Scaling and particulate fouling in membrane filtration systems

In the last decade, pressure driven membrane filtration processes; reverse osmosis, nano, ultra and micro-filtration have undergone steady growth. Drivers for this growth include desalination to combat water scarcity and the removal of various material from water to comply with increasingly stringent environmental legislation e.g. Giardia and Cryptosporidum removal guidelines of the Surface Water Treatment Rule (USA). Innovations in membrane manufacturing and process conditions have led to a dramatic decrease in membrane filtration costs. Consequently, membrane filtration has emerged as a cost competitive and viable alternative to conventional methods in drinking and industrial water production and in recycling and reuse. The potential of membrane filtration to solve our water quality problems is certainly only in its infancy as new applications and products emerge. However, membrane scaling, biofouling, organic fouling and particulate fouling (in this thesis scaling and particulate fouling were studied) exert severe limitations to the future growth of membrane technology. Scaling, occurring mainly in reverse osmosis (R0) and nanofiltration (NF), refers to the deposition of "hard scale" on the membrane due to the solubility of sparingly soluble salts e.g. BaS0 ₄ being exceeded. Whereas, particulate fouling is an especially persistent problem in all membrane filtration processes and refers to the deposition of suspended matter, colloids and micro-organisms on the membrane. Problems arising from scaling and particulate fouling are a reduction in product water flux or increasing operational pressures to maintain flux, which translates to increased operational costs. Membrane cleaning to remove sealants and foulants results in increased down time, energy and chemical use, and the production of waste water adding further costs attributable to fouling. Furthermore, if membrane cleaning is unsuccessful, the membranes have to be replaced to maintain production capacity.

It is widely recognised that the control of scaling and particulate fouling is instrumental in further membrane technology advancement and in decreasing costs associated with this process. However, this can only be achieved when reliable methods are available to predict and monitor the scaling and particulate fouling of feedwater and at present no such methods exist. Therefore, pilot plant operation is commonly used prior to designing full scale systems. Although this method generally provides reasonably good reproducibility, it is time consuming and expensive. The goal of this research was to develop methods to predict scaling (using barium sulphate as a model sealant) and particulate fouling in membrane filtration systems. These methods can be applied as tools to determine and monitor the efficiency of scaling and particulate fouling prevention techniques, for improvements thereof in the absence of expensive pilot plant studies and ultimately reduce costs.

Chapter 1 of the thesis gives an overview of membrane filtration in drinking and industrial water production and describes the most commonly occurring sealants and foulants and existing methods to predict and control these phenomena. Limitations of the existing methods in predicting scaling and fouling were illustrated. Whereby, at one RO pilot plant in the Netherlands treating River Rhine water, barium sulphate scaling occurred despite preventative measures i.e. antiscalant addition. While, under other operating conditions without antiscalant addition, no scaling occurred despite the high scaling tendency predicted for the concentrate. Similarly, the most widely used methods to predict particulate fouling i.e. the Silt Density Index (SDI), and the Modified Fouling Index (MFI _0.45 ) which simulate membrane fouling by filtering the feedwater through a 0.45μm microfilter in dead-end flow at constant pressure, are not sensitive to the presence of smaller particles. Furthermore, the

SDI is not based on any filtration mechanism and is not proportional with particle concentration. Therefore, it can not be used as the basis of a model to predict the rate of flux decline due to particulate fouling. In contrast, the MFI _0.45 index is based on cake filtration and is proportional to particle concentration and can be used to model particulate fouling. However, it does not satisfactorily correlate with particulate fouling observed in practice as it is not sensitive to the smaller particles which may be responsible for fouling. In order to carry out the research goal of this study, scaling and particulate fouling were split into two major research branches with specific research objectives to establish (1) the solubility and kinetics of scaling and to develop an approach for scaling prediction, using barium sulphate as a model scalant and (2) an accurate predictive test to determine the particulate fouling potential of a feedwater (further development of the Modified Fouling Index making use of ultrafiltration membranes with smaller pores). This was followed by the application of these methods to determine the efficiency of scaling and particulate fouling prevention techniques.

In Chapter 2, the accuracy of the most commonly employed method for predicting barium scaling i.e. the Du Pont Manual was examined. This method predicted the barium solubility of concentrate at the RO pilot plant of Amsterdam Water Supply (AWS) was exceeded by 14 times at 80% recovery at the fixed temperature of prediction of 25 °C. Yet no scaling occurred at the pilot plant for more than one year at this recovery. Possible explanations; inaccurate solubility prediction i.e. the RO concentrate were not really supersaturated and/or organic matter complexed barium were investigated. Seeded growth determination of barium solubility in the RO and synthetic concentrate (no organic matter) confirmed stable supersaturation and proved organic matter had no effect on solubility. Du Pont's method under predicted solubility by circa 30% at 25 °C. Finally, a more accurate method was developed and verified to predict solubility (and hence quantify supersaturation) in RO concentrate in the temperature range of 5-25 °C. This method uses Pitzer coefficients and an experimentally determined solubility product constant (K _sp ) for the RO concentrate.

In Chapter 3 the cause of stable supersaturation in the AWS RO concentrate, either slow precipitation kinetics and/or inhibition of kinetics by organic matter, was investigated. Barium sulphate precipitation kinetics; crystal nucleation, measured as induction time, and growth were investigated in batch experiments in RO concentrate and in synthetic concentrate containing (i) no organic matter and (ii) commercial humic acid. Supersaturation appeared to control induction time. Induction time decreased more than 36 times with a recovery increase from 80% to 90%, corresponding to a supersaturation of 3.1 and 4.9, respectively. Organic matter in 90% RO concentrate did not prolong induction time (5.5 hour). Whereas, commercial humic acid extended induction time in 90% synthetic concentrate to more than 200 hours. This was most likely due to growth inhibition as growth rates determined by seeded growth in synthetic concentrate containing commercial humic acid were reduced by a factor of 6. In comparison, growth rates were retarded only 2.5 times by organic matter in RO concentrate. However, growth rates measured for 80 and 90% RO concentrate were still significant and not likely to limit barium sulphate scaling. Results indicate that the nucleation rate expressed as induction time is governing the occurrence of scaling.

In Chapter 4 a more realistic method was developed to predict barite scaling based on the assumption that a threshold induction time can be defined which should not be exceeded to prevent scaling. Induction times were calculated for supersaturation (determined using the Pitzer model) and temperature data from the AWS RO pilot plant from a relationship derived from measured induction times at 2 5 ^o C. Safe (≥10 hours) and unsafe (≤5 hours) induction time limits, were derived from periods when scaling did and did not occur in the RO system at recoveries between 86-90%. Based on these induction times, safe and unsafe supersaturation limits were defined for 5-25 ^o C. Use of these limits allows more flexible operation in optimising RO recovery while avoiding scaling. The general validity of these limits should be verified in further pilot studies with feedwater of different quality and using different RO elements.

In Chapter 5, the Modified Fouling Index using ultrafiltration membranes (MFI-UF index) was developed. This index incorporates the fouling potential of smaller colloidal particles not measured by the existing MFI _O.45 or MFI _0.05 tests. In order to propose a suitable reference membrane for the MFIUF test, polysulphone and polyacrylonitrile UF membranes of a broad pore size expressed as molecularweight-cut-off (MWCO) 1 - 100 kDa were examined in tap water experiments. The measured MFI-UF (2000 - 13 3OOs/l ² ) were significantly higher than the MFI0.45 expected for tap water, (1 - 5 s/l ² ), indicating smaller particles were retained as the MFI is dependent on particle size through the CarmenKozeny equation for specific cake resistance. However, the MFI-UF appeared MWCO independent within the 3 -100 kDa MWCO range as most likely the cake itself acts as a second membrane, determining the size of particles retained and the resultant MFI-UF. The polyacrylonitrile membrane of 13 kDa MWCO was proposed as the most suitable reference membrane for the MFI-UF test as cake filtration, the basis of the MFI test, was proven to be the dominant filtration mechanism, demonstrated by linearity in the t/V versus V plot. This results in a stable MFI-UF value over time. Furthermore, field emission scanning electron microscopy of the membrane surface showed the pores were circa 1000 times smaller than the pores of the existing MFI _0.45 test membrane.

Chapter 6 investigated various aspects of the new MFI-UF test to establish its general use for characterising the fouling potential of feedwater. Namely, proof of cake filtration and linearity of the MFI-UF index with particulate concentration of low and high fouling feedwater, reproducibility of the MFI-UF index, methods to correct the MFI-UF index for test pressure and temperature differences to the standard reference conditions of 2 bar and 20'C, respectively and application of the MFI-UF as a monitor to detect feedwater changes over time. Cake filtration was demonstrated for high and low fouling feedwater as the MFI-UF was stable over time and proportional to particulate concentration for all feedwater tested. Reproducibility of the MFI-UF was found for 83% of the membranes tested from three different batches and in five tests using one membrane with chemical cleaning of the membrane between measurements. Correction to the reference temperature of the MFI-UF test required only correction of the feedwater viscosity. However, all the cakes formed by the filtration of the feedwaters tested were found to be pressure dependent i.e. cake compression occurred. Therefore, pressure compressibility coefficients were determined for a given feedwater and a global compressibility coefficient was calculated to correct to the standard reference pressure. At present the MFI-UF test can not be applied to quantify the fouling potential of a variable feedwater over time i.e. operate as a monitor, as the resultant MFI-UF value may be due to the combination of cake filtration with depth filtration and/or compression effects. Moreover, the delayed response in the MFI-UF index to a change in feedwater, may be due to the history effect in the calculation of the MFI-UF via the t/V vs V plot. More accurate measurement of time and volume is expected to resolve this problem and warrants further research. However, results in this chapter showed that the MFI-UF test can be used to characterise the fouling potential of a single given feedwater type and to register a change in feedwater quality.

In Chapter 7 the MFI-UF was applied to measure and predict the particulate fouling potential of reverse osmosis (R0) feedwater. MFI-UF measurements were carried out under constant pressure filtration at the IJssel Lake and River Rhine RO pilot plants of the influent feedwater and after pretreatment processes e.g. coagulation, sedimentation, conventional filtration, ultrafiltration, etc. Using the MFI-UF results, the pretreatment efficiency was evaluated and a comparison made with the MFI _0.45 which measures larger particles. The MFI-UF of the influent feedwater was circa 700 - 1900 times higher than the corresponding MFI _0.45 , due to the retention of smaller particles. A pretreatment efficiency of≥80%, was found by MFI-UF measurements at both plants. For the larger particles the MFI _0.45 gave a 90≈100% reduction. Minimum predicted run times for a 15% flux decline from MFI-UF measurements, assuming cake filtration occurs in the RO systems, were shorter than that observed at the IJssel Lake plant. This was most likely due to almost negligible particle deposition in the RO systems and/or particle removal from the cake formed under cross flow. Moreover, it was shown that cake resistance increased with ionic strength in MFI-UF tap water experiments and therefore, a correction of the MFI-UF index is required for salinity effects in RO concentrate. Finally, it was suggested that the MFI-UF be carried out under constant flux (CF) filtration to more closely simulate fouling in RO systems. Preliminary experiments were promising, the MFI-UF could be determined under CF filtration within≈2 hours for the low and high fouling feedwater examined and the fouling index I of the MFI-UF determined in the CF mode was linear with particulate concentration. In conclusion, the MFI-UF (measured at constant pressure or constant flux) was found to be a promising tool for measuring the particulate fouling potential of a feedwater. It can be used alone or in combination with the MFI _0.45 to compare the efficiency of various pretreatment processes for the removal of selected particle sizes and to determine the deposition of particles in target membrane systems.

In Chapter 8 the main conclusions of the research were summarised.