1. Introduction
In 2017, an estimated 29% of the global population did not have access to safely managed water supplies, and 785 million people lacked even a basic drinking service [
1]. While one aim of the United Nations Sustainable Development Goals is to ensure universal and equitable access to safely managed and affordable drinking water for all [
1], household water treatment (HWT) options are currently promoted as an interim solution for those without safe drinking water [
2].
HWT methods, such as boiling, chlorination, flocculant/disinfectant powder, solar disinfection, and filtration, have been shown to improve microbiological water quality and reduce diarrheal disease among users [
3,
4]. Among filtration methods, there has been extensive previous research and promotion of sand and ceramic filters [
5,
6]. More recently, hollow-fiber membrane filters (HFMFs) have been increasingly promoted. HFMFs consist of several hundred tubular fibers, with specified pore sizes, packed together. Water flows into the filter casing, through the porous membrane walls into the cores, and into a storage container. With advertised pore sizes between 0.1 and 0.02 μm [
7], HFMFs remove >99% bacteria from contaminated water in laboratory settings [
8]. As such, HFMFs can obtain at least a targeted protection rating from the WHO Certification Scheme for household treatment products without any other treatment [
8].
In Rwanda, one HFMF was shown to reduce diarrhea in users by 78% 12–18 months after distribution [
9]; in the Democratic Republic of the Congo, HFMF use reduced diarrhea by 15% [
10]. However, microbiological water quality improvement was lower than anticipated in both studies, as 29% and 36% of intervention households had detectable thermotolerant coliforms (TTCs) in filtrated water. HFMF effectiveness in removing microbiological contaminants was also lower than laboratory efficacy in other field studies. In Kenya, between 18 and 26 months after filter distribution, a geometric mean of 20.6
E. coli CFU/100 mL was observed in filtered water [
11]. In Honduras, 52% of filters tested after 23 months of use had filtrate with >10
E. coli CFU/100 mL [
12]. In South Sudan, 38–40% of filtrate samples were ≥1 TTC CFU/100 mL after the second visit for filters cleaned after previously having contaminated filtrate [
13]. In Fiji, 71% of filtered water samples had microbiological contamination [
14]. Other reported concerns included misuse, broken/missing parts without access to replacements, slow flow, difficulty with backwashing (due to lost parts, difficulty in use, or need for clean water), and bad water smell/taste [
11,
14,
15,
16]. Thus, there appears a consistent disparity between laboratory efficacy and field effectiveness results when using HFMFs for HWT in low- and middle-income countries (LMICs) and humanitarian contexts.
One potential reason for the discrepancy in microbiological membrane performance between laboratory and field results is differences in membrane fouling and the associated mechanisms to control fouling [
13,
16]. Fouling is the adhesion and accumulation of influent water components on membrane surfaces that can cause a decline in effluent quality and quantity [
17,
18,
19] and changes in membrane material properties [
20,
21]. Several principal fouling mechanisms have been identified, including [
17,
22] physical fouling, such as pore blockage by colloids of similar size to the pores, or cake formation, formed by larger particles depositing on the membrane surface; chemical fouling, such as organic adsorption (e.g., humic or fulvic acids), and inorganic precipitation (due to pH change, hydrolysis, or oxidation of Ca, Mg, CO
3, SO
4, Si, and Fe compounds); and biological fouling caused by microorganisms that adhere to the membrane surface and grow to create a biofilm. Fouling can be reversible, where particles are removed by physical processes, or irreversible, where the material is adsorbed to the pores and requires chemical cleaning to recover performance [
23]. Fouling is influenced by water chemistry, membrane properties, temperature, mode of operation, and hydrodynamic conditions [
17,
22,
23]. While membrane fouling has been extensively investigated in other applications, and it is known that membranes foul over time [
24], questions remain on how quickly fouling occurs on HFMFs for HWT in LMICs and humanitarian contexts and its reversibility by using locally available backwashing solutions. Membrane fouling can be controlled via design and operation, pretreatment of influent water, and regular maintenance (e.g., daily backwashing/forward flushing, chemical cleaning with alkalis or acids [
23,
25]). HFMFs use membranes initially developed for use within well-controlled, large-scale water treatment facilities operated by experts to manage influent water quality. A small-scale application of HFMFs is within cartridges designed for short-term use (e.g., treating surface water while camping). These small-scale cartridges have also been widely distributed for free to families living in LMICs and humanitarian contexts for HWT, intended to last a family for many months to years. In these contexts, it is not feasible to optimize a site-specific filter design, pretreat water, perform intensive maintenance using clean solutions and without potentially damaging fibers, or know when an HFMF has failed. To control fouling, some HWT HFMF manufacturers recommend forcefully backwashing or soaking the filters with clean water, diluted chlorinated solution, vinegar, salt water, or hot water [
26,
27,
28]. Manufacturers claim that HFMFs can be “cleaned and reused almost indefinitely” [
29] and used for “an extended period of time (years) if properly maintained” [
30]. However, laboratory evidence to support using these recommended locally available cleaning mechanisms to recover membrane performance while maintaining filter function is lacking.
With the goal of providing recommendations for the backwashing of HFMFs to inform HFMF distribution programs and membrane applications in LMICs and humanitarian contexts, this research explores how fouling caused by different influent water qualities and backwashing with locally available solutions impact the filters’ mechanical properties, water quantity, and quality performance.
2. Materials and Methods
To complete the research, (1) four commonly-distributed HFMFs were selected and purchased; (2) smaller modules were assembled for testing; (3) a system was built to test HFMFs with three influent water types; (4) a system was built to simulate backwashing with three cleaning solutions; (5) three data collection runs were conducted, including fouling (filtering with influent waters), influent and effluent water quality testing, and fouling removal (backwashing); (6) postmortem SEM and tensile strength tests were conducted; and (7) data were analyzed. Each step is described below. In total, four HFMF filters, three influent waters, and three backwashing solutions (plus no-backwashing control) were tested in duplicate for a total of 96 modules.
2.1. Selection of Commonly Distributed HFMFs
Four different commonly distributed HFMFs that met the minimum criteria for HWT filters were purchased [
7]. All HFMFs consisted of fibers bundled in a U-shape inside a plastic casing, and water moved via gravity flow from outside to inside the fiber during treatment. Filters 1, 2, and 3 (F1–3) consisted of plastic-encased 0.1-μm pore size microfiltration (MF) hollow fibers, with microporous outer surfaces and relatively symmetric cross-sections. For F1–F2, no pretreatment mechanism was provided. A 75-micron foam prefilter was included with F3. A 50–60 mL syringe to complete backwashing and a water storage container were provided with all three filters. Filter 4 (F4) consisted of 0.02-μm ultrafiltration (UF) membranes with a highly asymmetric cross-section and a dense outer selective layer encased in a plastic structure. The structure also contains an 80-μm prefilter, a pump for backwashing filtered water, and a water storage area.
2.2. Preparation of Modules for Testing
To prepare the test modules, plastic filter casings were carefully opened using a circular saw. Individual fibers were gently detached at the base with a razor. To obtain 24 modules of each HFMF type with similar flow rates, 3 (F1 and F3), 4 (F2), and 15 (F4) fibers were potted in a U-shape (F1–3) or straight (F4) into 1/4′′ internal diameter (ID) semiclear crack-resistant polyethylene tubing and glued to create a seal without blocking fiber pores (
Figure S1). The average fiber lengths for F1–4 were 33, 38, 41, and 186 cm, respectively. Modules were then tested to verify their integrity and ensure similar flow by filtering deionized (DI) water for one hour at 1 psi transmembrane pressure and weighing the effluent. Modules with the most similar initial flow (Q
0 determined with
Equation (S2); SI) were selected for inclusion.
2.3. Filtration System
Four filtration units, each holding eight modules from the same HFMF, were built to ensure a continuous flow of influent water through modules (
Figure S2a). Each system consisted of a plastic 20-L tank connected to the modules through a network of 1/4′′ ID pipes and manifolds, regulated by an air regulator (0.5–3.5 psi outlet pressure range) and a pressure gauge (Kodiak Controls KC25-3# Low-Pressure Gauge 3 psi) to maintain a constant transmembrane pressure of 0.5–1 psi, without air pockets. This pressure approximated a filled 5-gal bucket head pressure with a 1 ft tube, a common implementation of HFMFs in households. Tanks were placed on a stir plate to avoid the settling of influent water particles and bacteria. Before an experimental run, the system was cleaned with 5% sodium hypochlorite, rinsed with Milli-Q
® water to remove any chlorine, and left to air dry. Complete rinsing of chlorine from the system was confirmed (free chlorine residual equal to 0 mg/L) using a Lamotte colorimeter (Lamotte, Chestertown, MD, USA).
Fouling experiments were run with three influent water compositions. Influent #1, simulating biological fouling and termed “Bacteria”, consisted of Milli-Q water buffered with phosphate-buffered saline (PBS) solution with 10
6 CFU/100mL
E. coli (ATCC 11220).
E. coli cultures were prepared from streak plates in Luria−Bertani broth using standard methods [
31].
E. coli cell concentrations in exponential phase cultures were estimated through turbidometry using a spectrophotometer (OD = 600 nm; GeneQuant100, GE Healthcare Life Sciences, Hatfield, UK) as per Standard Methods [
31] and used to calculate the spike culture volume into 20 L buffer solution. Influent #2 simulated combined biological and chemical fouling and was termed “BacChem”. It consisted of Influent #1 with the addition of 0.63 g/20 L humic acid (HA; 6813-04-4, Alfa Aesar, Haverhill, MA, USA) and 1.1 g/20 L CaCl
2 (C20010–1000.0, Research Products International Corporation, Madison, WI, USA) to obtain theoretical total organic carbon (TOC) and Ca
++ concentrations of 15 mg/L. Influent #3 simulated combined biological and physical fouling and was termed “BacSed”. It was prepared by adding 2.2 g/20 L sediment (ISO spec. 12103-A2 fine test dust) to Influent #1 to reach 30 NTU turbidity. Fine test dust includes particles larger and smaller than HFMF pores (0.97–124.50 μm) and is composed of 77–91% silica and aluminum and traces of Fe, Na, Ca, Mg, Ti, and K.
2.4. Backwashing System
The backwash system consisted of three 1-L tanks filled with cleaning solution, connected to four modules via 1/4′′ ID pipes, and an air regulator (0.5–10 psi outlet pressure range) and a pressure gauge (Kodiak Controls KC25-3# Low-Pressure Gauge 10 psi) to maintain a constant backwash pressure of 5 psi (
Figure S2c).
Three locally available cleaning solutions in LMICs and humanitarian contexts were tested: bleach, vinegar, and clean water. Bleach solution was prepared by diluting 6% laboratory-grade commercial bleach (Pure Bright, KIK International, Houston, TX, USA) to 0.5% sodium hypochlorite (NaOCl) with Milli-Q
® water. Concentration was confirmed using iodometric titration (Hach Method 8209, Loveland, CO, USA). This 0.5% concentration was selected because a 10:1 dilution is simple to prepare, a 0.5% solution efficaciously removes
E. coli from surfaces [
32], and it was between the HFMFs’ manufacturer-provided ranges [
26,
27]. Commercial vinegar (6% acetic acid; Heinz, Pittsburgh, PA, USA) was purchased and used without modification, as recommended by manufacturers [
27]; it is available in LMICs and humanitarian contexts. DI water was used without modification as clean water.
2.5. Data Collection Runs
Each module was exposed to daily fresh influent water for 17.5 h/day for 10 consecutive days and backwashed for 5 min/day. This aimed to simulate a timeframe of ~3 months of normal use (2.5 h/day) with cleaning once per week, a timeframe shown to be before membrane failure begins in South Sudan [
13].
To ensure the consistency of the influent water, turbidity, E. coli concentration, total dissolved solids (TDS), and pH were tested at the beginning of daily filtration, and turbidity and E. coli concentration were tested again after filtration. Backwashing solution pH was also tested daily to ensure consistency.
Turbidity was measured with a calibrated Lamotte 2020we turbidimeter (Lamotte, Chestertown, MD, USA). Triplicate readings were averaged, with results of zero replaced with the minimum detection limit (0.05 NTU).
E. coli concentrations were measured by preparing appropriate dilutions of the samples by filtering through a 0.45 µm membrane, plating on m-ColiBlue24
® (Hach, Loveland, CO, USA) media, and incubating at 35 °C for 24 h [
31].
E. coli colonies were then enumerated and recorded. Plates with no detectable colonies were replaced with the minimum detection limit of 1 colony as a conservative estimate. The geometric mean of plates in the countable range (1–200 colonies) was calculated and reported in CFU/100 mL. TDS was measured with a PC60 Premium Multiparameter probe (APERA Instruments, Columbus, OH, USA). pH was measured with an Orion
® 9106BNWP probe (Thermo Electron Corporation, Waltham, MA, USA).
During the first 14 h of filtration, effluent water was collected in nonsterilized 2-L HDPE collection bottles and weighed (
Figure S2a). The pressure and exact duration of filtration were recorded. Then, effluent water was collected into sterile 118 mL WhirlPak™ bags (Nasco Company, Fort Atkinson, WI, USA) for
E. coli and turbidity testing and an additional flow reading using the methods described above (
Figure S2b). This “second” filtration lasted ~3.5 h, depending on flow rates, as a minimum volume of 118 mL was needed to conduct reliable
E. coli testing. Each module was tested for
E. coli and turbidity every other day, and when the effluent collected the day before showed
E. coli contamination breakthrough.
After filtration each day, all noncontrol modules were backwashed using the backwashing system and their assigned backwashing solution. Between backwashing and restarting filtration, modules were filled with PBS solution to maintain a wetted state to avoid damaging fibers and prevent the bacteria from desiccating. Nonbackwashed control modules were similarly maintained in PBS between filtration cycles.
The filtration/backwash cycle was repeated for nine consecutive days (
Figure 1). On the 10th day, only one duplicate was backwashed after filtration, and a final filtration with DI water was performed to prepare the modules for postmortem analysis.
2.6. Postmortem Analysis
After each run, modules were carefully cut open and the fibers separated for postmortem analyses. The fibers were left to dry at ambient temperature overnight (for F1–3) or for 24 h (F4) due to the larger fiber volume (
Figure S3). After drying, all fibers were stored at 4 °C. To assess the mechanical properties, tensile tests were conducted using an ARES-LN2 rheometer (TA Instruments, New Castle, DE, USA) on at least three fibers per module at room temperature. Fibers were clamped at both ends and pulled at a constant elongation velocity of 0.02 mm/sec until failure; break force was measured, and strain at break (
Equation (S1) in SI) was then calculated. These parameters relate to membrane material properties, and a decrease of these coupling parameters could indicate an increase in toughness and fiber embrittlement [
21]. Tests were also conducted on two additional modules of each membrane type soaked in DI water (no filtration) over the 10-day experimental period, to be used as a baseline (“Base Wet”) in postmortem analyses, as well as on unsoaked fibers (“Base Dry”) (
Figure 1). Please note that burst pressure was not assessed because the membranes are porous and somewhat anisotropic; as such, break force is a reasonable proxy to burst pressure.
Fiber morphology was characterized by scanning electron microscopy (SEM) using a Phenom G2 Pure SEM (Phenom-World BV, Eindhoven, The Netherlands) operating at 5 kV. Fiber cross-sections were obtained by freeze-fracturing in liquid nitrogen. Before imaging, all samples were coated with gold-palladium for 90 s using a Cressington Sputter Coater 108 (Cressington Scientific Instruments, Watford, UK).
2.7. Data Analysis
Data were entered into Microsoft Excel (Redmond, WA, USA) and analyzed with R (RStudio, Vienna, Austria). Qualitative SEM analysis was used to confirm fouling at the study’s end by imaging control (nonbackwashed) modules and backwashed modules, with and without a final backwash. Indicators developed to assess impacts of fouling and backwashing included (equations in
SI): strain at break and break force (mechanical properties), normalized flow (water quantity performance), and
E. coli log reduction value (LRV) and turbidity reduction value (TRV) (water quality performance). Baseline and endline were defined for each indicator and used to calculate the relative change in performance over the length of the study (
Table 1;
Equation (S5)).
First, to determine if MF filters could be combined for analysis, a Kruskal–Wallis test was used to assess differences in select baseline and all endline indicators between: all filters (F1–F4); MF (F1–F3) filters only; and averaged MF filters (F1–3avg) and F4 (UF). Based on the results, MF filter results were combined and averaged for subsequent analysis. Then, a Kruskal–Wallis test was used to compare “Base Dry” to “Base Wet” break force and strain at break for each individual filter to assess the impact of wetting.
To confirm fouling on nonbackwashed modules, a Wilcoxon signed-rank test was used to compare baseline to endline indicators for MF (F1–3avg) and UF (F4) filters. To assess fouling consequences on nonbackwashed modules, a Kruskal–Wallis test was used to compare the endline indicators between influent water conditions (Bacteria, BacChem, and BacSed) for MF (F1–3avg) and UF (F4) filters.
To assess the backwashing impact, a Kruskal–Wallis test was used to compare endline indicators between backwashed and nonbackwashed filters for each influent water condition (Bacteria, BacChem, and BacSed) stratified by filter type (MF (F1–3avg) and UF (F4)). Lastly, to assess the impact of solutions on backwashed filters, a Kruskal–Wallis test was used to compare the endline indicators between solution types for each influent water condition (Bacteria, BacChem, and BacSed), stratified by filter type (MF (F1–3avg) and UF (F4)). For all statistical tests, p < 0.05 was significant.
4. Summary and Recommendations
The effect of different influent water conditions and backwashing solutions was tested on fouling development and control for different HFMF types. Our results confirmed that fouling developed rapidly in HFMFs used with different influent waters and highlighted membrane fouling development and control depend on HFMF type, influent water characteristics, completion of backwashing, and backwashing solution type. A complete summary of fouling and cleaning consequences by assessed indicators, membrane types, influent water conditions, and cleaning solutions is provided in
Figure 5.
In this study, fouling rapidly decreased water quantity performance and mechanical properties (particularly, strain at break) but did not decrease water quality performance or cause fiber breakage. The lack of change in E. coli LRV and turbidity removal were attributed to not having fiber failures during the short study, regular cleaning with low backwash pressure, and/or unknown factors. Additionally, while all influent water led to flow reductions and changes in strain at break and break force across all filters, MF membranes were generally more impacted by fouling than UF membranes, and BacChem influent water (including bacteria, humic acid, and calcium) generally had more fouling consequences.
For fouling control, the results indicate the importance of backwashing to recover water quantity performance and mechanical properties, especially for MF membranes. Backwashing depends on frequency and pressure (mainly removing surface deposits rather than internal foulants [
22,
40]), and, in this study, backwashing once every 17.5 h, with 5 psi pressure for 5 min, was insufficient to fully recover water quantity performance and the fiber’s mechanical properties. Optimizing these parameters may lead to better results, although the severity of fouling also depends on the concentration and nature of the membrane; complex regimes may not be possible for users to implement. Importantly, for the development of field-based cleaning recommendations, cleaning efficacy with the manufacturers’ recommended solutions (bleach, vinegar, and water) depended on influent water quality and membrane type, and there was no universal cleaning method that was appropriate for all fibers and influent water qualities. Water quantity performance recovery was generally better with bleach than other cleaning methods for MF membranes for all influent waters, but 0.5% sodium hypochlorite bleach possibly degraded the UF membrane’s mechanical properties. For all HFMFs, 6% acetic acid vinegar recovered water quantity performance and mechanical properties when the influent water contained bacteria only but damaged fibers in the presence of sediment. Lastly, water was an acceptable alternative to backwash UF membranes to recover mechanical properties when fibers were fouled with influent water containing humic acid, calcium, and/or sediments.
The limitations of this work include: (1) filtration and backwashing were performed on small modules with a few membrane fibers instead of full-size filters; (2) in this 10-day study, E. coli breakthrough was not reached, and, therefore, the research could not determine whether fouling development or backwashing was the reason for E. coli breakthroughs seen in the field; (3) a constant low-pressure backwash was applied to constantly wetted fibers, which may not mimic field conditions where pressure could be variable and fibers could dry out; and, (4) some orange deposition was noted in vinegar-cleaned modules with BacSed influent water (potentially iron deposition, but more investigation is needed). Additionally, some parameters (including fiber material properties) were outside the scope of this study and could be responsible for the differences between membrane types and influent waters observed after filtration and backwash. Therefore, detailed mechanisms behind the observations could be further investigated. One reason why these parameters were not investigated is that companies manufacturing HFMFs for sale in LMICs and humanitarian contexts change the membranes inside their cartridges without notice; hence, this information may not be clear. Despite these limitations, the results from this large exploratory study show that fouling develops rapidly (<200 h of use) on fibers, depending on HFMF type and influent water quality, and that there is no one-size-fits-all backwashing solution. Further laboratory testing should include confirmatory testing of full-scale filters to confirm the results presented herein.
This work highlights the need for future research on fouling in HFMF filters for HWT. Additional laboratory work should focus on expanding testing with other field-relevant influent water qualities but also to test additional cleaning solutions (including other available and easy-to-use concentrations) and backwashing solution combinations (including soaking and the consecutive use of two different cleaning solutions, which could be more effective, although more difficult for the user) [
22,
41,
42]. Reaching
E. coli breakthrough by changing filtration and backwash conditions should also be considered (including the timeframe of experiment and backwashing pressure), as well as the performance when fibers continually go through wet–dry phases. Lastly, completing field studies seem necessary to determine if cleaning methods can be adopted but also to measure water quality and fouling indicators in collaboration with users over time. Results could help develop end-of-life indicators for HFMFs.
To conclude, based on these exploratory results that provide unique quantitative information on fouling and cleaning mechanisms on HFMFs in LMICs and humanitarian contexts, implementers of HFMFs should understand that all HFMFs will foul, and there are no one-size-fits-all cleaning mechanisms to prevent fouling in HFMFs. Thus, wide-scale distribution of HFMFs for long-term use in contexts where influent water quality is unknown cannot be recommended. This study highlights the importance of characterizing the water composition of the most probable water sources for each context HFMFs will be distributed. This would lead to a better selection of HFMF type and guide membrane cleaning procedures during HFMF distribution programs. Alternatively, organizations distributing HWT options could also consider alternate options, such as chlorination, other filtration options (e.g., ceramic filters), and/or solar disinfection [
43]. As also promoted by some commercial suppliers [
44], HFMFs can also be distributed as temporary filters, intended to be replaced after three months of use, before
E. coli breakthrough [
13], or until we know the reasons for breakthrough and how to prevent it.