Mechanism elucidation and scaling control in membrane distillation using 3D printed carbon nanotube spacer

The study addresses the persistent problem of inorganic scaling in membrane distillation (MD), particularly calcium sulfate (CaSO4) scaling, which leads to pore blockage, flux decline, and membrane wetting. MD inherently experiences coupled temperature and concentration polarization near the hydrophobic membrane surface, aggravating fouling and scaling. Conventional scaling mitigation strategies (chemical dosing, pH adjustment, antiscalants, or enhancing membrane hydrophobicity) have limitations, costs, and environmental burdens and often lack long-term efficacy. Feed spacers can reduce boundary layers and polarization. Carbon nanotubes (CNTs), with excellent mechanical, thermal, and interfacial properties, can enhance turbulence, heat transfer, and hydrophobicity. Prior work showed CNT-embedded spacers improved MD performance via multiscale roughness. The research question here is whether a 3D-printed CNT-embedded spacer can mitigate CaSO4 scaling in DCMD, how it alters the scaling mechanism, and whether observed behavior can be elucidated by in situ imaging and a mechanistic MD–crystallization model. The working hypothesis is that CNT spacer roughness enhances bubble generation and alters nucleation/attachment–detachment dynamics, promoting bulk crystallization and reducing surface scaling and flux loss.

Prior literature establishes that MD suffers from temperature and concentration polarization, which depress vapor pressure driving force and foster fouling (Martínez-Díez and Vázquez-González; Alsaadi et al.; Manawi et al.). CaSO4 is a common scalant with low solubility; surface and bulk crystallization can block pores and induce wetting, causing rapid flux decline (Christie et al.). Conventional mitigation through pH control, antiscalants, and increasing membrane hydrophobicity has operational and environmental drawbacks and limited longevity (He et al.; Elcik et al.; Warsinger et al.; Rolf et al.). Spacers act as turbulence promoters reducing polarization (Phattaranawik et al.; Chang et al.). CNT-based materials can enhance MD via heat transfer, surface roughness, and hydrophobicity (Tijing et al.; Roy et al.; Noy et al.; Lee et al.). CNTs may reduce scale formation and improve reversibility (Humoud et al.). However, studies on CNT-embedded spacers are limited; prior work by the authors demonstrated performance gains with CNT-embedded 3D-printed spacers. Nucleation and growth are sensitive to hydrodynamics and surface morphology/roughness (Grosfils & Lutsko; Jin et al.; Løge et al.). Bubble-induced mixing can mitigate scaling (Ye et al.; Ali et al.; Farid et al.). These studies motivate investigating CNT spacer-induced mechanistic changes (bubble formation, nucleation/attachment–detachment) and modeling crystallization coupled with MD.

Experimental set-up and materials: Direct-contact membrane distillation (DCMD) was performed in a transparent poly(methyl methacrylate) module with a viewing window for OCT. Feed and permeate flow rates were held at 0.5 L/min (laminar flow) using gear pumps. Feed channel dimensions: 0.015 × 0.06 × 0.002 m (L × W × H). Feed temperatures were 60, 70, or 80 °C (or set to achieve comparable initial flux among spacers), and permeate was maintained at 19.5 ± 0.5 °C via chiller. Permeate was continuously weighed to calculate flux, with data logged every minute; tests were repeated ≥3–5 times. The module was stirred (200 rpm) in the feed tank to avoid sedimentation. Volume concentration factor (VCF) and saturation ratio (SR) were used to track concentration and supersaturation. Membrane: Hydrophobic PVDF (Durapore) membrane; pore size 0.22 µm, porosity 70%, thickness 125 µm, effective area 0.0009 m². Spacers: 3D-printed CNT-embedded spacers fabricated per prior work: 1 mm thickness, 45° filament angle, 4 mm filament spacing. Initial print size 10 × 4 cm, cut to 6.0 × 1.5 cm for tests. PLA spacer of same geometry used as control. A no-spacer condition was also included. Feed solution: CaSO4 (as CaSO4·2H2O) at 0.01 M prepared by dissolving 1900 mg/L in DI water, stirred 1 day for complete dissolution. CaSO4 solubilities noted (0.255 g/100 mL at 20 °C; 0.244 g/100 mL at 60 °C) to contextualize supersaturation. Operating protocols: Experiments at feed 60, 70, 80 °C compared no-spacer, PLA spacer, and CNT spacer. Additional tests adjusted feed temperatures to equalize initial flux (~45–46 LMH) across spacer types (e.g., no-spacer ~65 °C, PLA ~80 °C, CNT 60 °C) to decouple initial drag effects. Flux vs VCF was the main performance metric. In situ and ex situ characterization: Optical coherence tomography (OCT; OQ Labscope 2.0) captured 2D cross sections (512 × 512 pixels over 1.0 × 1.0 mm) at intervals to visualize scaling thickness at the membrane surface; images processed in ImageJ for noise reduction, contrast/brightness adjustments, and thickness conversion (1 pixel = 2 µm). Scanning electron microscopy (SEM; Zeiss SUPRA40VP, 10 kV, 1000×) and EDS mapping (Oxford X-Max) characterized crystal morphology and composition; samples dried overnight (~24 ± 1 °C) and sputter-coated with Pt. Water contact angle (WCA) measured (Phoenix10) at five points per sample after drying to assess hydrophobicity loss over time. Bubble observations and tests: Visible flow imagery captured by smartphone documented bubble generation on the feed side with spacers during operation. Bubble counts over 1 h were tabulated for each case. To isolate bubble effects, auxiliary tests injected artificial bubbles (air pump DK-9000) into the PLA spacer case under otherwise identical conditions, comparing flux decline and post-operation cleanliness. Modeling: A mechanistic MD–crystallization model was developed using population balance equations (PBE) and moment methods to represent secondary nucleation and crystal growth in four domains: feed tank, module bulk, membrane surface, and spacer surface. Assumptions: secondary nucleation dominates; well-mixed tank and module; negligible breakage/agglomeration; growth rate independent of crystal size. The 0th–3rd moments (µ0–µ3) were solved in each domain, with nucleation and growth rates modeled as B(t)=knΔc^β and G(t)=kgΔc^α, where Δc is supersaturation relative to temperature-dependent solubility. Flux decline arises from water removal (concentrating feed), crystal formation (reducing dissolved solutes), and reduction of effective membrane area by surface scaling. Model parameters (nucleation/growth rate constants and orders, attachment–detachment related effects) were estimated by fitting experimental flux vs VCF data; supporting parameter studies are provided in Supplementary Information. Bubble effects were not included in this modeling phase.

• Flux behavior vs VCF and temperature:
  • At 60 °C: No-spacer initial flux ~28 LMH, sharp decline after VCF ~1.9, complete flux loss by VCF 2.4. PLA spacer maintained ~42 LMH from VCF 1.0–2.3, then gradual decline to VCF 3.1 followed by sharp drop; both no-spacer and PLA eventually lost flux. CNT spacer reached ~46 LMH initially, remained near-constant from VCF 1.0–3.0, then declined slowly; flux stayed >29 LMH up to VCF 4.0 and >20 LMH to end of test, indicating delayed scaling.
  • At 70 °C: Increased flux and faster decline overall. No-spacer and PLA fluxes dropped rapidly after VCF ~1.7 and ~2.1, respectively. CNT spacer exhibited gradual decline after VCF ~2.1, maintaining >58 LMH up to VCF 4.0.
  • At 80 °C: Highest initial fluxes and fastest scaling. CNT spacer showed a sharp drop from >95 LMH, then stabilized at ~30 LMH without total flux loss, indicating pores not fully blocked despite scaling.
• Same-initial-flux comparison (initial ~46 LMH): No-spacer (65 °C) and PLA (80 °C) started declining after VCF ~1.4 and ~2.2, respectively, whereas CNT (60 °C) maintained >29 LMH to VCF ~4.5, demonstrating superior scaling resistance at comparable initial concentration rates.
• Supersaturation tolerance: Despite reaching SR>1 (supersaturation), CNT spacer experienced only ~45% flux reduction, maximizing water production without complete performance loss, unlike PLA or no-spacer.
• OCT scaling thickness (Table 1): At VCF 2.4: No-spacer 30.8 ± 7.1 µm; PLA 10.0 ± 2.0 µm; CNT none (±0). After ~12 h: No-spacer 30.8 ± 7.1 µm (VCF 2.4); PLA 20.8 ± 5.2 µm (VCF 3.5); CNT 8.3 ± 4.5 µm (VCF 5.0). CNT spacer markedly reduced surface scale buildup even at higher VCF.
• Hydrophobicity (WCA, Table 2): Virgin membrane 111 ± 4°. After 4 h: No-spacer (VCF 1.5) 36 ± 2°; PLA (VCF 1.8) 75 ± 3°; CNT (VCF 2.0) 98 ± 2°. After 12 h: No-spacer (VCF 2.4) 37 ± 18° and effectively lost hydrophobicity; PLA not measurable; CNT (VCF 5.0) 37 ± 18°, indicating hydrophobicity retained longer with CNT than others.
• SEM/EDS: No-spacer and PLA showed progressive surface coverage by crystals with pore blockage by ~10–12 h; PLA reduced scaling under spacer filaments but not in open areas. CNT spacer case showed sparse, fragmented crystals on the membrane until ~8 h; larger crystals appeared later but were fewer and more weakly attached; bulk crystal sizes were larger with CNT (consistent with enhanced bulk growth). EDS confirmed CaSO4 as the scalant.
• Bubble generation: After 1 h, bubble counts were 0 (no-spacer), 3 (PLA), and 41 (CNT). CNT produced numerous, larger bubbles that increased over time (1 h vs 12 h), likely due to enhanced vaporization on rough CNT surfaces; bubbles altered near-wall flow and reduced CP and scaling. Artificial bubble injection in PLA case reproduced much of the CNT benefit (higher initial flux, slower decline; flux >30 LMH at VCF 3.5) but did not fully match CNT performance or its biphasic decline signature.
• Mechanistic modeling: The MD–crystallization model reproduced experimental flux–VCF trends across temperatures and spacer types. Parameter estimation suggested CNT spacers have lower membrane nucleation rate constants but higher spacer nucleation rate constants than PLA, consistent with nuclei detachment from rough CNT surfaces, enhanced secondary nucleation in bulk, and reduced dissolved solute concentration. The CNT case exhibited a biphasic flux decline (initial linear drop followed by a quasi-plateau), aligning with the hypothesized interplay of water removal and bulk crystallization.

The findings support the hypothesis that CNT-embedded spacers fundamentally alter CaSO4 scaling in DCMD. The multiscale roughness of CNT spacers increases local vaporization rates and promotes bubble formation, which enhances near-wall mixing, reduces concentration polarization, and physically hinders scale adhesion at the membrane. Concurrently, rough CNT surfaces favor nucleus detachment, shifting crystallization from membrane surfaces toward the bulk and spacer zones; detached nuclei serve as seeds for secondary nucleation, and subsequent bulk growth consumes dissolved solutes. This dual pathway explains sustained flux at high VCF and supersaturation, the reduced and delayed surface scaling seen by OCT/SEM, and the presence of larger, more weakly attached crystals with CNT. Comparisons at equal initial flux decouple simple hydrodynamic drag effects and underscore intrinsic anti-scaling advantages of CNT spacers over PLA or no-spacer. The mechanistic MD–crystallization model, while simplified, captures the flux decline trajectories and suggests quantitatively that CNT spacers reduce membrane nucleation while enhancing spacer/bulk nucleation. Bubble-assisted experiments with PLA corroborate the beneficial role of bubbly flow but also reveal that bubbles alone do not fully account for the CNT’s biphasic flux decline and superior performance, implicating surface roughness-driven nucleation/attachment–detachment dynamics as an additional key factor. Collectively, the results demonstrate that CNT spacers mitigate scaling by modifying both transport and crystallization mechanisms, maintaining productivity under conditions where conventional spacers fail.

This work experimentally and theoretically elucidates how 3D-printed CNT-embedded spacers mitigate CaSO4 scaling in DCMD. CNT spacers deliver higher initial fluxes and sustain significant permeate production at high VCF, with only modest flux loss even under supersaturation, while preventing total flux collapse seen with PLA or without spacers. OCT, SEM, WCA, and bubble observations reveal reduced and delayed surface scaling, enhanced bubble generation, and a shift toward bulk crystallization with larger, weakly attached crystals. A mechanistic population-balance-based MD–crystallization model reproduces observed flux–VCF trends, supporting a mechanism wherein CNT roughness reduces membrane nucleation but increases spacer/bulk nucleation via enhanced detachment of nuclei. Practical implications are that CNT spacers can extend MD operating windows, reduce reliance on chemical antiscalants, and improve long-term stability. Future research should: (i) conduct fundamental crystallization experiments to independently quantify nucleation and growth kinetics on membranes, PLA, CNT spacers, and crystals; (ii) incorporate multiphase CFD coupled with crystallization kinetics to explicitly account for bubble dynamics and optimize spacer architecture and operating conditions; and (iii) validate performance with complex waters and over extended durations to assess durability and generalizability.

Parameter estimation for the mechanistic model relied solely on flux–time (or flux–VCF) data; time-resolved solute concentrations and distributed crystal inventories (bulk, membrane, spacer) were not directly measured, leading to potential non-uniqueness of fitted parameters. Bubble effects and multiphase flow were not included in the current model, limiting mechanistic completeness. The mapping of nucleation and growth constants to specific physical processes remains partly inferential. Additional well-controlled experiments are needed to independently resolve nucleation/attachment–detachment mechanisms on each surface type and to incorporate bubbly-flow effects for more quantitative predictive capability.