Membranes are being applied with increasing frequency and on an ever greater scale in the preparation of drinking water. Ultrafiltration (UF) removes harmful micro-organisms such as bacteria and viruses for example. When a membrane is damaged it can be penetrated, particularly by viruses. This is obviously undesirable. How can you show in practice that the membrane installation is actually effective in keeping out viruses? And how can you monitor this?
One way of determining the efficiency with which UF membranes remove unwanted elements is to measure the turbidity of the water before and after it passes through the membrane. However, this method is not sufficiently sensitive to monitor the effectiveness of virus removal in the drinking water preparation and thus the reliability (integrity) of the UF membranes. The concentration of viruses in surface water is between 10 and 100 times greater than that of bacteria. So the efficiency of removal also needs to be higher. Therefore in order to establish whether UF membranes meet the requirements laid down for virus removal, a sensitive measuring method is needed. For this purpose so-called surrogate viruses such as the bacteriophage MS2 can be added for the measurements. This is viable for laboratory or pilot scheme testing, but is undesirable on a practical scale in view of the costs and the possible negative effect on drinking water quality.
KWR has developed a patented sensitive method for identifying natural viruses from surface water. This natural virus (NV) method [1, 2] uses natural viruses as indicators of reliability (integrity) of the UF membrane, by estimating virus removal performance.
The degree of removal of micro-organisms, including viruses, is expressed by the term log reduction value (LRV). It is a logarithmic scale: 6 LRV means that one in every million viruses survives; 5 LRV means one in 100,000 survives.
Depending on the virus concentration in the water analysed, the NV method gives an LRV of 7 or more  with small sample volumes and without the addition of surrogates. The method has been tested at the laboratory and pilot scheme scale for suitability for systematically monitoring the integrity of UF membranes. The practical applicability of the method at full scale was then tested. Based on earlier research  the virus markers NV2247, NV2310 and NV2314, commonly found in surface water, were chosen. Of these three, NV2310 consistently had the highest concentration in incoming water (1 x 108 V/L). For that reason, the results of this virus marker are compared with state-of-the-art turbidity measurements.
Laboratory testing of the effect of fibre damage
In order to determine the effect of fibre breakage on the integrity of UF membranes, KWR and membrane producer Pentair X-Flow carried out tests with small UF modules (one with 120 fibres, one of 0.08 m2) the fibres of which were deliberately and systematically damaged. The effect of the fibre breakage was calculated with a simple Excel model, based on existing knowledge. This included, among other things, the effect of fibre breakage on the permeability of the fibre. It was assumed that the LRV for viruses of a broken fibre was 0 and for an intact fibre 5. The model calculated that with one damaged fibre out of 120 the LRV declines sharply, from 5 to 1.3. The same calculations were also carried out for greater numbers of damaged fibres.
Then intact modules and modules with one or three damaged fibres were double-tested in the laboratory. Two different types of damaged fibres were studied: leaky fibres (0.5 mm hole) and shorter fibres (as simulation for complete fibre breakage). The incoming water was from the Lekkanaal (NV2310 1 × 108 V/L). The natural virus concentrations and the turbidity were determined before and after the UF module. The intact modules gave an LRV of between 5 and 6 for NV2310 (illustration 1, top), while the turbidity test resulted in an LRV of just 2.2. Damaging just one fibre by making a hole in it led to a decline in LRV to 1 for NV2310. Further damage led to further decreases in LRV. The LRV values are in line with the decline calculated by the Excel model, although the model predicted a slightly (0.4) higher LRV over the whole range, so in other words the model systematically overestimates the LRV.
Illustration 1. LRV of intact and damaged UF module, determined with NV2310, turbidity and model calculation. Top for small module (120 fibres), bottom for 8-inch UF module (64 m2).
Also, shortening (‘breaking’) a fibre leads to a bigger decline in LRV than making a hole. However, a hole in just one fibre is enough to cause a significant decline in LRV as shown with the help of the NV method. The turbidity measurement shows comparable LRVs for the different types and degrees of damage: a value of around 1 for damage to a single fibre with a single hole. An intact fibre however gives an LRV of 2.2, as a result of which the turbidity measurement seems to show a much smaller decline than the NV method.
Pilot testing of the effect of fibre breaks
The effect of a broken fibre on LRV was also determined for an 8-inch UF module (18,600 fibres, 64 m2), based on the model calculations and the results of the laboratory tests (illustration 1, bottom graph). With one damaged fibre we saw a sharp decline in LRV to 2.5.
We then hot tapped into epoxy or outer layer of the membrane and systematically cut through the fibres in the UF module. After each such intervention the pipe was sealed and a test was carried out on three filtration cycles, each of 20 minutes filtering and half a minute rinsing (back) to remove any contamination. Sampling of the natural viruses took place half-way through the second and third filtration cycles. Tests were carried out with the intact module and with 1, 3, 5, 10 and 50 severed fibres. The incoming water was once again from a canal, the Twentekanaal, with 3 × 107NV2314 V/L. As well as the NV analysis the turbidity was also measured before and after the UF module.
Just as in the laboratory, 5 LRV can be demonstrated for an intact module (illustration 1, bottom graph). As more and more fibres are severed, LRV declines to 2; this is in line with the model calculations. Just as in the laboratory, here too the model predicts a sharp fall (to 1.2 LRV). The turbidity measurements also show declining LRV – from 1.5 to 0.9 – but the differences are significantly smaller than with the NV method.
The model calculations agree with the test results and can be used to predict trends in fibre breaks. However the calculations are based on only a small number of measurements. It would be advisable to further optimise the model with more pilot tests, to determine the effect of a damaged fibre on permeability, for example, or test various types of UF membrane with different types of water.
And then in practice
Conclusions and practical significance
With the new NV method, the efficiency with which intact UF membranes remove viruses can be determined without using surrogates, with a range of 5 LRV at both laboratory and pilot and actual scale. With turbidity measurements in the laboratory and in pilot tests, an LRV of only about 2 could be demonstrated.
The NV method shows that both in the laboratory (120 fibres) and at pilot scheme scale (8-inch UF module, 18,600 fibres), even with just one broken fibre, there is already a significant reduction in LRV. This tallies with the model’s predictions. Further optimisation of this model will lead to better predictions. The NV method shows the influence of broken fibres more clearly than do turbidity measurements. Users, such as drinking water companies, can use this new method to monitor the performance of the UF membranes precisely and to determine when the performance deteriorates and it becomes advisable to replace the membranes or take other measures.
Danny Harmsen (KWR); Emile Cornelissen (KWR and Ghent University); Han Vervaeren (De Watergroep); Stefan Koel (Pentair X-Flow)
 European Patent Office (EPO), Method for determining the effectiveness of removal of viruses in a purification process, EP 3 486 650 A1
 Hornstra, L.M, Rodrigues da Silva, T., Blankert, B., Heijnen, L., Beerendonk, E.F., Cornelissen, E.R. & Medema, G.J. 2019 Monitoring the Integrity of Reverse Osmosis Membranes Using Novel Indigenous Freshwater Viruses and Bacteriophages . ES&T 5 (9), 1535-1544.
Much use is made of membranes in the preparation of drinking water. When membranes are damaged the efficiency of virus removal (log reduction value, LRV) declines as viruses pass through the membrane. In this article, the LRV of ultrafiltration (UF) membranes is described with a new method of measurement for natural viruses (NV method) in surface water (without applying surrogates). With this method the influence of broken fibres on the integrity of UF membranes was determined at laboratory and pilot scheme scale and compared with model calculations. The tests show that for an intact UF module, an LRV of 5 can be demonstrated. A damaged module leads to a reduction in LRV to between LRV 3 and LRV 1 depending on the degree and nature of the damage. The model proves well able to predict this. Measuring the integrity of a UF installation at full scale using the NV method results in an LRV of between 4 and 5. With this new method, users can accurately monitor the performance of UF membranes.