Locally manufactured ceramic filters can improve drinking water quality and reduce diarrheal disease burden in developing countries; however, production methods and quality control protocols vary at the >50 factories. We manufactured filter disks with varied clay, burn-out material, burn-out material sieved with different mesh sizes, and burn-out material to clay ratios and calculated filter characteristics, including porosity, density, shrinkage, and flow rate. Water was run through filters daily for 4 weeks, and flow rate and Escherichia coli reduction, as measured by log reduction value (LRV), were tested twice weekly. Our results suggest: (1) the first and last LRV test results do not correlate strongly (R2 = 0.38, p < 0.010); (2) there is not a strong association between flow rate and first, average, or last LRV results (R2 = 0.17, p = 0.090; R2 = 0.30, p = 0.020; R2 = 0.24, p = 0.040); and (3) first and average LRV are associated with burn-out material (R2 = 0.68, p < 0.001; R2 = 0.60, p < 0.001), and last LRV is associated with burn-out material and mesh size (R2 = 0.54, p < 0.050). Recommendations for filter factories, are to: (1) verify filtration efficacy with repeated bacteria reduction tests when materials, processing, or filter characteristics vary; (2) carefully control production variables; and (3) continue flow rate testing each filter to evaluate within and across batch production consistency.

INTRODUCTION

Worldwide, approximately 663 million people lack access to an improved water source and an estimated 1.2 billion more drink water at elevated risk of contamination at the source or during transport and storage (Onda et al. 2012; WHO/UNICEF 2015). Household water treatment (HWT) can be a cost effective means of improving drinking water quality (Clasen et al. 2007a) and reducing diarrheal disease in households that do not yet have access to water and sanitation infrastructure (Fewtrell et al. 2005; Clasen et al. 2007b; Clasen 2015). Locally manufactured ceramic ‘pot’ water filters (CWFs) are one promising HWT technology. CWFs are comprised of an ∼10 L capacity, silver-treated ceramic filter element that is suspended in a lidded receptacle. Water is poured into the filter, flows via gravity into a safe storage container, and is dispensed through a tap.

In laboratory investigations, CWFs have demonstrated >2 log reduction value (LRV, 99%) of protozoa and protozoan-sized particles (Lantagne 2001; van Halem et al. 2007) and 1–7 LRV (90–99.99999%) of bacterial organisms from drinking water (Lantagne 2001; van Halem et al. 2007; Brown & Sobsey 2010; Matthies et al. 2015). Virus reduction remains a challenge, with results ranging from <1–3 LRV (63–99.9%) (van Halem et al. 2007; Brown & Sobsey 2010; Matthies et al. 2015). In the field, water treated by CWFs is often improved to the World Health Organization's (WHO) low-risk classification (Roberts 2004; Brown et al. 2008) of <10 CFU Escherichia coli /100 mL (WHO 1997), and filter use has been associated with 49–80% reduction in diarrheal disease among users (Brown et al. 2008; Abebe et al. 2014).

Filters are typically manufactured by pressing a pre-determined ratio of locally-sourced clay and burn-out material, such as sawdust or rice husk, into the filter shape. After filters are pressed, dried, and fired to a ceramic state (∼800–900 °C) they are flow rate tested for quality control. Flow rate is defined as the volume of water that flows through a water-saturated filter after 1 hr. Filters that meet factory established flow rate criteria are coated with silver, a known antimicrobial agent, and packaged for sale or distribution.

The filter mixture ratio is determined by manufacturing batches of filters with different clay to burn-out material ratios. The filters are tested for flow rate and filters from the batch that meet the factory-established flow rate range (typically 1–3 L/hr) are tested for bacteria reduction. The specific recipe of the batch that meets both flow rate and bacteria reduction criteria is then used in production. Since it is not practical to test every filter for bacteria reduction, flow rate is used as the ongoing indicator of production consistency (CMWG 2011). Minimum flow rate is important to verify minimum rate of water treatment to meet a household's drinking water needs. Maximum flow rate is a topic of much research and debate (CMWG 2011).

Flow rate is affected by different input materials, such as the amount of burn-out material included in the filter mixture (Bloem 2008; Oyanedel-Craver & Smith 2008; Lantagne et al. 2010; Kallman et al. 2011; Yakub et al. 2012) and the type of burn-out material used (Lantagne et al. 2010). Flow rate is widely assumed to be an indicator of bacteria reduction efficiency. However, research on the relationship between flow rate and bacteria reduction has arrived at contradictory conclusions. Lantagne et al. (2010) found that total coliform reduction decreased to <2 LRV (<99%) when flow rates were >1.7 L/hr. Filters in this study had flow rates ranging from 0.25–10.2 L/hr and were manufactured with 40–60%, by volume, sawdust, coffee husk, or rice husk, processed through US #50 mesh sieves (0.30 mm openings) (Lantagne et al. 2010). Silver was not applied. In contrast, Bloem (2008) found no significant difference in E. coli LRV, independent of silver nitrate application, in filters with initial flow rates ranging from 1.7–7.6 L/hr. These filters were manufactured with 19–27% rice husk (mesh not specified), 5–17% laterite, and 64–74% clay, by weight. van der Laan et al. (2014) concluded that, in filters made with 24–31% rice husk processed through US #18 mesh sieves (1 mm openings), 2% laterite, and 67–74% clay, by weight, with initial flow rates ranging from 2.2–19.1 L/hr, the majority of bacteria reduction does not occur during filtration, but rather, due to contact time with silver nitrate during storage. LRVs of samples with <5 minutes storage time; however, ranged from 0.6–3.1, suggesting variability in the filtration ability of the filter material.

Previous research has also investigated relationships between input materials and bacteria reduction, and found that filter material produced with a fine-grained clay with uniform particle size distribution resulted in better bacteria reduction than material manufactured with clays containing larger, non-uniform particles (Oyanedel-Craver & Smith 2008). Kallman et al. (2011) suggest that an increased proportion of sawdust (4–17%, US #60 mesh, without silver) leads to lower bacteria reduction, yet a recent study found no correlation (R = −0.06) between LRV and rice husk proportion (24–31%, US #18 mesh, 2% laterite, 67–74% clay) (Soppe et al. 2015).

Lastly, silver nanoparticle application has been shown to improve bacteria reduction during filtration (Oyanedel-Craver & Smith 2008) and recent studies have found that the rate of silver elution varies depending upon the type of silver (silver nanoparticles or silver nitrate) and influent water quality characteristics (Rayner et al. 2013b; Mittelman et al. 2015). Since we are unable to predict silver longevity in filters, the filtration ability of the filter material should be relied on as the primary means of bacteria reduction.

In summary, manufacturing recommendations are limited by these seemingly contradictory research results. A 2009 survey of filter manufacturers found that manufacturing and quality control protocols vary widely both between and within factories, including: (1) criteria for modifying filter mixture ratio; (2) flow rate criteria and test protocols; (3) amount of silver application; and (4) bacteria reduction test protocols (Rayner et al. 2013a). Since 2009, the number of factories has grown from 35 to 50, and continued growth is anticipated. Factory visits and water quality testing in households suggest manufacturing variability may have resulted in poor quality filters reaching the market (Rayner et al. 2016).

The WHO has recently launched an international scheme to evaluate HWT products to support member states in product selection; however, the possible variability both across and within factories poses a challenge in the selection of representative filters for testing. To address this, the Ceramics Manufacturing Working Group (CMWG) has committed to developing a certification process to support factories in developing and maintaining effective quality assurance and quality control processes. A better understanding of the influence of input materials on filter characteristics and bacteria reduction is needed in order to further guide manufacturing recommendations, support factories in consistently manufacturing high quality filters, and responsibly promote decentralized CWF production.

We carried out an exploratory investigation into the influence of input materials on filter characteristics, flow rate, and LRV by: (1) manufacturing filter disks as surrogates for full-sized filters with varied input materials (clay source, burn-out type, burn-out mesh size, percent burn-out material); (2) calculating filter characteristics measurable at the factory level (porosity, density, shrinkage, and flow rate); and (3) testing disks for flow rate and LRV. Disks were tested without silver to evaluate filtration ability of the filter material. The main aims of this study were to evaluate: (1) the effects of input materials on filter characteristics; (2) the effects of input materials on LRV; and (3) the relationship between filter characteristics, including flow rate, and LRV.

METHODS

Batches of six, 10-cm diameter disks were manufactured at Advanced Ceramics Manufacturing (ACM, Tucson, AZ, USA) with clay imported from filter factories in Indonesia (Pelita Indonesia), Tanzania (SafeWAterNow), and Nicaragua (Filtrón). Burn-out materials, sawdust and rice husks, were purchased locally in Tucson, AZ. Burn-out material was processed in a Nutrimill® grain mill (Hampton, NE, USA) and sieved with US #8/16, 16/30, or 30/60 mesh screens (2.38/1.19, 1.19/0.60, and 0.60/0.25 mm openings, respectively). The material that passed through the larger mesh (smaller number) and was retained on the smaller mesh screen (larger number) was used in the filter mixture. Percent burn-out material for initial disk recipes was selected based on factory reported filter mixture ratios (Rayner et al. 2013a), then iteratively for subsequent recipes based on LRVs with the objective of meeting >2 LRV guideline value (CMWG 2011; WHO 2011).

Dry materials were measured by weight and mixed in a KitchenAid® mixer (Greenville, OH, USA) and, after adding water, further mixed until visually homogeneous. Filter material was hydraulically pressed into a disk shaped metal die with 4,500 lbs of force (2 × 104 N) to simulate pressure reportedly applied to press full-sized filters (CMWG 2011). Disks were air-dried and fired in an electric furnace. Target fired disk thickness was 1.8 cm. Firing profiles followed Best Practice Recommendations (CMWG 2011) and peak firing temperatures and soak times were established for each clay source to achieve sufficient strength to mount the disks for testing. The absence of a black core, indicative of incomplete firing, was confirmed at the selected profiles.

The height, diameter, and weight of each disk were measured after pressing and after firing. After firing, disks were saturated, by boiling in water for 1 hr, and weighed before and after saturation. Porosity was calculated by dividing the difference in weight by the geometric disk volume. Density was calculated by dividing the fired disk weight by the geometric disk volume. Shrinkage in diameter and height were calculated as the percent change between freshly pressed and fired disk diameters, and heights, respectively.

Three disks from each of 25 batches were shipped to Lehigh University (Bethlehem, PA). Disks were tested for pH, flow rate, and LRV eight times over 4 weeks. Disks were soaked in de-chlorinated tap water for 24 hr, attached to a PVC connector with plumber's putty, and then to a 7.62 cm (3-inch) diameter PVC column with a flexible coupling. Between test days, columns were filled with de-chlorinated tap water several times daily to maintain saturation and to flush out E. coli remaining from the previous test both to prevent clogging and to allow for more accurate performance evaluation. A sterile inoculating loop was dipped into filtered water on non-test days, placed in sterile LB broth, and incubated for 24 hr at 38.5 °C to check for presence of E. coli (detection level >15 CFU/100 mL).

Influent and effluent pH were measured using an OAKTON® pH/CON 510 benchtop meter calibrated with 4, 7, and 10 standards. Flow rates were calculated by measuring the volume of water that filtered from a saturated disk with a 24 cm falling head after 30–60 min and are presented in mL/hr. Testing was discontinued if flow rates were <10 mL/hr or >1,500 mL/hr.

To test LRV, E. coli was grown in LB broth to ∼1011 CFU/100 mL, determined by spectrometer reading at a wavelength of 600 nm, confirmed by plate counts, and diluted to 1.1 × 107 CFU/100 mL with de-chlorinated tap water. Columns were filled to a 24 cm head (∼2 L) with E. coli-spiked water twice per week. Influent and effluent samples were collected 1–2 hr after filling columns, diluted appropriately, tested by membrane filtration, and incubated on m-FC broth media at 44.5 °C ± 0.5 °C for 22–26 hr. Standard Methods were followed (APHA/AWWA/WEF 2012). LRV was calculated as the difference between the log10 of the influent and the log10 of the effluent E. coli CFU/100 mL.

Single and multivariable linear regression analyses were run using Stata 10.1 (College Station, TX, USA) to identify relationships between: (1) input materials (clay source, burn-out type, burn-out mesh size, and percent burn-out material) and disk characteristics (porosity, density, shrinkage, and flow rate); (2) input materials and LRV; and (3) disk characteristics and LRV. As flow rates and LRV varied, the three-disk average of the first, average of the tests, and last (8th) results were used. Variables were considered significant at p < 0.050; insignificant variables were removed from models. The half-way point between larger and smaller mesh openings is linear, therefore mesh sizes were coded as 30/60 = 1, 16/30 = 2, and 8/16 = 3 for regression analysis.

RESULTS

Filter disks were manufactured from filter material containing clay from Nicaragua, Tanzania, or Indonesia and 7.5%–25% of either sawdust or rice husk (by weight to clay) processed using 8/16, 16/30, or 30/60 mesh screens. Peak firing temperature varied per clay. The firing temperature and soak time selected for the Nicaraguan clay was 1,085 °C/60 min; the Tanzanian clay, 950 °C/60 min; and the Indonesian clay, 800 °C/180 min. The high temperature required to sinter the Nicaraguan disks resulted in over-fired, non-representative filter material and therefore Nicaraguan clay disks were dropped from the study.

In total, 25 sets of disks manufactured with Tanzanian or Indonesian clay were sent to Lehigh University for testing. The porosity and density of Tanzanian clay disks were similar after firing to 800 °C/180 min or 950 °C/60 min, but after being fired to 950 °C/60 min, the flow rates tripled. Five sets of disks manufactured with Tanzanian clay were shipped for testing but only one set completed testing due to the excessive flow rates.

Of the 20 disk sets manufactured with Indonesian clay 18 completed testing, as two sets did not filter enough water for testing (Table 1). Nine sets were manufactured with sawdust and nine with rice husk. Burn-out material to clay ranged from 11%–24% sawdust and 7.5%–25% rice husk, by weight. Fired disk thickness averaged 1.87 cm (min-max: 1.78–2.0 cm).

Table 1

Input materials, filter characteristics, and LRV

Clay sourceBurn-out typeMesh sizePercent burn-outDensity (g/cc)
AVG (SE)
Porosity (%) AVG (SE)Shrinkage (D) (%) AVG (SE)Shrinkage (H) (%) AVG (SE)Flow (mL/hr) AVG (SE)LRV AVG (SE)
Indonesia Sawdust 8/16 11.0% 1.18 (0.004) 55.5 (0.001) 11.7% (0.000) 15.4% (0.002) 87 (7.4) 1.79 (0.131) 
13.7% 1.13 (0.008) 56.3 (0.004) 11.3% (0.002) 15.6% (0.004) 70 (5.2) 1.87 (0.261) 
16/30 13.7% 1.16 (0.011) 53.2 (0.002) 11.8% (0.002) 17.3% (0.006) 27 (1.5) 4.43 (0.402) 
17.0% 1.04 (0.012) 56.2 (0.004) 10.8% (0.003) 15.1% (0.006) 100 (4.7) 2.37 (0.239) 
20.0% 1.00 (0.005) 60.1 (0.004) 10.5% (0.002) 15.6% (0.002) 126 (4.8) 2.83 (0.265) 
30/60 13.7% 1.17 (0.014) 50.4 (0.220) 11.7% (0.002) 18.7% (0.005) 17 (1.7) 2.06 (1.330) 
17.0% 1.11 (0.001) 52.7 (0.009) 10.9% (0.000) 18.4% (0.003) 17 (0.8) 4.00 (0.285) 
20.0% 1.05 (0.003) 57.0 (0.005) 11.3% (0.002) 19.4% (0.005) 27 (0.7) 3.41 (0.232) 
24.0% 0.95 (0.003) 64.6 (0.003) 10.8% (0.002) 17.4% (0.001) 172 (19.0) 2.78 (0.156) 
Rice husk 8/16 10.0% 1.30 (0.006) 48.0 (0.005) 9.3% (0.000) 12.1% (0.002) 387 (25.2) 0.98 (0.136) 
17.0% 1.12 (0.013) 53.8 (0.006) 8.4% (0.001) 13.7% (0.010) 671 (34.0) 0.86 (0.048) 
25.0% 0.91 (0.003) 60.1 (0.007) 7.7% (0.001) 12.9% (0.004) 1,252 (127.8) 0.89 (0.075) 
16/30 7.5% 1.41 (0.007) 47.9 (0.000) 11.8% (0.002) 14.5% (0.001) 35 (4.5) 1.28 (0.105) 
12.0% 1.28 (0.004) 51.6 (0.003) 10.5% (0.003) 13.9% (0.002) 175 (14.4) 1.47 (0.118) 
17.0% 1.14 (0.005) 59.0 (0.046) 10.6% (0.001) 14.3% (0.002) 283 (14.5) 0.96 (0.079) 
30/60 a10.0% 1.43 (0.008) 39.5 (0.019) 11.9% (0.002) 15.5% (0.006) 3.4 (0.7)b b 
a17.0% 1.36 (0.002) 46.9 (0.004) 11.8% (0.004) 15.2% (0.009) 7.5 (0.9)b 2.78 (0.229)b 
18.0% 1.18 (0.004) 56.2 (0.002) 10.0% (0.002) 14.7% (0.007) 171 (10.9) 1.93 (0.110) 
19.0% 1.16 (0.012) 56.5 (0.003) 10.8% (0.001) 14.6% (0.006) 180 (10.8) 1.26 (0.166) 
25.0% 1.02 (0.007) 58.7 (0.007) 9.9% (0.001) 13.5% (0.006) 374 (22.4) 1.26 (0.097) 
Tanzania Sawdust 8/16 a13.7% 1.03 (0.005)c 57.8 (0.007)c 5.0% (0.001)c 3.4% (0.002)c 3,030 (n/a)c b 
16/30 a13.7% 1.11 (0.000)c 57.6 (0.004)c 7.0% (0.002c 8.3% (0.002)c 3,195 (285.0)c b 
  1.11 (0.000) 57.3 (0.002) 6.2% (0.001) 8.0% (0.003) 805.7 (81.7) b 
30/60 a11.2% 1.36 (0.005)c 51.2 (0.003)c 7.7% (0.004)c 8.6% (0.001)c 525 (36.6)c 1.89 (0.159)c 
a20.0% 0.98 (0.003)c 63.0 (0.001)c 7.6% (0.001c 10.9% (0.005)c 1,925 (152.3)c b 
  0.97 (n/a) 62.9 (n/a) 6.7% (0.001) 9.6% (0.005) 921.2 (n/a) b 
Rice husk 30/60 a19.0% 1.16 (0.004)c 55.1 (0.000)c 5.7% (0.01)c 5.5% (0.001)c 2,060 (102.5)c b 
Clay sourceBurn-out typeMesh sizePercent burn-outDensity (g/cc)
AVG (SE)
Porosity (%) AVG (SE)Shrinkage (D) (%) AVG (SE)Shrinkage (H) (%) AVG (SE)Flow (mL/hr) AVG (SE)LRV AVG (SE)
Indonesia Sawdust 8/16 11.0% 1.18 (0.004) 55.5 (0.001) 11.7% (0.000) 15.4% (0.002) 87 (7.4) 1.79 (0.131) 
13.7% 1.13 (0.008) 56.3 (0.004) 11.3% (0.002) 15.6% (0.004) 70 (5.2) 1.87 (0.261) 
16/30 13.7% 1.16 (0.011) 53.2 (0.002) 11.8% (0.002) 17.3% (0.006) 27 (1.5) 4.43 (0.402) 
17.0% 1.04 (0.012) 56.2 (0.004) 10.8% (0.003) 15.1% (0.006) 100 (4.7) 2.37 (0.239) 
20.0% 1.00 (0.005) 60.1 (0.004) 10.5% (0.002) 15.6% (0.002) 126 (4.8) 2.83 (0.265) 
30/60 13.7% 1.17 (0.014) 50.4 (0.220) 11.7% (0.002) 18.7% (0.005) 17 (1.7) 2.06 (1.330) 
17.0% 1.11 (0.001) 52.7 (0.009) 10.9% (0.000) 18.4% (0.003) 17 (0.8) 4.00 (0.285) 
20.0% 1.05 (0.003) 57.0 (0.005) 11.3% (0.002) 19.4% (0.005) 27 (0.7) 3.41 (0.232) 
24.0% 0.95 (0.003) 64.6 (0.003) 10.8% (0.002) 17.4% (0.001) 172 (19.0) 2.78 (0.156) 
Rice husk 8/16 10.0% 1.30 (0.006) 48.0 (0.005) 9.3% (0.000) 12.1% (0.002) 387 (25.2) 0.98 (0.136) 
17.0% 1.12 (0.013) 53.8 (0.006) 8.4% (0.001) 13.7% (0.010) 671 (34.0) 0.86 (0.048) 
25.0% 0.91 (0.003) 60.1 (0.007) 7.7% (0.001) 12.9% (0.004) 1,252 (127.8) 0.89 (0.075) 
16/30 7.5% 1.41 (0.007) 47.9 (0.000) 11.8% (0.002) 14.5% (0.001) 35 (4.5) 1.28 (0.105) 
12.0% 1.28 (0.004) 51.6 (0.003) 10.5% (0.003) 13.9% (0.002) 175 (14.4) 1.47 (0.118) 
17.0% 1.14 (0.005) 59.0 (0.046) 10.6% (0.001) 14.3% (0.002) 283 (14.5) 0.96 (0.079) 
30/60 a10.0% 1.43 (0.008) 39.5 (0.019) 11.9% (0.002) 15.5% (0.006) 3.4 (0.7)b b 
a17.0% 1.36 (0.002) 46.9 (0.004) 11.8% (0.004) 15.2% (0.009) 7.5 (0.9)b 2.78 (0.229)b 
18.0% 1.18 (0.004) 56.2 (0.002) 10.0% (0.002) 14.7% (0.007) 171 (10.9) 1.93 (0.110) 
19.0% 1.16 (0.012) 56.5 (0.003) 10.8% (0.001) 14.6% (0.006) 180 (10.8) 1.26 (0.166) 
25.0% 1.02 (0.007) 58.7 (0.007) 9.9% (0.001) 13.5% (0.006) 374 (22.4) 1.26 (0.097) 
Tanzania Sawdust 8/16 a13.7% 1.03 (0.005)c 57.8 (0.007)c 5.0% (0.001)c 3.4% (0.002)c 3,030 (n/a)c b 
16/30 a13.7% 1.11 (0.000)c 57.6 (0.004)c 7.0% (0.002c 8.3% (0.002)c 3,195 (285.0)c b 
  1.11 (0.000) 57.3 (0.002) 6.2% (0.001) 8.0% (0.003) 805.7 (81.7) b 
30/60 a11.2% 1.36 (0.005)c 51.2 (0.003)c 7.7% (0.004)c 8.6% (0.001)c 525 (36.6)c 1.89 (0.159)c 
a20.0% 0.98 (0.003)c 63.0 (0.001)c 7.6% (0.001c 10.9% (0.005)c 1,925 (152.3)c b 
  0.97 (n/a) 62.9 (n/a) 6.7% (0.001) 9.6% (0.005) 921.2 (n/a) b 
Rice husk 30/60 a19.0% 1.16 (0.004)c 55.1 (0.000)c 5.7% (0.01)c 5.5% (0.001)c 2,060 (102.5)c b 

SE = Standard error.

aNot included in regression analysis.

bDisk set did not complete testing due to too low or excessive flow rates, or fragility.

cMeasurements taken after 950 °C/60min firing.

Input materials, filter characteristics and LRVs are presented for the 25 disk sets shipped to Lehigh in Table 1. Both influent and effluent pH averaged 6.5 (ranging from 6.0–6.8 and 5.9–7.3, respectively). Please note only the 18 disk sets manufactured with Indonesian clay that completed flow rate and LRV testing were included in the regression analysis.

Disk characteristics

Density and porosity

The average density of disks manufactured with Indonesian clay (n = 18) ranged from 0.91–1.41 g/cc (mean 1.13 g/cc) and average porosity ranged from 47.9–64.6% (mean 55.4%) (Table 1). Standard error within disk sets ranged from <0.001–0.014 for density and <0.001–0.046 for porosity. Density was greater in disks made with rice husks, with smaller burn-out mesh, and lower percent burn-out material (R2 = 0.99, p < 0.001) (Table 2). Density and porosity had a strong inverse correlation (R2 = 0.73, p < 0.001).

Table 2

Linear regression model results

NR2Outcome: filter characteristicsConstantInput materials
18 0.70 Porosity 43.332*** 0.725***(ratio)   
18 0.99 Density 1.484*** −0.025***(ratio) 0.092***(burnout) −0.042***(size) 
18 0.79 Shrinkage (D) 0.161*** −0.001***(ratio) −0.012***(burnout) −0.007**(size) 
18 0.83 Shrinkage (H) 0.219*** −0.029***(burnout) −0.011***(size)  
18 0.69 First Flow Rate −2,071*** 346.168*(burnout) 389.450**(size) 67.252**(ratio) 
18 0.84 Average Flow Rate −1,300*** 265.438***(burnout) 238.889***(size) 41.554***(ratio) 
18 0.88 Last Flow Rate −960.44*** 221.610***(burnout) 185.463***(size) 28.592***(ratio) 
NR2Outcome: LRVConstantInput materials
18 0.68 First LRV 6.717*** −2.534***(burnout)   
18 0.60 Average LRV 4.47*** −1.63***(burnout)   
18 0.54 Last LRV 3.852*** −0.725*(burnout) −0.598**(size)  
NR2Outcome: flow rate, LRVConstantFilter characteristics
18 0.65 First Flow Rate 4,148.38*** −36,410.36***(Shrinkage(D))   
18 0.82 Average Flow Rate 2,781.49*** −24,165.08***(Shrinkage(D))   
18 0.87 Last Flow Rate 2,159.90*** −18,576.83***(Shrinkage(D))   
18 0.54 First LRV −5.735* 56.218***(Shrinkage(H))   
18 0.65 Average LRV −4.522** 42.546*** (Shrinkage(H))   
18 0.59 Last LRV −3.516** 33.472***(Shrinkage(H))   
NR2Outcome: filter characteristicsConstantInput materials
18 0.70 Porosity 43.332*** 0.725***(ratio)   
18 0.99 Density 1.484*** −0.025***(ratio) 0.092***(burnout) −0.042***(size) 
18 0.79 Shrinkage (D) 0.161*** −0.001***(ratio) −0.012***(burnout) −0.007**(size) 
18 0.83 Shrinkage (H) 0.219*** −0.029***(burnout) −0.011***(size)  
18 0.69 First Flow Rate −2,071*** 346.168*(burnout) 389.450**(size) 67.252**(ratio) 
18 0.84 Average Flow Rate −1,300*** 265.438***(burnout) 238.889***(size) 41.554***(ratio) 
18 0.88 Last Flow Rate −960.44*** 221.610***(burnout) 185.463***(size) 28.592***(ratio) 
NR2Outcome: LRVConstantInput materials
18 0.68 First LRV 6.717*** −2.534***(burnout)   
18 0.60 Average LRV 4.47*** −1.63***(burnout)   
18 0.54 Last LRV 3.852*** −0.725*(burnout) −0.598**(size)  
NR2Outcome: flow rate, LRVConstantFilter characteristics
18 0.65 First Flow Rate 4,148.38*** −36,410.36***(Shrinkage(D))   
18 0.82 Average Flow Rate 2,781.49*** −24,165.08***(Shrinkage(D))   
18 0.87 Last Flow Rate 2,159.90*** −18,576.83***(Shrinkage(D))   
18 0.54 First LRV −5.735* 56.218***(Shrinkage(H))   
18 0.65 Average LRV −4.522** 42.546*** (Shrinkage(H))   
18 0.59 Last LRV −3.516** 33.472***(Shrinkage(H))   

Ratio = percent burnout material to clay by weight. Burnout = burnout type, coded as: sawdust = 1, rice husk = 2. Size = US mesh number, coded as: 8/16 = 3, 16/30 = 2, 30/60 = 1. D = diameter, H = height.

*p < 0.050.

**p < 0.010.

***p < 0.001.

Shrinkage

Shrinkage in diameter ranged from 7.7%–11.8% with standard error within disk sets ranging from <0.001–0.003 (Table 1). Shrinkage in height ranged from 12.1%–19.4% and standard error ranged from 0.001–0.010. Shrinkage in diameter and height were moderately correlated (R2 = 0.42, p = 0.004, n = 18) (Table 2). This correlation was stronger for disks made with rice husks (R2 = 0.43, p = 0.05, n = 9), than for disks manufactured with sawdust (R2 = 0.01, p = 0.53, n = 9). Shrinkage in height was greater in disks made with sawdust, sieved with a smaller mesh (R2 = 0.83, p < 0.001).

Flow rate

The average flow rates of disk sets made with Indonesian clay (n = 18) ranged from 17–1,252 mL/hr (Table 1). Flow rates fluctuated over the 4 weeks of testing (Figure 1), however there was a strong correlation between first and last flow rates (R2 = 0.70, p < 0.001). Flow rates were faster in disks made with rice husks (n = 9), processed with a larger mesh, and with greater percent burn-out material (R2 = 0.69, p < 0.050; R2 = 0.84, p < 0.001; R2 = 0.88, p < 0.001) (Table 2 ). Flow rate declined with increased shrinkage in diameter (R2 = 0.65, p < 0.001; R2 = 0.82, p < 0.001; R2 = 0.87, p < 0.001).
Figure 1

Flow rates (dots) and LRV (bars) with standard error bars of Indonesian clay disks with (a) sawdust and (b) rice husk. Bold horizontal line indicates 2 LRV.

Figure 1

Flow rates (dots) and LRV (bars) with standard error bars of Indonesian clay disks with (a) sawdust and (b) rice husk. Bold horizontal line indicates 2 LRV.

E. coli reduction

LRV fluctuated throughout the 4 weeks of testing (Figure 1), and the association between first LRV and last LRV was weak (R2 = 0.38, p < 0.010, n = 18). LRVs in the first test ranged from 0.86–6.34 (mean 2.92), average LRVs ranged from 0.86–4.43 (mean 2.03), and the last LRVs ranged from 0.61–3.14 (mean 1.63).

For disks manufactured with sawdust (n = 9), first LRVs ranged from 2.4–6.34 (mean 4.18), average LRVs ranged from 1.79–4.43 (mean 2.84), and last test results ranged from 0.61–3.14 (mean 2.06) (Table 1). For disks manufactured with rice husks (n = 9), first LRVs ranged from 0.86–2.39 (mean 1.65), average LRVs ranged from 0.86–1.93 (mean 1.21), and last test results ranged from 0.67–2.26 (mean 1.21).

In the regression analysis, first and average LRVs were greater in disks made with sawdust (R2 = 0.68, 0.60, p < 0.001, respectively), and last LRVs were greater in disks made with sawdust using a smaller mesh (R2 = 0.54, p < 0.010). Shrinkage in height was associated with first, average, and last LRVs (R2 = 0.54, 0.65, 0.59, p < 0.001, respectively).

Single variable linear regression results for flow rate and LRVs for the 18 disk sets manufactured with Indonesian clay were: (1) first flow rate with first LRV: R2 = 0.17 (p = 0.090); (2) average flow rate with average LRV: R2 = 0.30 (p = 0.020); and (3) last flow rate with last LRV: R2 = 0.24 (p = 0.040).

Discussion

We manufactured 25 filter disk sets (without silver application) to evaluate relationships between input materials (clay source, burn-out type, mesh size, and percent burn-out material) and filter characteristics measurable at the factory level (density, porosity, shrinkage and flow rate) against filter quality criteria of LRV over 4 weeks of testing (8 tests). Investigating these relationships offers potential to better understand relationships between input materials, filter characteristics, and LRV, and is needed to guide manufacturing quality control recommendations.

Our primary findings include: (1) the first LRV test did not correlate strongly with the last test; (2) we did not find a strong association between flow rate and LRV for non-silver treated filter material; and (3) we did find an association between burn-out type and mesh size with LRV. These are discussed below, along with the effects of input materials on filter characteristics.

All disks manufactured with sawdust and some rice husk disks achieved >2LRV on the first test; overall, 8th LRVs were lower than initial results. While it appears that the decline in LRV over time was greater in disks manufactured with sawdust, the lower initial LRV in disks manufactured with rice husk possibly limited the extent of decline in LRV, this also may have diluted some of the regression results. LRV in sawdust disks generally appeared to stabilize after about four tests; however, it is not clear whether performance would be maintained if testing had continued. This suggests a need for further long-term performance studies and factories should consider this when designing their quality control protocols.

There was only a weak association between flow rate and LRV. Flow rate is a primary quality control evaluation carried out at most factories to evaluate production consistency, minimum flow rate criteria, and filter quality. Flow rate is affected by input materials including burn-out type, percent burn-out material, and mesh size. While in disks manufactured with either sawdust or rice husks flow rate increased with increased percent burn-out material or mesh size, rice husk disks demonstrated a steeper change in flow rate, suggesting, in terms of flow rate, control of percent rice husk and mesh size is more sensitive than for sawdust. The high aspect ratio (oblong shape) of rice husk may allow long, thin particles to pass through the sieve, possibly resulting in a lower threshold for connected pores and a less tortuous filtration matrix than with more spherical sawdust particles.

Neither porosity nor density were associated with LRV and while flow rates were faster, disks manufactured with rice husk were less porous (and more dense). Materials were measured by weight and as rice husk is more dense than sawdust, the smaller volume of rice husk added to the filter mixture resulted in less porous material. Furthermore, while porosity and density remained similar, flow rates nearly tripled in the two Tanzanian disk sets that were re-fired to a hotter temperature. This supports findings by Soppe et al. (2015) who also measured a change in flow rates with firing temperature. This suggests that while changes in filter characteristics such as porosity and density may detect variability in input materials, they are unlikely to capture effects of wide variation in firing temperature, which may also impact LRV (Soppe et al. 2015). Shrinkage in diameter was associated with flow rate and shrinkage in height was associated with LRV; however, shrinkage direction in full-sized filters might vary according to the angle of pressure during pressing. We conclude that while flow rate is not a reliable indicator for LRV, it should still be used as an indicator for production consistency (and to confirm minimum rate of water treatment), as it can capture variation both in input materials and firing temperature.

While burnout type, mesh size, and percent burn-out material influence flow rate, input materials most associated with LRV were burn-out type and mesh size. Factory-produced filters manufactured with rice husks have demonstrated effective bacteria reduction (with and without silver) (Brown & Sobsey 2010); however, disks made with rice husk in this investigation did not meet >2 LRV criteria throughout eight tests. Disks manufactured with sawdust performed better than disks manufactured with rice husks, and sawdust sieved with a smaller mesh was associated with better LRV. Theoretically, filter material of similar porosity but created by a greater number of smaller void spaces will have a greater surface area to trap bacteria than material with fewer large void spaces. While a smaller mesh was associated with better LRV in this study, there is likely a threshold for the proportion of sawdust that can be added before pores become interconnected and LRV decreases (Kallman et al. 2011). If the higher aspect ratio (oblong shape) of rice husk impacts the pore structure as discussed above, it could also impact LRV. LRV was associated with shrinkage in height, which was also associated with burn-out type and mesh size. Filter manufacturers have commented on a difference when working with hard versus soft woods, rice husk versus bran, and rice husk versus sawdust and it could be that burn-out material characteristics, such as absorbency, affect shrinkage and/or the resulting pore structure.

Based on the results of this research, recommendations to factories include: (1) verify filtration efficacy with repeated bacteria reduction tests when a new recipe is developed, or when there is variation in input materials, processing, or filter characteristics; (2) carefully control production variables, with special attention to sawdust particle size; and (3) continue flow rate testing filters to evaluate both within and across batch production consistency (and verify minimum flow rate requirements), and if flow rates are not consistent, identify and remedy the variation in production. We suggest quality assurance and quality control protocols address these recommendations to consistently produce high quality filters.

Our results are limited by: (1) the use of disks as surrogates for full sized filters; and (2) the small sample size available for statistical analysis. These are further discussed below.

The use of disks as surrogates for full-sized filters can improve research efficiency on CWFs; however, it would be useful to know if full-sized filters produced with the same input materials would meet the minimum flow rate criterion of ≥1 L/hr. The flow rate model presented in van Halem (2006) was used to project full-sized filter flow rates, and filter recipes were iteratively selected based on flow rate predictions. At the conclusion of this research, an error in the reported filter dimensions was identified, calling into question the empirical model validation. The analysis was therefore carried out on non-converted disk flow rates. Additional models have since been proposed (Plappally et al. 2009; Schweitzer et al. 2013) and are currently being evaluated empirically for use in predicting full-size filter flow rates.

The small sample size was primarily a result of the challenges of working with different clays, and remaining resources did not allow us to push input variables to reach thresholds. A post-study analysis of the sand-silt-clay distributions found the Indonesian clay had 50.5:18.5:31.0, the Tanzanian clay had 64:7.5:28.5, and the Nicaraguan clay had 82:17.5:0.5 (Duocastella & Morrill 2012). Variability in clay is a challenge faced by factories. When the Nicaraguan clay used in this research, which contained <1% clay-sized particles, reached production, the factory struggled to produce filters and suspended production until a new clay source was identified. Whenever a new clay source is introduced, factories need to establish a new recipe and test for bacteria reduction, which can be time consuming and costly.

Despite these limitations, we believe that our investigatory research results improve our knowledge of the relationships between input materials, filter characteristics, and quality criteria. We identified key areas for manufacturing recommendations at the factory level and believe this is a promising model to further evaluate effects of input materials on filter characteristics and quality. Further research is recommended to empirically validate flow rate prediction models, as the use of disks to model full-size filters can reduce cost and time of research and may also be useful to improve efficiency in initial prototype development at factories. Evaluation of pore network morphology and variation in pore structure created by different clay characteristics, burn-out material characteristics, burn-out material processing, and firing profile and temperature on the effects on bacteria reduction are needed.

Conclusions

With the growing global interest in scaling up CWF production, ensuring quality-controlled production in decentralized filter factories is of increasing interest. We manufactured 25 sets of disks with different input materials (clay source, burn-out type, percent burn-out material, and mesh size) and measured the disk characteristics (porosity, density, shrinkage, and flow rates) against LRV. We found that the first LRVs were not strongly correlated with the last (8th) LRVs, flow rate did not correlate strongly with LRV in non-silver treated filter material, and sawdust sieved with a smaller mesh was associated with better LRV compared with rice husk and a larger mesh.

Recommendations to factories include: verify bacteria reduction with repeated tests when a new recipe is developed or when there is variation in input materials, processing, or filter characteristics; carefully control production variables, with special attention to sawdust particle size; and continue flow rate testing all filters to evaluate both within and across batch production consistency and minimum flow rate requirements.

ACKNOWLEDGEMENTS

We would like to thank the filter factories that provided us with clay, Advanced Ceramics Manufacturing for manufacturing the disks used in this study, and PATH for supporting this work.

REFERENCES

REFERENCES
APHA/AWWA/WEF
2012
Standard Methods for the Examination of Water and Wastewater
.
American Public Health Association, American Water Works Association, and the Water Environment Federation
,
Washington, DC, USA
.
Bloem
S.
2008
Silver Impregnated Ceramic Water Filter: Flowrate Versus the Removal Efficiency of Pathogens
.
Delft University of Technology
,
Delft
,
Netherlands
.
Brown
J.
Sobsey
M. D.
Loomis
D.
2008
Local drinking water filters reduce diarrheal disease in Cambodia: a randomized, controlled trial of the ceramic water purifier
.
The American Journal of Tropical Medicine and Hygiene
79
(
3
),
394
400
.
Clasen
T.
Cairncross
S.
Haller
L.
Bartram
J.
Walker
D.
2007a
Cost-effectiveness of water quality interventions for preventing diarrhoeal disease in developing countries
.
Journal of Water and Health
5
(
4
),
599
608
.
CMWG
2011
Best Practice Recommendations for Local Manufacturing of Ceramic Pot Filters for Household Water Treatment
,
1st edn
.
CDC
,
Atlanta, GA
,
USA
.
Duocastella
M. d. M.
Morrill
K.
2012
Particle Size Distribution Analysis for Ceramic Pot Water Filter Production
.
Potters without Borders
,
Enderby, British Columbia
,
Canada
.
Fewtrell
L.
Kaufmann
R. B.
Kay
D.
Enanoria
W.
Haller
L.
Colford
J. M.
Jr.
2005
Water, sanitation, and hygiene interventions to reduce diarrhoea in less developed countries: a systematic review and meta-analysis
.
The Lancet Infectious Diseases
5
(
1
),
42
52
.
Kallman
E.
Oyanedel-Craver
V.
Smith
J.
2011
Ceramic filters impregnated with silver nanoparticles for point-of-use water treatment in rural Guatemala
.
Journal of Environmental Engineering-ASCE
137
(
6
),
407
415
.
Lantagne
D. S.
2001
Investigation of the Potters for Peace Colloidal Silver Impregnated Ceramic Filter Report 1: Intrinsic Effectiveness, Report 1
.
Alethia Environmental
,
Allston, MA
,
USA
.
Lantagne
D.
Klarman
M.
Mayer
A.
Preston
K.
Napotnik
J.
Jellison
K.
2010
Effect of production variables on microbiological removal in locally-produced ceramic filters for household water treatment
.
International Journal of Environmental Health Research
20
(
3
),
171
187
.
Matthies
K.
Bitter
H.
Deobald
N.
Heinle
M.
Diedel
R.
Obst
U.
Brenner-Weiss
G.
2015
Morphology, composition and performance of a ceramic filter for household water treatment in Indonesia
.
Water Practice & Technology
10
(
2
),
361
370
.
Mittelman
A. M.
Lantagne
D. S.
Rayner
J.
Pennell
K. D.
2015
Silver dissolution and release from ceramic water filters
.
Environmental Science and Technology
49
(
14
),
8515
8522
.
Onda
K.
LoBuglio
J.
Bartram
J.
2012
Global access to safe water: accounting for water quality and the resulting impact on MDG progress
.
International Journal of Environmental Research and Public Health
9
(
3
),
880
894
.
Oyanedel-Craver
V. A.
Smith
J. A.
2008
Sustainable colloidal-silver-impregnated ceramic filter for point-of-use water treatment
.
Environmental Science and Technology
42
(
3
),
927
933
.
Plappally
A.
Yakub
I.
Brown
L.
Soboyejo
W.
Soboyejo
A.
2009
Theoretical and experimental investigation of water flow through porous ceramic clay composite water filter
.
Fluid Dynamics and Material Processing
5
(
4
),
373
398
.
Rayner
J.
Skinner
B.
Lantagne
D.
2013a
Current practices in manufacturing locally-made ceramic pot filters for water treatment in developing countries
.
Journal of Water, Sanitation and Hygiene for Development
3
(
2
),
252
261
.
Rayner
J.
Zhang
H.
Schubert
J.
Lennon
P.
Lantagne
D.
Oyanedel-Craver
V.
2013b
Laboratory investigation into the effect of silver application on the bacterial removal efficacy of filter material for use on locally produced ceramic water filters for household drinking water treatment
.
ACS Sustainable Chemistry & Engineering
1
(
7
),
737
745
.
Rayner
J.
Murray
A.
Joseph
M.
Branz
A.
Lantagne
D.
2016
Evaluation of household drinking water filter distributions in Haiti
.
Journal of Water Sanitation and Hygiene for Development
6
(
1
),
42
54
.
Roberts
M.
2004
Field Test of A Silver-Impregnated Ceramic Water Filter
.
WEDC
,
Vientiane
,
Lao PDR
.
Schweitzer
R. W.
Cunningham
J. A.
Mihelcic
J. R.
2013
Hydraulic modeling of clay ceramic water filters for point-of-use water treatment
.
Environmental Science and Technology
47
(
1
),
429
435
.
Soppe
A.
Heijman
S.
Gensburger
I.
Shantz
A.
van Halem
D.
Kroesbergen
J.
Wubbels
G.
Smeets
P.
2015
Critical parameters in the production of ceramic pot filters for household water treatment in developing countries
.
Journal of Water and Health
13
(
2
),
587
599
.
van der Laan
H.
van Halem
D.
Smeets
P.
Soppe
A.
Kroesbergen
J.
Wubbels
G.
Nederstigt
J.
Gensburger
I.
Heijman
S.
2014
Bacteria and virus remzoval effectiveness of ceramic pot filters with different silver applications in a long term experiment
.
Water Research
51
,
47
54
.
van Halem
D.
2006
Ceramic Silver Impregnated Pot Filters for Household Drinking Water Treatment in Developing Countries
.
Delft University of Technology
,
Delft
.
WHO
1997
Guidelines for Drinking-water Quality, 2nd Edition: Volume 3; Surveillance and Control of Community Supplies
.
Geneva
,
Switzerland
.
WHO
2011
Evaluating Household Water Treatment Options: Health-based Targets and Microbiological Performance Specifications
.
WHO/UNICEF
2015
Progress on Sanitation and Drinking Water - 2015 Update and MDG Assessment
.
Yakub
I.
Plappally
A.
Leftwich
M.
Malatesta
K.
Friedman
K. C.
Obwoya
S.
Nyongesa
F.
Maiga
A. H.
Soboyejo
A. B.
Logothetis
S.
2012
Porosity, flow, and filtration characteristics of frustum-shaped ceramic water filters
.
Journal of Environmental Engineering
139
(
7
),
986
994
.

Author notes

Consultant for PATH during the research portion of this project