Abstract
Empirical observation of filter beds at South African water treatment plants showed that the filters were insufficiently cleaned by the backwash system and that media losses were unexpectedly high. Specific deposit tests developed by the RAU Water Research Group indicated that the dirtiness correlated with the organic content of the water being treated. This led to the hypothesis that biofilm is present on the media, somehow causing both the media loss and the difficulty to attain efficient backwashing.
Biofilm consists of organisms surrounded by a sticky, gelatinous polysaccharide matrix. This matrix, also known as extra-cellular polymeric substances (EPS), is the bulk (50-90%) of the biofilm. Biofilm plays an important role in the establishment and maintenance of organisms in a hostile environment. From the above it doesn¡¦t make sense trying to measure biofilm from the numeration of the organisms. A more reliable direct but tedious measure is quantifying the EPS. A new alternative method developed by the RAU Water Research Group is to mechanically strip the specific deposit off the filter media and then determine the organic fraction by combusting the sample at 500¢XC.
Two aspects of mechanical behaviour are deemed important in this study. First, headloss, because an under prediction in headloss will result in a higher than expected backwash frequency. Second, bed expansion, because an under prediction in bed expansion will lead to media washout. Literature indicates that both headloss and bed expansion increase with increasing biofilm growth. However, all those studies were conducted at waste water treatment plants with high organic and solids loading. With the exception of one reference which only discusses headloss, nothing on this topic is available in the literature for potable water treatment.
Mathematical models were used to reduce the data from multiple headloss and bed expansion experiments. For the headloss data the Ergun equation was used and the sphericity (ƒÚ) was retained as the only unmeasured calibration constant. For the bed expansion data the Dharmarajah equation was used and the sphericity was retained as the only unmeasured calibration constant. Calibration of the mathematical models was done with least square fitting. The two values of sphericity as determined by Ergun and Dharmarajah are not necessarily the same for the same media sample. The sphericity was used as a calibration constant without any physical meaning, which accounts for different sets of complex unknowns.
Samples for experimental work were drawn from full scale operating water treatment plants. The treatment plants were spread over four provincesof South Africa with different raw water sources, but using approximately the same media. The sampling was done on three occasions, Winter 2003, Summer 2003 and Winter 2004, to cover the extreme temperatures experienced in South Africa. Samples collected at the plants were tested for headloss and bed expansion, then transported back to the laboratory and placed in the oven for 24 hours at 110¢XC. The sample was then sieved and the density determined. The headloss and bed expansion tests were then repeated in the laboratory. Parallel to these tests, EPS and volatile fraction quantification tests were done.
Direct methods of measuring biofilm, namely EPS and volatile fraction, yielded measurable results, thereby confirming the presence of biofilm. Plants that had large quantities of EPS also had a high volatile fraction, thereby confirming the expectation that the volatile fraction is an excellent method to rapidly quantify biofilm presence. EPS made up 41% of the volatile fraction, which is roughly comparable with the 50-90% quoted in literature. Where large quantities of EPS were found at a plant, a high TOC reduction also occurred through the filters.
The indirect methods of measuring biofilm, namely headloss and bed expansion, also yielded measurable results. The filter media with biofilm as sampled from the treatment plants had a higher headloss and bed expansion than the same sample after drying and sieving, which resembles virgin filter media. The sphericity values for headloss decrease by as much as 26% which translates to a headloss gradient increase of 150mm/m at typical filtration rates. The sphericity values for bed expansion decrease by as much as 18% which translates to a bed expansion increase of 17% at normal backwash rates. The conditions at the treatment plants sampled suggest that biofilm growth is stimulated by eutrophic raw water and the presence of pre-ozonation and inhibited when the high pH lime process is used. The mechanism which causes the increased headloss and bed expansion with increased biofilm is hypothesised to be media grains sticking together causing clumping, and not grains which are individually and uniformly covered with a smooth, uniform layer of biofilm.
Designers can compensate for this increase in headloss and bed expansion in two ways. They could either apply a correction factor after application of the models to allow for more headloss or bed expansion during eventual plant operation, or they could adjust parameters within the models to account for the larger headloss or bed expansion. As the surface area sphericity was used as a calibration factor in this study and could account for different sets of complex unknowns, it is suggested that this factor is used for adjustment of the model.
Operational practice in South Africa often includes in-situ chlorine or acid treatment to alleviate the problem of dirty filter beds. In this study, however, where high and efficient backwash rates were used during tests, no significant improvements in media cleanliness could be attributed to the use of either chlorine or acid. It seems that a good backwash system doesn¡¦t need such remediation, but plants with a backwash system which underperforms might find such remediation useful.
Prof. J. Haarhoff