A stage-based framework for closed-loop recycling of polymeric composite membranes

Abstract

Polymeric composite membranes are essential in modern separation technologies but pose critical challenges for sustainable end-of-life (EOL) management due to their chemically heterogeneous, multilayer structures. Existing approaches, such as mechanical processing or catalytic depolymerization, exhibit limited efficiency in separating and recovering multiple components. Here, we present a stage-based physical-chemical recycling strategy enabling multicomponent recovery and closed-loop regeneration of composite membranes through sequential solvent extraction, catalytic depolymerization, and membrane recasting. Using real EOL membranes as feedstock, the process yields regenerated membranes with mechanical robustness and separation performance comparable to commercial counterparts. Life-cycle assessment and technoeconomic analysis reveal substantial reductions in environmental impacts and material costs relative to conventional membrane manufacturing. By engineering a scalable process system for the circular recovery of complex membrane components, this work bridges materials science and process engineering, offering a viable route for sustainable management of polymeric composite waste streams.

A recycling strategy enables closed-loop regeneration of polymeric membranes with reduced environmental and material costs.

INTRODUCTION

Polymeric composite membranes are cornerstones of modern separation technologies, with widespread applications in product purification (1–5), electrochemical systems (6–8), and biomedical devices (9, 10). These membranes typically consist of a mechanically robust support and a selective separation layer, enabling independent optimization of structural integrity and separation performance. A prominent example is the thin-film composite (TFC) membranes used for reverse osmosis (RO) desalination, comprising an ultrathin polyamide (PA) selective layer (typically 100 to 300 nm) formed via interfacial polymerization on a microporous polysulfone (PSf) support reinforced with polyethylene terephthalate (PET) nonwoven fabric. The cross-linked PA layer provides precise molecular separation [>99% sodium chloride (NaCl) rejection], while the PSf-PET substrate maintains structural integrity under hydraulic pressures up to 80 bar during desalination operation.

However, the increasing deployment of composite membranes has raised environmental concerns regarding their end-of-life (EOL) management. For instance, in water treatment alone, more than 8.56 million membrane modules—including RO, nanofiltration (NF), ultrafiltration (UF), and microfiltration (MF)—are deployed globally in desalination and wastewater treatment facilities (11) (Fig. 1 and table S1), corresponding to 318 million m² of membrane area. With projected doubling of this figure by 2030 and typical life spans of 3 to 5 years (12), EOL membrane disposal presents an urgent sustainability priority. Landfilling and incineration are two common practices for EOL membrane management. Landfilling likely contributes to the generation of microplastics and the release of persistent organic pollutants (13, 14), while incineration requires substantial energy input and poses the risk of producing highly toxic by-products (15). Moreover, complex residuals—such as bacteria, viruses, and micropollutants—have led many counties or regions to classify EOL membranes as hazardous waste (16), thereby substantially increasing disposal costs and regulatory limitations. Compounding this issue is the inherent heterogeneity of composite membranes (e.g., PA-PSf-PET mixture in TFC membranes), which limits the efficiency of conventional recycling techniques and underscores the need for scalable, low-impact recovery strategies.

Fig. 1. Global potential for recycling composite membranes.

The global development of composite membranes in seawater desalination and sewage treatment is largely observed in regions with water scarcity issues. Key regions include Asia, North America, and Oceania—particularly countries like Saudi Arabia, the United Arab Emirates, China, the United States, and Singapore—which together account for ~50% of global membrane usage.

Recent advances in polymer recycling have largely focused on single-component plastic systems, such as PET (17–26), polyethylene (PE) (27–37), and polyvinyl chloride (PVC) (38–41), using physical and chemical approaches. Physical approaches—via shredding, melting, and reprocessing—are effective for pure polymers but often fail in composite systems because of immiscibility. Chemical strategies, such as solvolysis and pyrolysis, have enabled closed-loop recycling and upcycling. For instance, solvolysis of PET using metal-based catalysts achieves efficient monomer recovery (>90%) for repolymerization (26). Nevertheless, these approaches still face fundamental limitations in handling multicomponent systems. First, thermodynamic incompatibility among polymers degrades product quality in mechanical recycling (42). Second, complex feedstock likely deactivates catalysts—e.g., in PSf-PET blends, sulfur compounds generated from PSf decomposition can poison metal catalysts designed for PET recycling (43). Third, high energy demand combined with the low yield of high-value products impose barriers to the viability and implementation of pyrolysis-based processes (44).

Notably, recent progress in chemical recycling have shown growing interest in the management of binary polymer systems, such as PET/PVC (45), polypropylene (PP)/PVC (46), and polystyrene/PVC (47). One notable study used chlorine released from PVC to drive PET depolymerization, achieving 98% material recovery via a tandem dechlorination-depolymerization (44). However, these binary systems, largely composed of pure polymer blends, are far less complex than real-world polymeric composites, which contain chemically distinct layers, foulants, additives, and fillers. Recent efforts have also exploited solubility differences to recover polymeric components in different dissolution stages (48), yet polymers like PA and PET remain poorly soluble under mild conditions (49). For example, PET exhibits a solubility of only 13.3 wt % in dimethyl sulfoxide (DMSO) at 135°C, while PA solubility is even lower (3.1 wt %) (49), making selective separation inefficient. Another approach—chemical cleaning—leverages strong oxidants (e.g., chlorine) to remove contaminants or degrade the selective layer, thus enabling partial reuse of composite membranes (50) (fig. S1). This strategy has enabled the extension of life span of water treatment membranes. However, its effectiveness strongly relied on the structural integrity of the membrane, limiting its applications to EOL membranes that have undergone severe chemical aging or physical degradation—conditions frequently encountered after prolonged industrial use. Moreover, repeated exposure to oxidative agents may accelerate structural and chemical degradation. Although these treatments may temporarily extend membrane functionality—typically by 1 to 2 years (51)—the underlying degradation processes are irreversible. Consequently, the membranes ultimately fail to meet performance requirements and are relegated to EOL disposal routes such as incineration or landfill, incurring both material loss and environmental burden.

In this study, we demonstrate a stage-based strategy that integrates physical (solvent dissolution) and chemical (catalytic depolymerization) processes for efficient recovery of polymeric composite membranes. Solvent selection, guided by Hansen solubility parameters (HSPs) while also accounting for interfacial adhesion, and fouling constraints, enables precise dissolution of specific soluble layers, such as PSf and polyvinylidene difluoride (PVDF). Moreover, catalyst-activated depolymerization facilitates the conversion of poorly soluble components, notably PET, into reusable monomers under realistic and contaminant-rich conditions. The recovered constituents are subsequently reprocessed into new composite membranes with comparable separation performance and mechanical strength, thereby demonstrating functional closed-loop recycling. Mechanistic insights are obtained through in situ Fourier transform infrared spectroscopy and mass spectrometry, elucidating the depolymerization pathways of PET substrates and the influence of residual contaminants on recycling efficiency. Application of the strategy to real EOL RO membrane modules confirms its practical scalability. Furthermore, life cycle assessment (LCA) and technoeconomic analysis (TEA) reveal substantial reductions in both environmental impact and material cost relative to conventional membrane manufacturing. This study establishes a framework that combines targeted separation, contaminant-tolerant depolymerization, and functional remanufacturing into a cohesive and scalable process, offering a robust and versatile platform for the sustainable management of structurally complex, composite polymeric wastes.

RESULTS

Development and initial study for stage-based recycling composite membrane

We developed a stage-based strategy that enables closed-loop recycling of polymeric composite membranes. Taking RO membrane as an example (Fig. 2A, left), this TFC membrane consists of a PET substrate (79 wt %), a PSf support layer (20 wt %), and a PA selective layer (1 wt %). Its corresponding recycling process is divided into five distinct stages (Fig. 2A, right, and fig. S2), each targeting specific membrane components: (I) Solvent extraction: The PSf layer from the EOL membrane is recovered through solvent dissolution, producing a PSf solution for membrane casting. Because of the poor solubility in N,N′-dimethylformamide (DMF), the PA selective layer, along with the PET substrate, remains undissolved and is physically retained as a solid residue. (II) Depolymerization: The undissolved solid mixture is processed through glycolysis, yielding bis(hydroxyethyl) terephthalate (BHET) monomers from PET. The PA layer remains largely inert under these conditions and is subsequently removed during product purification process. (III) Repolymerization: The recycled BHET is subsequently subjected to polycondensation to produce recycled PET, which is then used to fabricate a new substrate through spinning. (IV) Nonsolvent-induced phase separation (NIPS): The casting solution from step I, together with the recycled PET substrate from step III, are used to prepare UF membranes via NIPS. This process involves film casting, phase separation, solidification, and washing, ensuring the production of UF support layer with robust structural and functional properties. (V) Interfacial polymerization: A PA selective layer is formed on top of the UF support through interfacial polymerization, resulting in high-performance RO membranes. This process involves monomer impregnation, interfacial polymerization, washing, and curing to achieve TFC membranes with desired performance.

Fig. 2. Stage-based recycling process and performance for composite membranes and films.

(A) Schematic diagram of the stage-based strategy using EOL RO membrane closed-loop recycling as an example. (B) Recycling performance for composite membranes and films. The process involves initial solvent extraction (middle), followed by catalytic PET depolymerization to yield BHET monomer (right). The glass bottles contain casting solutions obtained by dissolving polymeric components, with adjacent poorly soluble PET substrates retained after extraction. Depolymerization of these PET substrates yielded BHET with >90% efficiency under optimal conditions (175°C, 40 min, ethylene glycol/PET mass ratio of 4:1, and 1 wt % of Mo-MgO catalyst). PC, polycarbonate; EVA, ethylene-vinyl acetate.

To comprehensively evaluate the effectiveness and robustness of the proposed stage-based framework, we conducted recycling experiments on a broad range of purification membranes and composite films used in energy storage, flexible electronics, agricultural insulation, and food packaging (Fig. 2B). These materials included RO (BW30), NF (NF90 and NF270), UF [PSf and polyacrylonitrile (PAN)], MF (PVDF) membranes, as well as PE-PET, polycarbonate (PC)–PET, ethylene-vinyl acetate (EVA)–PET, and PVC-PET composite films (specific compositions of these materials are provided in table S2). During the solvent extraction step, the PSf, PVDF, and PAN layers of the purification membranes were completely dissolved in DMF at room temperature within 5 min. For the composite films, polymer-specific solvents were selected on the basis of their chemical compatibility to recover the soluble components: PE was selectively removed using toluene at 110°C for 4 hours; EVA and PVC were dissolved in tetrahydrofuran (THF) at 60° and 50°C for 20 min, respectively; and PC was efficiently solubilized in dichloromethane (DCM) at 40°C for 10 min.

Following the solvent extraction process, the remaining materials underwent catalytic depolymerization using ethylene glycol, yielding the BHET monomer. The structure of the BHET product was confirmed through by nuclear magnetic resonance analysis (fig. S3). Optimization of reaction conditions resulted in a BHET yield >90% (fig. S4 and table S3). All the polymeric composites exhibited BHET yields exceeding 90% (fig. S5 and table S4). Moreover, our Mo-MgO catalyst demonstrated a space-time yield (STY) of 184.4 g_BHET g_cat⁻¹ hour⁻¹, surpassing most catalysts reported in the literature (52) (fig. S6 and table S5). For context, representative glycolysis catalysts exhibit STYs that mostly range from 1 to 130 g_BHET g_cat⁻¹ hour⁻¹. Notably, the process maintained high efficiency when scaled up to module-scale RO membranes (100 g, ~1 m²). The solvent extraction process achieved complete recovery of PSf, while the depolymerization effectively converted ~80 g of PET substrate to 97 g of BHET (fig. S7 and table S3). The Mo-MgO catalyst remains reusable under EOL membrane–relevant conditions, maintaining a BHET yield above 80% over five consecutive cycles, and a simple calcination-based regeneration restores the catalytic performance to near its initial level (fig. S8). Structural characterization indicates that the Mo-MgO catalyst features abundant surface defects and strong basic sites, together with atomically dispersed Mo centers (figs. S8 to S12), supporting a cooperative interaction between the basic MgO domains and Mo active centers that drives its high catalytic activity and excellent reusability under contaminant-rich conditions. Overall, the stage-based recovery strategy achieved effective performance in recovering commercial purification membranes and composite films.

Selection of extraction solvents and elucidation of depolymerization mechanisms

The effectiveness of solvent extraction largely depends on selecting appropriate solvent systems to separate and recover specific layers within composite structures, such as the PSf layer in TFC RO membranes. HSPs play a pivotal role in this solvent extraction process, as they quantify interactions between solvents and materials through three independent solubility components: dispersion interactions (δ_D), polar interactions (δ_P), and hydrogen bonding interactions (δ_H) (53). On the basis of experimental data and HSP values, solvents such as DMF, DMSO, DCM, tetrahydrofuran (THF), and N-methylpyrrolidone were screened for their process effectiveness (Fig. 3A and table S6). Moreover, the applicability of HSP extends beyond PSf to other composite material layers, such as PVDF and PAN, further demonstrating its utility in designing solvent systems for material recycling (fig. S13).

Fig. 3. Methods for solvent system selection and depolymerization mechanisms.

(A) Prediction of solvents that selectively dissolve PSf using HSPs. (B) Schematic diagram of the synchrotron radiation facility and in situ high-temperature (HT) infrared (IR) spectrometry. (C to E) In situ HT-IR spectra to investigate the interaction between catalyst and solvent, and the catalytic depolymerization pathways of the model compound (Mode) in various solvents. EG, ethylene glycol; EG-d₄, deuterated ethylene glycol–d₄. (F to H) Mass spectra and corresponding chemical structures of depolymerization products of the Mode and PET substrate in deuterated and nondeuterated solvents. m/z, mass/charge ratio.

While HSP analysis provides a thermodynamic basis for solvent-polymer compatibility, the actual efficiency of selective dissolution in multilayer systems is further influenced by interfacial adhesion, fouling, and dissolution kinetics (fig. S14). Comparative extraction tests revealed that PSf layers in three-layer TFC membranes dissolved less efficiently than those in two-layer analogs, suggesting that strong interlayer adhesion can hinder solvent infiltration despite favorable HSP compatibility. Likely, PSf extraction from EOL membranes was markedly impeded compared with pristine counterparts, primarily because of surface-bound foulants that physically obstruct solvent access. Time-resolved dissolution experiments further revealed a three-stage behavior: an initial slow-release phase governed by interfacial diffusion resistance, a rapid dissolution phase as the polymer becomes solvent accessible, and a final plateau corresponding to solubilization completion. These findings underscore the importance of balancing thermodynamic compatibility (HSP) with kinetic accessibility and structural constraints to achieve efficient and selective dissolution in realistic composite systems.

To support the practical viability of our solvent-based separation process, we developed a multicriteria solvent selection framework that balances polymer selectivity, material cost, and environmental impact. While HSP compatibility provided the thermodynamic foundation for assessing dissolution efficiency, candidate solvents were further evaluated on the basis of toxicity (ecotoxicity and human toxicity), volatility (boiling point), economic feasibility (selling price), environmental impact [global warming potential (GWP)], and recovery potential (recoverability and unit recovery energy consumption). For instance, although DCM and THF demonstrated favorable solubility characteristics, their high volatility and environmental hazard profiles made them less suitable for large-scale use. In contrast, DMF combined excellent compatibility with PSf, lower toxicity, and higher recyclability, making it the most viable for scaled implementation (table S7).

The subsequent depolymerization stage involves complex interactions among catalysts, solvents, and substrates. We synthesized a dimer (i.e., 1,2-ethylene glycol dibenzoate), as a model compound (Mode) for elucidating reaction pathways (fig. S15). Using in situ high-temperature infrared spectroscopy (Fig. 3B), we observed a temperature-dependent increase in the intensity of characteristic peaks associated with ethylene glycol (Fig. 3C). Specifically, the vibration of v(C─OH) at 1084 and 1032 cm⁻¹, as well as the bending of β(C─O─H) at 1415 cm⁻¹, notably more pronounced as temperature increased, indicating the activation of ethylene glycol at elevated temperatures. Moreover, a new peak emerged at 1545 cm⁻¹, with its intensity increasing as temperature rose. Given that ethylene glycol lacks characteristic peaks in this spectral region and that metal oxide peaks typically occur below 600 cm⁻¹, we surmised that this peak arises from an interaction between hydroxyl groups and metal oxides. This interaction likely corresponds to the formation of a chemical bond analogous to O─H─Mg (Mo), reflecting a critical step in the catalytic activation process.

Depolymerization of Mode in ethylene glycol led to notable peak shifts associated with chemical transformations (Fig. 3D and fig. S16). The double peaks of v(C─O) at 1283 and 1261 cm⁻¹ merged into a single peak at 1261 cm⁻¹, while the double peaks of ω(─CH₂) at 1202 and 1124 cm⁻¹ merged to a single peak at 1140 cm⁻¹. In addition, the peak of ω(─CH₃) at 976 cm⁻¹ shifted to 968 cm⁻¹. These spectral changes indicate that the Mode molecule was converted into BHET monomer through the cleavage of the C─O bond in the ester group, followed by the incorporation of dehydrogenated ethylene glycol molecules. Simultaneously, the terminal groups combined with hydrogen atoms to form methanol following bond cleavage. Likewise, depolymerization of Mode in deuterated ethylene glycol–d₄ environment resulted in similar peak changes (Fig. 3E and fig. S17), verifying the activation of the deuterated reagent and the subsequent conversion of the Mode molecule into BHET monomer. These findings validate that the C─O bond in the ester group was the primary bond cleavage site during the glycolysis of the PET substrate.

We also applied isotope labeling assay to verify the bond cleavage site and the role of ethylene glycol in depolymerization. Liquid chromatography–orbitrap mass spectrometry identified the glycolysis products of the Mode and the PET substrate under real reaction conditions, using either ethylene glycol (Fig. 3F) or deuterated ethylene glycol–d₄ (10 vol % in ethylene glycol; Fig. 3H) as the solvent. Notably, in deuterated ethylene glycol–d₄ environment, the depolymerization products were identified as HOD₂CD₂CO(O)C–C₆H₄–C(O)CD₂CD₂OH, corresponding to a signal at mass/charge ratio (m/z) = 263/285 (Fig. 3G and fig. S18). In comparison, when using ethylene glycol as the solvent, the products were HOH₂CH₂CO(O)C–C₆H₄–C(O)CH₂CH₂OH, with a corresponding signal at m/z = 255/277. These findings demonstrate that the C─O bond in the ester group is the bond cleavage site in the depolymerization of the Mode compound and the PET substrate. Hence, the formation of BHET monomer formation involves the attachment of ethylene glycol molecules, coupled with the removal of one hydrogen from the hydroxyl group.

Together, the depolymerization of PET substrate begins with the activation of ethylene glycol by the catalyst. Upon activation, the lone pair of electrons on the oxygen atom of ethylene glycol attacks the carbonyl carbon of PET, leading to the cleavage of the ester C─O bond. This reaction facilitates the binding of ethylene glycol molecules to the carbonyl carbon while one hydrogen atom is removed from the hydroxyl group. Consequently, the substrate undergoes progressive depolymerization, breaking down sequentially into oligomers, dimers, and ultimately into BHET monomers.

Stage-based recycling of real EOL composite membranes and modules

After revealing the key mechanisms for the solvent selection and the depolymerization pathways, we further evaluated the practical applicability of the stage-based strategy for recycling EOL membranes (Fig. 4A) through introducing typical foulants in membrane applications. The foulants included bovine serum albumin (BSA), sodium alginate (SA), humic acid (HA), and gypsum, corresponding to proteins, polysaccharides, humic substances, and mineral foulants observed on EOL membranes from desalination and wastewater treatment facilities (54) (fig. S19). These foulants were applied at two concentrations, 0.5 and 1.0% (i.e., foulant-to-membrane weight ratio), to mimic mild and severe fouling conditions, respectively. The presence of BSA still led to a high yield (91 to 92%), implying its negligible impact on PET-substrate depolymerization. In contrast, the presence of SA or HA resulted in a slightly lower yield (~90%) in the low-fouling concentration and a more substantial decrease at severe-fouling conditions (~75%). This decrease was likely due to complexation between the carboxyl groups of the foulants and metal oxide catalysts that lowered catalytic activity (55). Such decreased activity could be compensated through extending the reaction time from 40 to 60 min, which restores the yield to ~92% (table S8). For gypsum, the BHET yield is ~80% under both the mild and severe fouling conditions. Such minor decrease (~12%) is likely due to physical coverage of catalyst surfaces and impeded substrate-catalyst contact. However, extending the reaction time restored BHET yields to ~90%, indicating that the observed inhibition is kinetically, not chemically, driven (fig. S20). Overall, extending reaction time effectively mitigated the impact of membrane foulants, demonstrating the robustness of the stage-based strategy for recycling EOL membranes for various applications.

Fig. 4. Stage-based recycling of EOL composite membranes in practical applications and performance evaluation of the recycled membranes.

(A) Depolymerization performance of EOL membranes under varying fouling conditions. The tested foulants included bovine serum albumin (BSA), sodium alginate (SA), humic acid (HA), and gypsum, each applied at two concentrations (0.5 and 1.0% by weight relative to the membrane, top) to simulate mild and severe fouling conditions, respectively. The mild fouling condition (i.e., blue columns at the bottom) had a negligible impact on PET substrate depolymerization, with the largest decrease of 9% in BHET yield observed for gypsum. The severe fouling condition (i.e., red columns at the bottom) exhibited a slightly more pronounced effect, with SA causing a maximum yield reduction of ~20%. Notably, these reductions in BHET yield could be fully restored by extending reaction time to 60 min (striped bar, data provided in table S8). (B) Process flow of stage-based recycling EOL membranes from real-world membrane modules. Digital photos and scanning electron microscopy images of both membrane module, EOL membranes, and recycled membranes are also illustrated. (C) Molecular weight cutoff (MWCO) test comparing separation performance of recycled and commercial UF membranes. MWCO is as the molecular weight at which 90% of the solute is rejected. (D) Mechanical properties of the recycled RO membrane and the commercial RO membrane. (E) Water permeance and NaCl rejection of the recycled RO membrane compared with the commercial RO membrane.

We further evaluate the efficiency of the stage-based strategy in recycling an EOL RO membrane module from a real-world desalination facility (Fig. 4B and fig. S21). The EOL membrane exhibited a water permeability of 1.53 ± 0.44 liters m⁻² hour⁻¹ bar⁻¹ and a NaCl rejection of 85.2 ± 2.0%, confirming its degraded performance before recycling. Surface analysis of the membrane by scanning electron microscopy (SEM) coupled with energy-dispersive x-ray spectroscopy revealed the presence of both organic, biological foulants (elements such as C, N, O, P, and S) and inorganic deposits (e.g., Ca, Mg, and Al), suggesting severe multitype fouling typically encountered in long-term industrial operation (fig. S22). Despite this contamination, our process successfully extracted the PSf layer during the solvent extraction stage, as confirmed by the disappearance of characteristic PSf peaks in the Fourier transform infrared spectra of the insoluble residue (fig. S23), leaving the remaining components to proceed to the depolymerization stage. After a 40-min reaction at 175°C, the PET substrate was depolymerized into BHET monomers, while inert materials—including acrylonitrile butadiene styrene, glass fiber, epoxy resin, PP, and ethylene propylene diene monomer—were separated (figs. S24 and S25). BHET was subsequently extracted using hot water (60°C), and the inert materials were isolated via filtration. The collected permeate containing BHET was concentrated at a low temperature (4°C) to yield BHET crystals, demonstrating the effectiveness of our stage-based strategy in recovering high-purity monomers and inert materials.

The recovered materials underwent repolymerization, spinning, phase separation, and interfacial polymerization to produce UF and RO membranes (Fig. 4B). SEM images of the recycled UF membrane showed uniform pore structure without defects. Polyethylene glycol (PEG) filtration tests (Fig. 4C) indicated a ~90% retention for PEG-300k, corresponding to a molecular weight cutoff of 300k (56), largely consistent with commercial UF membranes. In addition, water contact angle measurements suggested slightly enhanced hydrophilicity compared with commercial UF membranes (fig. S26). Such hydrophilic surface will impart the recycled UF membrane with improved anti-fouling property. SEM images of the recycled RO membrane (Fig. 4B) displayed the characteristic ridge-and-valley structure (57), indicating the successful fabrication of PA selective layer via interfacial polymerization. Tensile testing revealed that the RO membrane exhibited high stiffness (E = 1266.5 MPa), strength (σ = 45.9 MPa), and ductility (ε = 10.3%), with a higher ultimate tensile stress and a higher elastic modulus compared with commercial membranes (Fig. 4D and fig. S27). Furthermore, we achieve a water permeability of 2.63 ± 0.29 liters m⁻² hour⁻¹ bar⁻¹, with a NaCl rejection of 98.6 ± 0.5% for the recycled membranes (Fig. 4E), comparable to those of commercial RO membranes (water permeability, 2.90 ± 0.10 liters m⁻² hour⁻¹ bar⁻¹; NaCl rejection, 99.1 ± 0.3%). A combination of structural and surface analyses confirmed the removal of fouling and the recovery of membrane morphology and interfacial chemistry (58), validating full structure-function restoration after recycling (figs. S28 to S31). These findings validate the effectiveness of the stage-based strategy for closed-loop recycling membrane materials from EOL modules and underscore its potential for large-scale applications.

Sustainability evaluation of the stage-based recycling process

A cradle-to-gate LCA using the cutoff approach was conducted to evaluate the environmental impact of the stage-based membrane recycling process. Background data, including the production of chemicals and energy, were sourced from the ecoinvent v3.11 database. Activity data, including material and energy flows, were primarily obtained by constructing a conceptual plant model using Aspen software (Fig. 5A and figs. S32 to S34). The simulated plant, designed with an annual capacity of 10,000 metric tons, covered the entire process from EOL RO membranes to recycled RO membranes. The modeling framework included the following stages: (i) collection, transportation, and pretreatment of EOL RO membrane modules; (ii) solvent extraction; (iii) depolymerization and repolymerization; (iv) NIPS for producing recycled UF membranes; and (v) interfacial polymerization for fabricating recycled RO membranes (fig. S35).

Fig. 5. Layout of scaled-up process and material flow of the stage-based recycling strategy.

(A) Process flow diagram illustrating the scaled-up stage-based recycling of polymeric composite membranes. Fresh catalyst input was assumed for all TEA and LCA calculations. (B) Mass flow analysis showing the primary material inputs, conversions, and outputs across the recycling stages. MPD, m-phenylenediamine; TMC, trimesoyl chloride; Oli-PET, polyethylene terephthalate oligomer.

Mass flow analysis (Fig. 5B) revealed that 2000 tons of PSf was extracted from 10,000 tons of EOL RO membranes during the solvent extraction stage, yielding 11,111 tons of casting liquid (18% mass fraction). The remaining 7900 tons of PET substrate underwent depolymerization, repolymerization, and spinning, resulting in the production of 7568 tons of recycled PET substrate. The NIPS process combined the casting liquid with the recycled PET substrate to produce 9568 tons of recycled UF membranes. These UF membranes were further used as supports in interfacial polymerization, yielding 9665 tons of RO membranes. Notably, to ensure system sustainability, we recycled solvents including DMF, ethylene glycol, and water through rectification or vacuum distillation processes. This closed-loop approach thus maintained a balanced system of solvent consumption and replenishment, largely enhancing the environmental efficiency of the recycling process.

The environmental impacts of the stage-based recycling process were compared with those of commercial membrane production across 18 indicators, classified into three categories, including human health, resource consumption, and ecosystem impacts (Fig. 6A and figs. S36 and S37). The recycling process demonstrated reductions in environmental impacts ranging from 4 to 93% across all product systems. The indicators near the lower end of this range are primarily constrained by background burdens from energy use and chemical inputs during the recycling process, which are not fully offset by avoided virgin-material production. This results in a trade-off whereby large reductions are achieved for categories dominated by avoided virgin materials, whereas improvements in some indicators driven by energy use and additional chemical inputs remain limited. In Asia, the GWP and fossil resource scarcity (FRS) for producing recycled RO membranes are reduced by 22 and 29%, respectively, compared to commercial production process. Similarly, the GWP and FRS for recycled UF membranes are reduced by 34 and 47%, respectively. Globally, recycling the 8.56 million membrane modules used in seawater desalination and sewage treatment could yield ~307 million m² of recycled membranes. Such efforts in recycling would prevent the emission of 80,000 tons of CO₂ and reduce fossil oil consumption by 50,000 tons. According to the contribution analysis (Fig. 6B and figs. S38 and S39), energy use (largely electricity and steam) and chemical consumption are key contributors to these impacts, highlighting the importance of optimizing steam heating and substituting recycled materials.

Fig. 6. LCA and TEA for the stage-based recycling process.

(A) Comparison of LCA results of recycled membranes and commercial membranes across 18 indicators categorized into: Ecosystem, Resources, and Human health. AS, Asia. (B) Contribution analysis of the LCA results for the recycled RO membrane. (C) TEA comparing total capital investments and unit production costs for the stage-based recycling process versus conventional membrane manufacturing process. (D) Comparative analysis of three EOL management pathways for RO membranes: the proposed closed-loop recycling process (EOL RO to RO), incineration, and landfill. TA, terrestrial acidification; TE, terrestrial ecotoxicity; FEu, freshwater eutrophication; FEc, freshwater ecotoxicity; MEu, marine eutrophication; MEc, marine ecotoxicity; SOD, stratospheric ozone depletion; OFTE, ozone formation–terrestrial ecosystems; WC, water consumption; MRS, mineral resource scarcity; LU, land use; OFHH, ozone formation–human health; IR, ionizing radiation; HNCT, human noncarcinogenic toxicity; HCT, human carcinogenic toxicity; FPMF, fine particulate matter formation.

We conducted a TEA to evaluate the capital investment and unit production costs of the stage-based recycling process and compared the data with that of a conventional industrial production process (Fig. 6C and fig. S40). While the recycling process required a higher initial capital investment because of additional equipment for depolymerization, repolymerization, and solvent separation, the whole process largely offset these expenses by reducing raw material costs through the use of recycled PET and PSf. The analysis revealed substantial cost savings in unit production. For RO membranes, unit production cost decreased by 25%, from $5.33/m² to $4.00/m². Similarly, for UF membranes, the cost was reduced by 36%, from $3.41/m² to $2.18/m². These findings demonstrate that, despite the higher initial investment, the stage-based recycling process offers substantial economic advantages through material reuse and cost-efficient production.

Furthermore, a comprehensive comparison of the final disposal pathways (incineration and landfill) (59) and the proposed recycling method revealed that the stage-based recycling process outperformed conventional processes across all six key indicators, including GWP, FRS, marine eutrophication, freshwater eutrophication, terrestrial acidification, and stratospheric ozone depletion (Fig. 6D and fig. S41). Notably, even without accounting for the environmental benefits associated with the recycled products, the stage-based process consistently exhibited the lowest impacts across all categories. Hence, the stage-based recycling process represents a promising, eco-friendly, and cost-effective technology for sustainable management of EOL membranes, offering notable improvements in environmental performance and economic viability over conventional methods.

DISCUSSION

This study established a stage-based strategy for efficient recovery and functional remanufacturing of EOL polymeric composite membranes via sequential solvent extraction, catalytic depolymerization, and membrane recasting. Beyond conventional solubility screening, the extraction process was evaluated in the context of real-world constraints, including interlayer adhesion, surface fouling, and dissolution kinetics—factors that critically govern separation efficiency but are often overlooked in prior studies. The depolymerization step was implemented under contamination conditions representative of actual EOL membranes, demonstrating strong tolerance to inorganic, organic, and biological foulants without the need for precleaning. These efforts shift the focus from idealized model systems toward process-relevant conditions, offering strategic insights for scaling up recycling strategies under practical complexity. The stage-based framework allows the flexible integration of solvent-based separation and polymer-specific depolymerization, offering a versatile and scalable solution for structurally and chemically diverse composite wastes. Hence, this approach provides a broadly applicable platform for advancing closed-loop recycling of multicomponent polymer systems beyond the traditional water filtration membrane. Given the inherent complexity of polymeric waste streams and the urgency of a low-carbon economy, future efforts could focus on exploring alternative green solvents, expanded catalytic systems, and integrating renewable energy sources (60). These advancements could further enhance the scalability, efficiency, and sustainability of the approach, providing a robust framework for sustainable management of mixed or composite polymer wastes across various applications.

MATERIALS AND METHODS

Materials

The membrane materials used in this experiment was received from a commercial source. The RO membrane (BW30) and NF membrane (NF90 and NF270) were purchased from DuPont, the UF membrane (PSf and PAN) was purchased from OriginWater (China), and the MF membrane (PVDF) was purchased from RisingSun (China). The EOL RO membranes were retrieved from a desalination plant after ~5 years of continuous operation. The commercial UF membrane used for performance comparison is the PSf membrane from OriginWater (China), and the commercial RO membrane used for benchmarking is the BW30 membrane from DuPont. Reagents used in solvent extraction process, including DMF (≥99.8%), 1-methyl-2-pyrrolidone [analytical reagent (AR)], chloroform (AR), DCM (AR), THF (AR), and DMSO (AR), were all purchased from Sinopharm Chemical Reagent (China). Chemicals used for the synthesis of the depolymerization catalyst included magnesium chloride hexahydrate (AR), ammonium molybdate (Mo: 56.5%), and l-alanine (≥99.5%), which were obtained from Aladdin Biochemical Technology (China). In addition, sodium bicarbonate (AR), and ethanol (AR) were purchased from Sinopharm Chemical Reagent (China). Ethylene glycol (98%, Macklin Biochemical Company) was used as a solvent in depolymerization process, while ethylene glycol antimony (Sb: 57 ± 1%, Aladdin Biochemical Technology) was used as a catalyst in the repolymerization.

Stage-based recycling process

Solvent extraction of PSf

Dried waste membranes (20 g) were first cut into 1 cm–by–1 cm fragments. These fragments were then mixed with DMF (30 g) and agitated for 10 min. After agitation, the mixture was filtered using microporous membranes (0.45 μm) to separate the casting solution from the polluted PET substrate.

Depolymerization of PET substrate

The PET substrate obtained from the solvent extraction process was thoroughly dried in oven at 60°C. Subsequently, the dried substrate (0.5 g), Mo-MgO catalyst (5 mg), and ethylene glycol (2 g) were added to a single-necked flask to initiate a depolymerization reaction at 175°C with a continuous stirring at 300 rpm for 40 min. After the reaction, the resulting solution was extracted and introduced into 80°C water, maintaining a mass ratio of water to BHET of 50:1. The mixture was briefly stirred for 1 min, and then promptly filtered using microporous membranes (0.45 μm) while maintaining the high temperature of 80°C. The weight of unreacted PET substrate was measured to calculate substrate conversion by

$$\text{Substrate conversion}=\frac{m_{\mathrm{i}}-m_{\mathrm{r}}}{m_{\mathrm{i}}}\times 100\%$$ (1)

where m_i is the quality of PET substrate before reaction and m_r is the quality of PET substrate after reaction. The filtrate was collected, and 1 ml of the filtrate was analyzed using high-performance liquid chromatography to determine the BHET yield and STY of the catalyst by

$$\text{BHET yield}=\frac{n_{\text{BHET}}}{\text{theoretical} n_{\text{BHET}}}\times 100\%$$ (2)

$$\text{STY}=\frac{m_{\text{BHET}}}{m_{\text{cat}.}\times t}$$ (3)

where n_BHET and m_BHET are the amount and quality of BHET produced by the reaction, theoretical n_BHET is the amount of BHET generated by complete reaction of PET, and m_cat. is the quality of Mo-MgO catalyst. The remaining filtrate was concentrated by rotary evaporation to remove 90% of the water. The residual solution was then refrigerated at 4°C for 12 hours, followed by additional filtration using microporous membranes (0.45 μm). The resulting solid was thoroughly dried to yield BHET, which was then used in repolymerization as a monomer source for producing PET.

Repolymerization

The recycled BHET (100 g) and ethylene glycol antimony (0.15 g, 0.15 wt %) were introduced into a 250-ml custom-made titanium polyester reactor. First, the system was purged with nitrogen for 5 min and then heated to 230°C, which was maintained for 2 hours. Following this step, the temperature was further elevated to 270°C and maintained for 1 hour. During the second heating phase, the reaction system gradually underwent decompression, reaching ~60 Pa. The resulting repolymerized PET was then spun to produce clean PET substrates. The melt spinning temperature was 330°C, and the hot rolling temperature was 260°C.

Nonsolvent-induced phase separation

Initially, the casting solution (18 wt %) was degassed in a desiccator for a minimum of 15 hours before casting. The cleaned PET substrate was affixed to a sterile glass plate (20 cm by 25 cm) using waterproof tape. DMF was applied to prewet the substrate, and any excess DMF was subsequently removed using Kimwipes (Kimberly-Clark, Roswell, GA). The casting solution was spread over the wetted substrate using a casting knife (Gardco, no. 10548), with the gate height calibrated to 10 mils (~254 μm). Subsequently, the entire composite was immersed in a precipitation bath containing 3–wt % solvent in deionized (DI) water at room temperature (25°C) to facilitate phase separation. The composite was retained in the precipitation bath for 10 min before being transferred to a deionized water bath for storage to obtain a recycled UF membrane.

Interfacial polymerization

The PA active layer was synthesized via interfacial polymerization on the recycled UF membrane. Initially, the membrane was immersed in a m-phenylenediamine (MPD) solution (2 wt % in DI water) for 3 min. After the excess MPD was removed from the membrane surface using a rubber roller, the membrane was immersed in a trimesoyl chloride (TMC) solution (0.1 wt % in n-hexane) for 1 min to form a PA selective layer on the PSf support; subsequently, the excess TMC solution was drained vertically for 2 min. The membrane was then cured in a deionized water bath at 60°C for 10 min, resulting in the formation of a recycled RO membrane. The recycled RO membrane was thoroughly rinsed and stored in deionized water at 4°C.

Evaluation of membrane performance

PEG rejection test

PEG with varying weight-average molecular weights (M_w)—10,000, 20,000, 100,000, and 300,000 g mol⁻¹—was dissolved in deionized water at a concentration of 0.1 g liter⁻¹. Following the addition of 50 ml of each PEG solution to the feed chamber, compressed nitrogen (N₂) was applied to the cell to achieve a steady pressure of 0.6 bar. Stirring at 600 rpm minimized the effects of concentration polarization. Initially, 10 ml of permeate was discarded, and subsequent 15-ml portions were collected to determine the PEG concentration. The concentrations of PEG in the feed, permeate, and retentate were quantified using a total organic carbon analyzer (Shimadzu, Japan). The PEG retention rate (R_PEG) was calculated using the formula

$$R_{\text{PEG}}=(1-\frac{C_{\mathrm{p}}}{C_{\mathrm{f}}})\times 100\%$$ (4)

where R_PEG is the retention rate, C_p is the concentration of PEG in the permeate, and C_f is the initial concentration of PEG in the feed solution.

NaCl rejection test

Water flux and salt retention rate were measured using a cross-flow system operating at a pressure of 15.7 bar. Initially, the system was stabilized with pure water for 8 hours, after which the feed solution was replaced with a 2 g liter⁻¹ NaCl solution. Following a 12-hour stabilization period, NaCl concentrations of both the feed and permeate were determined using a conductivity meter (FiveEasy Plus, FE38). Water permeability, A, was calculated by

$$A=\frac{V}{\mathit{StP}}$$ (5)

where A is the water flux, V is the volume of permeated solution, S is the effective filtration area, t is the permeation time, and P is the operating pressure during the measurement. NaCl rejection, R_NaCl, was determined by

$$R_{\text{NaCl}}=(1-\frac{C_{\mathrm{p}}}{C_{\mathrm{f}}})\times 100\%$$ (6)

where R_NaCl is the NaCl retention rate, C_p is the concentration of NaCl in the permeate, and C_f is the initial concentration of NaCl in the feed solution.

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S41

Tables S1 to S59

References