Design considerations for implementing ceramics in new and existing polymeric UF systems

The robustness of ceramic ultrafiltration and microfiltration (UF/MF) membranes has been widely demonstrated in a variety of process applications over the past 30 years (Bhave 1991). The barrier to their application in mainstream water treatment applications has been the substantial price premium of ceramic membranes (Tan 2010), in some cases approaching nearly an order of magnitude difference in price (Guerra & Pellegrino 2012). In recent years, ceramic membrane manufacturers have made significant improvements to this price difference, with improvements to both product design and production efficiencies. Ceramem (Goldsmith 1988) developed the concept of the higher surface area monolith, which dramatically increased the surface area of the membrane available per piece relative to the ceramic tubes that had previously found widespread adoption. This design was later improved for use in water treatment by Metawater (Hattori 2010). More recently, Nanostone Water has developed a high surface area ceramic UF monolith shown in Figure 1 that further reduces the ceramic UF membrane cost (Göbbert & Volz 2010) through the use of potted individual ceramic segments housed in a fiberglass reinforced plastic (FRP) membrane vessel, and reduced production costs compared to other ceramic membrane modules.

With the higher total suspended solids (TSS) tolerance of ceramic membranes and the ability to utilize more aggressive chemical and hydraulic cleaning methods, the risk of irreversible fouling is lower than with polymeric UF/MF membranes (Lee et al. 2013). In some cases this can lead to less pre-treatment, such as reduced need for a clarification system, which has a significant impact on the initial capital cost, operating cost, and footprint. As a result of their high permeability and hydrophilic surfaces, ceramic UF/MF membranes can typically operate at higher design fluxes than is typical for polymeric UF/MF membranes (Lerch et al. 2005; Loi-Brugger et al. 2006), which can financially offset some or all of the capital cost premium of the ceramic membrane modules.

As outlined above, with lower prices and a higher design flux on the ceramic membranes the capital cost of the membrane modules can be much more competitive than was previously possible. However, in a system-level comparison of polymeric and ceramics, the hydraulic cleaning methods can have a significant impact in the overall costs and need to be considered in a techno-economic comparison of the different systems.

For most industrial and small- to mid-size municipal drinking water and wastewater reuse projects, the pressurized hollow fiber UF/MF membrane module is a commonly applied technology (Adham 2005). Pressurized UF/MF modules use a range of membrane materials, typically polyvinylidene fluoride (PVDF) or polyethersulfone (PES), and configurations (inside/out or outside/in) (Pearce 2007). Most outside/in membranes operate in dead end mode with all water feeding the membrane and coming through as filtered water, while leaving suspended materials on the membrane surface. Periodic backwash cycles remove the retained suspended materials and divert that water to waste. This is often performed with an air scour to help break up the cake layer within the fibre bundle. Inside/out polymeric membranes typically rely on the backwash flow alone to remove suspended solids between filtration cycles.

Backwash flow rates vary between manufacturers, with some requiring as little as 10% of filtration flow while others require up to 400% of filtration flow. Feed flush is another option often used as part of the backwash sequence to help move suspended solids out of the membrane modules. Some suppliers use a drain down step as well. Backwash pumps are typically controlled with a variable frequency drive and are operated to provide a gentle increase in flow and pressure during backwash, due to the potential for fiber breakage or other damage to the polymeric hollow fiber UF/MF membranes. PVC piping systems are typically used to reduce capital costs.

Due to the higher capital costs of ceramic membranes, the flux is pushed higher through the use of aggressive hydraulic cleaning designs. Recirculation is common for ceramic membrane systems, especially for high TSS and oily waste applications. With these designs, there is typically a concentrate bleed rate that sets the recovery rate of the system. The crossflow velocity reduces the rate at which suspended solids build up on the membrane surface. Periodically, a backwash cycle is employed to reverse the direction of permeate flow while the crossflow stream continues to operate. The periodic backwash moves suspended material off the membrane surface where it is swept away by the high crossflow in the channels. This backwash method is often driven by compressed air, either directly or through a diaphragm, to deliver a fast step change increase in both pressure and flow for short durations (Hofs et al. 2011).

Typical crossflow velocities for ceramic UF/MF membranes is up to 4 m/s (Guerra & Pellegrino 2012). To achieve this velocity with relatively wide capillary channel diameters (2–3 mm), the recirculation rate is often 10 to 35 times higher than the filtration flow rate. System designs with such high crossflow rates typically have some membranes in series to minimize the associated pumping costs. For a system with a feed flow of 114 m³/hr producing 108 m³/hr of permeate, even with two modules in series a recirculation rate of 636–1,930 m³/hr is required to reach a crossflow velocity of 1–3 m/s. For small-scale systems with high fouling potential or high suspended solids the crossflow design can be effective. However, it is generally more expensive considering the additional crossflow pumps and associated energy costs.

With increasingly more competitive ceramic prices there are now references for ceramic membrane systems designed for dead end filtration for mainstream drinking water and wastewater reuse applications (Hattori 2010). Typically the process design is dead end filtration mode and the hydraulic cleaning method will employ a high pressure, high flow backwash for short durations, with compressed air as the motive force. In some cases the system designs also use compressed air to purge the dislodged solids from the capillary channels in a fast draining step. While these high pressure hydraulic cleaning methods are very effective for ceramic membranes, they do represent design and cost challenges to the overall system. Compared to a polymeric UF/MF system designed with nearly 100% plastic piping, a ceramic system using compressed air and high rate backwash would require a substantial amount of larger diameter steel piping and this can be a significant system level cost difference and increase in system complexity.

This work involved a series of experiments to evaluate different methods of hydraulic cleaning to optimize the balance between design flux and the hydraulic cleaning system expense considerations. This was conducted to determine whether there could be operational conditions where it would be both technically and financially feasible to replace polymeric hollow fibre modules with ceramic membrane by using hydraulic cleaning methods more similar to those found in polymeric systems. The experimental work is augmented by financial projections to highlight the effect these operation conditions have on plant economics.

The pilot system used a single ceramic test membrane from Nanostone Water with 3 m² of active surface area housed in a 100 mm diameter PVC vessel. This is a representative version of the full-scale module from Nanostone Water with 24.3 m² of active surface area housed in a FRP membrane vessel. In either configuration the membranes operate under pressure and utilize an ‘inside-out’ flow. The membrane used has a nominal rating of 30 nm (0.03 micron).

The pilot system includes controls and data logging to allow pilot conditions to be set ahead of an experiment and run conditions (flux set point, dead end filtration duration, feed dosing, backwash flush, backwash duration, feed flush flow rate, feed flush duration, and cleaning conditions and frequency) to be executed and data collected (flow rates, pressures, temperature, pH, and turbidity) without direct operator intervention.

In this mode 100% of the water feeding the membrane went through the membrane as permeate. All suspended solids in the water built up as a cake layer on the membrane surface. Periodically this water was backwashed (reverse direction) to clean the membrane.

To discharge solids from the pilot system there were periodic backwashes of permeate water in the reverse direction of filtration flow. This water traveled through the membrane surface from the clean side to the dirty side to remove suspended solids that had built up on the surface. This water was diverted to remove solids. No chemicals were used for backwashing.

To purge solids that had been lifted from the surface of the membrane during backwashing out of the membrane channels, feed water was directed through the channels to discharge remaining solids. During this step, permeate valving was closed to prevent re-deposition on the membrane surface during the feed flush.

Maintenance chemical cleaning was used periodically, typically with a frequency of 2–3 days. Maintenance cleaning was used to prolong production time between recovery cleanings. Sodium hypochlorite (NaOCl) was the standard cleaning agent, in some cases NaOH was used to adjust pH of the sodium hypochlorite solution. Maintenance cleaning was performed as a short automatic maintenance clean-in-place (mCIP) step where collected permeate was adjusted with cleaning chemicals and recirculated through the feed system and membrane followed by a short soak and discharge. Backwash and feed flush were performed before a maintenance cleaning to remove bulk solids, and after to flush cleaning chemicals from the system.

Recovery CIP was used to restore the permeability of the membranes by removing the foulants from the membrane surface that were not completely removed by the backwashes and maintenance cleanings. Typical cleaning was with NaOCl with NaOH (pH 12) followed by cleaning with citric acid. Both cycles use heat to raise the temperature to 10 °C above the operating temperature.

Each cycle began with a dead end filtration run, followed by a hydraulic cleaning step that included a backwash followed by a feed flush. As specified in a given test, a mCIP may be conducted during a test condition. At the conclusion of each test, a recovery CIP was conducted to restore initial performance. Times and conditions of each operation are included in the results section.

The pilot test location was a direct river water feed with a turbidity of 3–15 NTU and total organic carbon (TOC) values <4 mg/L. The same site feeds a polymeric hollow fibre UF system from which comparative data is drawn.

Test 1: The first test condition used a one-hour dead end filtration period with a flux set point of 170 lmh. 1 mg/L as Al⁺³ of polyaluminium chloride (PACl) was used as a coagulant, (the same coagulant conditions are used on the polymeric UF membrane system). The backwash method used a fast step change in pressure, using a permeate tank pressurized with compressed air to 5 bar for a duration of one second. The estimated backwash flow rate was 14 times higher than the filtration flow, or 2,400 lmh. Following the backwash was a feed flush at 100% of filtration flow for 40 seconds. The estimated water recovery for this scenario was 98%. The test was run for 120 hours with an average pressure increase of 0.0017 bar/h. A plot of net driving pressure as a function of hours of operation is shown in Figure 2. No mCIP was utilized during this test.

Test 2: The next test condition was a direct mimic of the conditions used by most polymeric UF/MF membranes, with a pump used to pressurize permeate for the backwash. Due to the pump ramp time this resulted in a slow build up in backwash flux, which was then maintained for a longer period of time. Dead end filtration was performed for 20 minutes at a flux of 170 lmh. Again PACl (1 ppm as Al⁺³) was used as coagulant. The backwash pump was set to reach a flow of two times the filtration flow or 340 lmh for a duration of 11 seconds. Following the backwash cycle was a feed flush at 100% of filtration flow for 10 seconds. The estimated water recovery in this scenario is 97%. The test was run for just over two days, with an average pressure increase of 0.003 bar/h. A plot of net driving pressure as a function of hours of operation is shown in Figure 3. No mCIP was utilized during this test.

Test 3: The third test was a hybrid of the first two backwash methods. This test used the fast step change of backwash water pressure of the first test, but at the lower flux rates of the second. At scale this backwash method would be accomplished by ramping the backwash pump ramp against a closed valve to build pressure prior the valve opening. For the pilot system this scenario was accomplished with the same compressed air backwash system described for the first test, but at a low pressure determined by the membrane permeability to achieve the target backwash flux. Dead end filtration was conducted for 20 minutes at a flux of 170 lmh with PACl (1 mg/L as Al⁺³) added as coagulant. The backwash flux was twice the filtration flow rate, or 340 lmh, for a duration of 10 seconds. After backwash, a feed water flush at 100% of filtration flow for 40 seconds' duration was performed. The estimated water recovery in this scenario was 95%. The test was run for just over 100 hours, with an average pressure increase of 0.002 bar/h. A plot of net driving pressure as a function of hours of operation is shown in Figure 4. No mCIP was utilized during this test.

Hydraulic cleaning optimization summary: Table 1 summarizes the conditions chosen in the first three tests along with the resulting system recovery and a linear fit of the rise in net driving pressure with operational time that were described in Tests 1–3.

Test 1 takes advantage of the mechanical strength of ceramic membranes to maximize the efficiency of the hydraulic cleaning through the use of a very fast, extremely high flux, backwash (often referred to a backpulse). Although efficient, it requires permeate piping systems that have the ability to tolerate compressed air at high pressure, as well as an air compression system, both of which add complexity and cost to the overall system.

The conditions of Test 2 resulted in a higher pressure increase over time than the more aggressive backwash method used in the first test and a slightly reduced recovery rate. However, the operation was stable and would be manageable with the addition of a mCIP. Further, the overall cost of the system is reduced with the more conventional backwash system.

The hybrid design approach of Test 3 was an effective way to get the benefits of the fast backwash step change in flux without the associated high capital costs. Although relatively simple to implement, this is not normally performed with polymeric hollow fibre systems due to the risk of the rapid pressure rise causing fibre rupture and loss of system integrity.

At the conclusion of the hydraulic backwash optimization, a series of experiments to push operating flux were conducted. The conditions for these tests were based on the hybrid design approach of Test 3. These tests also added a mCIP every 2–3 days to extend the testing period and confirm longer-term stability. The mCIP used UF permeate with sodium hypochlorite (300 ppm) at pH 12 (adjusted with NaOH). To stabilize operating flux at higher levels the dose of coagulant was increased based on earlier work, which suggested the optimum conditions were not the same as that required for charge neutralization (Gaulinger 2007).

Test 4: Dead end filtration cycles of 20 minutes were used with a flux set to 195 lmh. PACl (1 mg/L as Al⁺³) was again used as coagulant. The backwash flow rate was double the filtration flow, or 390 lmh. During the test the backwash pressure was adjusted to maintain this target flow rate. Following each backwash, the feed water flush at 100% of filtration flow was performed for 40 seconds. The estimated water recovery in this scenario was 96%. The test was run for 260 hours. The result was an average pressure increase of only 0.0000007 bar/h. A plot of net driving pressure as a function of hours of operation is shown in Figure 5.

Recovery cleaning would be utilized as a preventative maintenance measure, with an anticipated time between cleanings of 6–12 months.

Test 5: This test condition utilized a hybrid backwash method at a higher flux. Dead end filtration lasted for 15 minutes at a flux of 313 lmh. For this condition the PACl dosage was increased to 2 mg/L as Al⁺³. The backwash flux rate was kept at double that of forward flow, or 626 lmh, for a duration of 15 seconds. The feed flush was again kept at 100% of filtration flow, but with a reduced duration of 15 seconds to maintain recovery at 96%. The test was run for just over 100 hours, with one mCIP cleaning occurring during the third day of operation. The result was an average pressure increase of 0.0021 bar/h. A plot of net driving pressure as a function of hours of operation is shown in Figure 6.

It is likely that after several mCIP cycles the profile would flatten to a lower pressure increase profile over time due to the shorter duration of this test and the fact that it ended prior to a mCIP.

Test 6: The final test was conducted at an increased flux of 391 lmh, still using the hybrid backwash method. Dead end filtration cycles of 15 minutes' duration were chosen with a PACl dosage 5 mg/L as Al⁺³. The backwash flux was kept at double the forward filtration, or 782 lmh, for a duration of 15 seconds. The feed water flush was 100% of filtration flow for a 20-second duration. The estimated water recovery in this scenario is slightly lower at 94–95% to offset the increased solids from the higher coagulant dose. The test was run for just over 200 hours. The result was an average pressure increase of 0.0003 bar/h. A plot of net driving pressure as a function of hours of operation is shown in Figure 7.

The expected recovery cleaning frequency in this case would be every 3–6 months. At these flux levels, the capital cost of the ceramic UF modules is less than what is typical for polymeric UF in most applications.

Based on these experimental results, the analysis below is a paper exercise to compare the process designs of the ceramic membrane to several polymeric pressurized UF modules to demonstrate the feasibility of retrofits. The design scenario outlined assumes a theoretical system producing a peak flow of 114 m³/h. The selected peak flux is a nominal value of 60 lmh for polymeric UF membranes and would be typical for a surface water source.

Table 2 summarizes three polymeric hollow fiber systems with data taken from the manufacturers' data sheets, and a corresponding design illustrating how ceramic membranes could be configured to replace the polymer membranes.

In Case #1, the PES inside/out model has 60 m² or 2.5 times the area of the Nanostone ceramic membrane. The permeability of the ceramic membrane is 800–900 lmh/b and the PES membrane is listed at 350–400 lmh/b. Given the higher permeability of the ceramic (2.27×), but at three times the flux, the feed pressure is anticipated to be slightly higher for the ceramic membrane in this example. Depending on the fouling profile of the application and design of the previously installed feed pump, an upgrade may be required to provide higher pressures. The backwash flow for the UF skid in Case #1 is equivalent to that used with the ceramic example. The flux multiple of 3.8 times the filtration flow is within the stated backwash range for the ceramic membrane of 2–4 times.

In Case #2 another PES inside/out module is compared. In this case, the area is only 46 m². Consequently, fewer ceramic modules would be required to retrofit the polymeric modules in this model. In this case the flux of the ceramic membrane is 2.5 times the PES membrane, which closely matches the permeability difference. As a result, the feed pump would have sufficient pressure to run the ceramic membrane at the stated fluxes without needing to increase the design pressure. The backwash rate for this case is lower than for Case #1, but still yields a backwash flux multiple of 3.3 times the filtration flow. This is also in the design range of 2–4 times the filtration flow stated above for the ceramic membrane.

Case #3 is an outside/in PVDF hollow fiber membrane with 77 m² of area. In this case, the module count for the ceramic retrofit remains the same since the flux difference is inversely proportional to the area change. The permeability advantage of the ceramic membrane in this case is 3.5 to 4 times higher. Depending on the application and the fouling potential, the feed pump would provide adequate pressure. The backwash rate for this PVDF module is lower than the other two cases, given that this module uses air scour in the backwash mode. However, the backwash rate is 2.9 times the filtration flow rate and so falls within the typical design range for the ceramic membrane. Again, it is practical to assume that the existing backwash system would be sufficient in a ceramic retrofit. There are several PVDF outside/in hollow fiber modules on the market with low backwash rates that rely primarily on air scour to dislodge solids from the fiber bundle.

These three cases demonstrate that ceramic membrane retrofits of polymeric systems could utilize existing backwash designs of all three hollow fiber membranes modeled. Each retrofit would need to be evaluated on a case by case basis to determine if the feed pump and backwash system could be employed for optimum performance. Depending on the design of the piping connections to the modules, a direct module retrofit maybe possible, but would need to be evaluated for each case. Replacing the existing membrane rack with a new ceramic membrane rack may be more practical to limit the complexity of field work. Regardless of the approach, much of the infrastructure of the incumbent system could be utilized.

To evaluate the economics of a polymeric vs ceramic system, a techno-economic model is constructed below to determine the capital and operating costs of the three different fluxes rates studied in the tests above and compared with a benchmark polymeric MF/UF membrane plant. Costs for a specific retrofit are too case specific, so new plants are considered herein. The design and cost assumptions are shown in Table 3. For this example, a plant size of 2 million gallons per day (MGD) (7571 m³/day) is selected. The water source considered is direct river water at 15°C with turbidity less than 10 NTU. The capital cost of the primary components, not including membranes, is estimated at $0.50/GPD and is the same for the polymeric and ceramic membrane plant due to the similarity in system design and components when using the hybrid design described previously.

A $2000 credit for changes to the system cost as a function of changes in membrane module count account for the difference in pipe manifolds, connections, couplings, labor, and frame material needed for a skid that has fewer membrane modules. The model adds a charge for the cost of a higher pressure feed pump if needed for higher pressure operation. This is determined from the ratio of the clean TMP of the ceramic to the polymeric design case (e.g. for Case #1, the clean water permeability is about three times higher while the flux is about three times higher and so pressures will be about the same). For ceramic UF Cases #2 and #3 there is a penalty for the higher pressure pump required.

There are a wide range of end user prices for polymeric MF/UF membranes; a price of $35 USD/m² is used in this case as a typical value of end user prices for an initial system purchase.

With the higher flux Cases #2 and #3, the savings in membrane modules far outweighs the cost of the higher pressure pump in terms of initial capital cost, and savings up to 6% or $70,000 USD are observed in this 2 MGD model.

In Table 4 operating assumptions and cost inputs are listed. Power cost for all cases is $0.10 per kilowatt hour. Chemical costs used in the analysis: aluminium hydrochlorate (ACH) coagulant $0.57/Kg, hydrochloric acid $0.50/Kg, sodium hydroxide $0.10/Kg, sodium hypochlorite $0.09/Kg. These costs are then annualized, and 20 years of operating costs plus the initial capital costs are compared for a total cost of ownership analysis of the various cases in Table 5.

For the chemical cleaning, it is assumed that all cases use the same levels of chemical cleaning including an ambient maintenance cleaning with sodium hypochlorite (1,000 mg/L) and sodium hydroxide (600 mg/L) for 30 minutes every three days, as well as an ambient hydrochloric acid (1,000 mg/L) maintenance cleaning for 30 minutes every nine days. A full recovery cleaning is assumed to take place every three months for all cases using the same concentrations but at elevated temperature for four hours. The pilot data suggested that full recovery cleanings could take place with a frequency of six months to as long as a one-year interval. However, for the operating cost evaluation, a more conservative recovery cleaning frequency is used.

For membrane replacement costs, the typical pressurized polymeric MF/UF membrane life is five to seven years and the lifecycle model assumes a life span of six years. The polymeric membrane replacement price is also set at $35 USD/m² based on industry benchmarks. For ceramic UF membranes, the lifespan is estimated at 20 years and a replacement set is included in the financial analysis.

The power consumption model uses a clean water permeability value of 200 lmh/b for the pressurized polymeric membrane and 600 lmh/b for the ceramic membrane. The operating pressure is adjusted for flux and temperature and includes an allowance for fouling. Small pressure losses are also assumed for feed screen filters and piping. The backwash pump pressure is included in the power consumption as well as general power requirements such as the control system.

The main cost savings for the ceramic is the reduction in membrane replacements. With the higher flux cases the capital cost savings are offset by increases in power and chemical costs. This analysis shows that the most cost effective design for the ceramic membrane system was case two at a flux rate of 309 lmh and 2 mg/L of ACH coagulant dosage. Although not the lowest in capital cost, it had the optimum balance of capex and opex to give the lowest lifecycle cost. Comparing Case #2 to the benchmark polymeric membrane case, the total savings in this example is $0.16/GPD or $320,000 USD over a 20-year period and does so at a 3% lower initial capital cost compared to the polymeric membrane plant.

With recent reductions in ceramic UF membrane module prices now available, there is now opportunity to retrofit many mainstream water and wastewater applications where polymeric UF/MF membranes are more commonly applied. Ceramic membranes have several performance advantages over polymeric membranes, such as a longer life span, higher pressure tolerances and greater chemical stability. However, the overall system design around the ceramic membranes needs to be cost effective as well. The aggressive hydraulic cleaning methods typically used for ceramic membranes, while technically effective, have prevented system costs from reaching that of polymeric systems and have kept ceramic membranes from more widespread adoption. The hybrid hydraulic cleaning methods described represent an approach that can achieve an optimized balance of hydraulic cleaning effectiveness and low system cost.