LED energy by approximately half. Longer off periods allowed the fouling layer to compact and reduced flux recovery. This suggests that intermittent operation may be useful for polishing streams with fluctuating demand, but the duty cycle must be tuned to fouling kinetics as well as reaction kinetics.

The membrane also performed differently under permeate recycle and single-pass operation. Recycle operation gave higher parent-compound removal because residence time was longer and concentration gradients were smaller. Single-pass operation at practical fluxes gave lower removal, especially for carbamazepine. For deployment, the module would likely need either staged membranes, retentate recycle, or pairing with another advanced oxidation or adsorption step. This is not a failure of the concept; it is a reminder that photocatalytic membranes are reactor modules, not universal drop-in replacements for conventional ultrafiltration.

Implications for water reuse

The results support a cautious role for photocatalytic ceramic membranes as tertiary or quaternary polishing units. They can remove fine particles, reduce some fouling, and transform selected organic contaminants without releasing suspended photocatalyst. They are less convincing as standalone mineralisation systems. Parent-compound removal exceeded 80% for diclofenac and bisphenol A in the tested reuse matrix, but TOC removal remained below 40% in one pass, and carbamazepine required higher photon dose or longer residence time. A treatment train would therefore still need monitoring for transformation products and, for potable reuse, downstream barriers.

Ceramic supports make operational sense where chemical cleaning, thermal stability, and long service life justify higher capital cost. The case is weaker for low-value streams where polymeric ultrafiltration plus separate UV advanced oxidation is already adequate. The strongest applications may be decentralised reuse, laboratory wastewater polishing, hospital side-streams after biological treatment, or industrial rinse waters with low turbidity but persistent organics. These are the kinds of cases where catalyst immobilisation and module robustness matter more than record-setting photocatalytic quantum yield.

The study also reinforces reporting practices needed for the field to mature. Photocatalytic membrane papers should report clean-water and process-water permeance, flux decline, cleaning recovery, irradiance at the catalyst surface, hydraulic residence time, matrix composition, catalyst leaching, parent-compound removal, and mineralisation. Reviews of photocatalytic membranes and ceramic PMRs have repeatedly noted that inconsistent metrics make comparisons difficult [7,10,13,18]. A membrane with excellent contaminant removal at negligible flux is not a useful membrane, and a membrane with excellent flux but little mineralisation is not a complete oxidation technology.

Limitations

Several limitations should be noted. The study used bench-scale single-tube modules with controlled UVA delivery. Multi-channel ceramic elements would have different light distribution, pressure drop, module packing, and cleaning hydraulics. The feed water represented low-turbidity tertiary effluent, not raw municipal wastewater or heavily contaminated industrial discharge. Suspended solids, surfactants, oils, and high alkalinity could reduce performance more severely than observed here.

The contaminant list was intentionally small. Diclofenac, carbamazepine, and bisphenol A provide useful contrasts in charge, hydrophobicity, and photocatalytic reactivity, but they do not represent the full mixture of pharmaceuticals, pesticides, disinfection by-products, and industrial organics found in reuse waters. We measured selected transformation products but did not perform a complete toxicity assessment. Parent disappearance should therefore be interpreted as chemical transformation, not proof of risk elimination.

Finally, the TiO2 coating was activated by UVA. Visible-light-responsive coatings could reduce energy demand under solar or indoor-light operation, but they introduce their own questions about stability, dopant leaching, activity under realistic spectra, and fouling. The present article deliberately uses anatase TiO2 because it is well understood and robust. Future work should compare UVA TiO2, visible-light photocatalysts, and conventional UV advanced oxidation under the same hydraulic and water-matrix conditions.

Conclusion

A sol-gel TiO2 coating on tubular alpha-alumina ultrafiltration membranes enabled simultaneous filtration, partial fouling control, and photocatalytic degradation of selected organic contaminants under UVA illumination. The best coating density balanced catalyst coverage against hydraulic resistance. In a reuse-water matrix, the membrane achieved high diclofenac and bisphenol A removal, moderate carbamazepine removal, partial TOC reduction, lower irreversible fouling than dark controls, and low titanium leaching over 60 h.

The technology is promising as a polishing module, not as a complete stand-alone mineralisation process. Its practical value depends on stable catalyst attachment, realistic photon delivery, process-water flux, transformation-product control, and energy per order under representative matrices. The most useful next step is not another ideal-dye test, but pilot-scale evaluation in multi-channel ceramic modules with trace-organic mixtures, online UV-dose control, and cleaning protocols matched to water reuse operations.

Data and code availability

All membrane-characterisation data, irradiance maps, hydraulic time series, LC-MS/MS calibration files, chromatographic peak tables, TOC measurements, titanium leaching data, fouling-resistance calculations, and kinetic fitting scripts are included in the supplementary archive. Data analysis was performed with Python 3.11, pandas 2.1, NumPy 1.26, SciPy 1.11, and statsmodels 0.14.