New techniques for real-time monitoring of reverse osmosis membrane integrity for virus removal

Monitoring and control of pathogens in water treatment processes are a daunting challenge for the water industry and governmental regulators (Lozier 2003; Mi et al. 2004; Kumar et al. 2007; Phattaranawik et al. 2008; Brehant et al. 2010; Guo et al. 2010). Of the different pathogens, waterborne viruses are especially challenging because of their small size, high mobility, and resistance to chlorination (Grabow 2001; Theng-Theng Fong 2005; Pontius et al. 2009). Waterborne enteric viruses have been linked to a variety of diseases, including poliomyelitis, heart diseases, encephalitis, aseptic meningitis, hepatitis, gastroenteritis, and even paralysis in immunocompromised individuals (Theng-Theng Fong 2005). Enteric viruses, which are nucleic acid strands surrounded by protein protective coats (capsids), are oblate intracellular parasites, requiring host cells in order to replicate. In the absence of host cells, enteric viruses are essentially inert nanoparticles (NPs), typically in the size range of approximately 30 to 100 nm (See Figure 1).

The application and integration (with conventional processes) of pressure-driven membrane processes have grown significantly, over the past few decades, as part of a multibarrier water treatment strategy to safeguard water supplies from harmful pathogens and impurities. Low-pressure membrane (LPM) processes such as microfiltration (MF) and ultrafiltration (UF) typically provide physical barriers for particles larger than >0.1 to 5 mc and >0.05 to 0.1 mc, respectively. High-pressure membrane (HPM) processes such as nanofiltration (NF) can reject multivalent ions and materials of >0.0005 to 0.001 mc, whereas reverse osmosis (RO) can reject materials as small as monovalent ions. Given the typical size of enteric viruses (30 − 100 nm), LPM processes such as UF can be effective in providing near-complete rejection of enteric viruses. In principle, HPM processes (RO/NF) should provide a complete physical barrier to nanosized enteric viruses. Membrane and membrane module imperfections and/or damage, however, may render both LPM and HPM membrane processes ineffective in removing viruses.

• 99% (2-log) removal and/or inactivation of Cryptosporidium

• 99.9% (3-log) removal and/or inactivation of Giardia

• 99.99% (4-log) removal and/or inactivation of viruses

Some states implement a more stringent regulation for removal and/or inactivation of microorganisms from water. In California, for example, 4-log removal and/or inactivation of Cryptosporidium and 99.999% (5-log) removal and/or inactivation of viruses are required for disinfected recycled water (Haymore et al. 2001; California Department of Public Health 2009b).

The typical efficiencies of MF and UF membranes for virus removal have been reported in the literature, with LRV ranges of 0 to 4 for MF and 5.4 to 7.9 for UF (Haymore et al. 2001). A survey conducted by the U.S. EPA in 2001 (See Figure 2) indicated that most of the 29 U.S. states with policies regarding the use of membrane filtration have given virus removal credits for both MF and UF processes. These virus removal credits can be counted toward the required regulatory LRV for virus removal from surface and recycled waters. In awarding virus removal credits, states rely on membrane integrity monitoring (MIMo) techniques that have been well established for LPM processes (California Department of Public Health 2009c). In contrast, virus removal credits are rarely given to NF and RO processes because of the difficulty of online MIMo of HPM processes.

The absence of reliable membrane integrity testing (MIT) procedures for HPM limits the evaluation and thus regulatory acceptance of the typically high removal efficiencies that are provided by membrane technologies. Overcoming the above challenges is paramount to greater acceptance and use of membrane technologies to purify impaired water sources while optimizing virus inactivation with a multiple-barrier design.

A summary of various MIT methods that are applicable to HPM systems (i.e., NF and RO) is summarized in Table 1.

Practical online membrane integrity methods for HPM systems should be as listed in Table 2.

Given these desired attributes, direct MIMo techniques based on fluorescent marker monitoring appear to be most promising. Such markers have been shown to be detectable at sufficient resolution and sensitivity (using commercial spectrofluorometers) and thus should be suitable for monitoring membrane integrity. The existing approaches, however, rely on constant marker concentration in the feed stream, which is an impractical approach in water treatment facilities because of the (a) high cost of markers and (b) temporal variations in water quality. Moreover, comparison of LRV of the marker for a constant feed marker concentration between intact and compromised membranes may be insufficient for assessing the extent and characteristics of membrane breaches.

The proposed MIT technique (Frenkel & Cohen 2014) was focused on advancing current fluorescent-based MIMo approaches via a PM-MIMo approach as a practical, online monitoring strategy. Pulse injection of markers into the RO feed stream has been proposed, studied and tested to allow monitoring of the dynamic change in marker concentration in the permeate stream in real time. The dynamic change in marker concentration in the permeate stream, coupled with analysis of marker membrane transport behavior, can then be utilized to both detect and assess the extent of membrane integrity breaches.

Five types of commercially available molecular fluorescent markers were selected after initial screening of over 20 markers as shown in Figure 3. Fluorescein (C₂₀H₁₂O₅), uranine (C₂₀H₁₂Na₂O₅), eosin B (C₂₀H₆Br₄Na₂O₅), and lissamine green B (C₂₇H₂₅N₂NaO₇S₂) were obtained from Fisher Scientific (Pittsburgh, PA) in a powder form. Rhodamine WT (C₂₉H₂₉N₂O₅Na₂Cl) was obtained from Keystone Aniline Corp. (Chicago, IL) in the form of a 20% (w/w)- aqueous solution. In addition to these molecular markers, a fluorescent-labeled macromolecule (FITC − dextran) was also tested in this study. The marker solutions were prepared by dispersing a predetermined mass or volume in ultrapure deionized water (conductivity of 0.18 μS) obtained by filtering distilled water through a Milli-Q water system (Millipore Corp., San Jose, CA).

The fluorescence spectrometer (spectrofluorometer) in conjunction with a fluorescence flow cell, pulsed light source, and filters was obtained from Ocean Optics, Inc. (Dunedin, FL). The spectrometer was based on a charge-coupled device (CCD) detector suitable for the range of 175 to 1,100 nm. During the initial phase of the study, the spectrofluorometer system consisted of the pulsed xenon light source, which emitted light in the range of 300 to 750 nm, and the monochromator, which transmitted a selectable narrow band of desired wavelength selected from a wider range of wavelengths from the input light. The fluorescence flow cell was made of chemically resistant polyether ether ketone PEEK material and was designed specifically to accommodate low flow rates (up to 6 mL/min). The detected fluorescence from the samples was emitted to the spectrometer, where the light intensity was quantified in a relative unit called ‘counts’. A diagram of the spectrofluorometer setup is shown in Figure 4. The pulsed light source, coupled with the monochromator, enabled varying the wavelength of the light source entering the fluorescence flow cell. This work enabled determination of the appropriate excitation wavelength to optimize the spectrometer sensitivity for detecting the target tracer.

The sensitivity of the initially assembled spectrofluorometer system was sufficient for screening of the range of suitable excitation wavelengths. However, the intensity of the filtered light (at the selected wavelength) was low, resulting in lower-than-expected excitation light intensity. As a consequence, the sensor output integration time had to be increased from 500 ms to 3 s and with increased light intensity, thereby resulting in a spectrometer temperature increase of 2 to 4 °C. The spectrofluorometer system was improved (see Figure 5) by using an LED light source to replace the previous wide-band white light source. The selected LED (based on the optimum excitation wavelength of the selected fluorescent markers) emitted light between 460 and 520 nm, which was suitable for detection of the selected markers. In addition, the monochromator was replaced with optical transmission filters (on both the excitation and emission sides) in order to further sharpen the excitation and emission spectrum. The optical filters on the excitation and emission sides were at wavelengths of 490 ± 20 nm and 530 ± 20 nm, respectively. Finally, the spectrofluorometer system was mounted inside a temperature-controlled enclosure.

Continuous real-time PM-MIMo monitoring using PFRO (plate-and-frame RO) and SPRO (spiral wound RO) systems was conducted as single-pass RO desalting runs. The RO systems were set to operate at a desired cross-flow velocity and permeate flux, and the fluorescence backgrounds of the permeate stream were determined once the RO system reached a steady-state condition (no significant fluctuation in the permeate flux). Once a stable fluorescence background signal was attained, the marker solution was injected into the feed stream by a metering pump at a predetermined frequency and a prescribed concentration–time profile. Marker injection dose profiles were at concentrations of up to 40 ppm. The marker permeate concentration was monitored as a function of time for the duration of each marker injection event. At least 10 min was allowed between individual marker runs until the fluorescence signal returned to background level.

Two basic types of marker injection experiments were carried out. The first consisted of continuous injection of a tracer solution for a prescribed period of sufficient length to attain steady-state tracing of permeate concentration response. These runs were utilized to determine the mass transfer coefficient (k_f), the reflection coefficient, and solute B values for the intact and membranes with induced integrity breaches. The second type of marker injection experiments were injection pulses (duration of 60 s) in which the permeate marker concentration was monitored to determine the fraction of tracer passage through breached membranes. Data from these experiments served to evaluate the marker permeation time distribution (MPTD) as well as the marker rejection and thus the LRV for both the intact and compromised membranes. During the experiments membranes were impacted mechanically and chemically, impacts were measured and detection limits of the proposed technique were studied.

Uranine was selected as the marker for developing the PM-MIMo approach. Among the candidate markers, uranine demonstrated relatively stable fluorescence intensity at a typical RO process pH operating range (pH 6–8) along with a high level of chlorine tolerance. The detection limit for uranine if using the initial spectrofluorometer setup was 10 ppb with a detection limit of <1 ppb, attained with the use of an LED light source (at the optimal wavelength).

Following the selection of uranine as a final candidate marker, the spectrofluorometer system was upgraded with an LED light source and the optical filters designed specifically for the detection of uranine. Because the optimum uranine excitation and emission wavelengths are 485 and 520 nm, respectively, the optical filters were set at 490 nm on the excitation side and 530 ± 20 nm on the emission side. Such a combination of an LED light source and excitation–emission filters allowed sufficient separation of fluorescence excitation and emission lights for detecting uranine down to 2.5 ppb (compared to 13.0 ppb detected with the initial spectrometer setup). The higher sensitivity of the improved fluorescence detection system facilitated the use of a lower marker dose to the RO feed. The calibration curve correlating uranine concentration with fluorescence intensity was generated as shown in Figure 6, which compares the uranine calibration curve generated by the original spectrometer setup and the upgraded spectrometer setup with the LED light source. These reported limits of detection were based on the 99% confidence interval limits of the background fluorescence intensities, which were determined by using the intercept values and the linear regression statistics of the calibration lines shown in Figure 7(a) and 7(b).

The upgraded detection system, at 50% LED intensity setting and 500 ms sensor output integration time, enabled sufficient separation of fluorescence signals of the low-concentration uranine solutions (1 ppb and 5 ppb) relative to DI water (see Figure 6). Indeed, on the basis of the background fluorescence intensities (i.e., intercepts of the calibration lines in Figure 7(a) and 7(b)), the upgraded spectrometer setup was about four times more sensitive than the initial setup. The results suggest that the upgraded spectrometer setup requires a shorter sensor output integration time (500 ms) and a lower LED intensity than the significantly longer (3 s) integration time with the initial detection system. This improvement enabled more-rapid and more-sensitive measurement of tracer concentrations.

In order to further quantify the extent of increased marker passage across the membranes and to identify the mechanism of marker passage due to various types of integrity breaches, the marker overall LRV (LRV_overall) and LRVs due to solution–diffusion (LRV_diff) and to convection (LRV_conv), as well as the reflection coefficient (σ), were extracted from the data obtained from marker injection runs. Both the LRV_diff and LRV_conv were determined through the analysis, and σ was determined via the rearranged Kedem–Katchalsky model.

Analysis of the experimental marker data (see Table 3) demonstrated that, with the presence of a membrane integrity breach, there was an increased level of convective marker transport across the membranes. This finding was demonstrated by the decrease in both the reflection coefficient and the marker LRV. The reflection coefficient decreased with the increasing extent of the membrane integrity breach. For example, σ decreased from 0.988 to 0.985 when the number of pinholes increased from one to two in the first (i.e., lead) membrane module. A decrease in σ typically indicates an increasing level of convective solute transport across the membrane. In addition, it is also apparent that, in the presence of the membrane integrity breach, the increased marker passage was controlled by convection because LRV_total was nearly identical to LRV_conv, whereas LRV_diff was in the range expected for an intact membrane.

It is important that, even though the marker LRV for the tested intact SPRO membranes was below 4, it was possible to establish an LRV of greater than 4 for the intact SPRO system by using the PM-MIMo approach with the current spectroflurometer setup. This task can be accomplished by utilizing the membrane modules that have higher solute rejection than do the membrane modules that were utilized in the present study. Measurement of a higher LRV for a high-rejection membrane was possible, given the low detection limit of the present high-sensitivity online spectroflurometer setup (i.e., detection limit of 2.5 ppb). For example, Figure 8 shows that, given the present intact XLE membrane properties (i.e., B), the current SPRO flow conditions, and 20 ppm uranine dosing in the feed stream, a 4 to 6 LRV_conv of marker (i.e., which would represent the lower LRV level for viruses) could be attained when the marker concentration in the permeate stream was approximately between 14 and 16 ppb. The above concentration level could be easily detected, given the current spectrometer setup detection limit of 2.5 ppb.

In summary, MIMo, the newly proposed MIT process, demonstrated that the PM-MIMo approach can be utilized to detect and provide information on the characteristics of various types of membrane integrity breaches in the SPRO membrane system via real-time monitoring. The PM-MIMo approach is sensitive to minor breaches and should be able to demonstrate greater than 4 LRVs of the marker through the intact membranes of the SPRO system. The PM-MIMo approach clearly has potential for use as a real-time integrity monitoring technique for full-scale applications. In this regard, long-term pilot studies would be beneficial to demonstrate the accuracy, versatility, and robustness of this novel method of MIMo.