Applications of Computational Fluid Dynamics Modelling in Water Treatment
Report No 1075/1/05
March 2005

EXECUTIVE SUMMARY

Almost all water and wastewater treatment equipment rely on continuous through flow of water. Some equipment requires this flow to be well-mixed, whereas other equipment requires plug-flow. Examples of well-mixed systems are activated sludge plants, chemical dosing zones and anaerobic digesters while sand filters (in both filtration and back-wash modes), clarifiers, adsorption columns (ozone, activated carbon and ion exchange) and dissolved air flotation cells are examples of the plug-flow systems.  Some processes such as nutrient removal activated sludge plants require the combination of both plug flow and completely mixed reaction zones.
The laboratory-scale experiments that are used to obtain design data for a plant are usually operated under ideal flow conditions; unfortunately it is usually not feasible to carry this through to full-scale plants, due to the greater difficulties and expense of achieving similar ideal conditions on a large scale. The complexity of the flow patterns, and the uncertainties about how they affect the relevant performance indicators for the process involved have led designers of equipment to use safety factors based on experience to ensure that the process achieves its required objectives. This means that equipment that is installed is often larger and more expensive that it needs to be.
Computational fluid dynamics (CFD) is a numerical procedure to calculate the properties of moving fluid. Most water treatment processes involve the movement of water. This motion is often complex and difficult or very expensive to observe. The prediction of the flow patterns and other properties of flowing fluids would provide insight into processes which otherwise would not have been possible. A previous WRC project (No 648) indicated the value of CFD modelling of clarifiers and an anaerobic compartment. It was able to both logically explain the unexpected behaviour of the clarifier and in designing features to modify the undesirable flow pattern.
Apart from its use in design of water treatment equipment, CFD modelling can also assist in research into water treatment processes.  The project on which this report was based was unusual in that it was initiated to provide a CFD modelling service to assist water researchers who felt that it could enhance their investigations. As a result the project did not have a specific research focus of its own, but adapted to the objectives of each research project that it became involved with.  Furthermore, not all the collaborations that were started were fruitful. The main content of the report is a series of case studies, each corresponding to a different investigation. To give the report some thematic consistency, it has been compiled with a view to illustrating the kinds of situations where CFD modelling is useful. To this end, the case studies that were selected for inclusion in the report are those which best fitted this purpose, i.e. they each involved an appropriate application of CFD, and yielded some useful conclusions. Various investigations which for one reason or another did not fit these criteria have been left out. The investigations which are presented have also been cleaned up to reflect a logical development which did not always take place in reality; i.e. the various misapplications, misconceptions and dead ends which occurred along the way have been removed from the narratives.

1 Project Objectives
2 Overall Course of the Project
At the inaugural meeting in 1999, a list of water-related projects with potential for CFD input was put forward. Seven activities were originally placed in the work programme for 1999, which would have involved collaborations with researchers from around South Africa.  These were:
  • Ozone Contactor
  • Anaerobic baffled reactor (ABR)
  • Secondary settling tanks
  • Passive mine water treatment
  • Filter back washing
  • Capillary membrane modules
  • Ultrafiltration manifolds
    
Not all of these evolved into fruitful collaborations; the direction and extent of progress depended on unpredictable factors which emerged. During the course of the project, reasons for making progress, or failing to do so, largely revolved around personnel capacity to undertake the work, though the logistics of collaborating with remote groups also played a part in some cases.  No postgraduate students were directly associated with the project during 2000, although some work was done by undergraduate students taking vacational employment.

Between December 2000 and January 2001 modelling or the secondary settling tanks at Durban's Northern Wastewater Wastewater Treatment Works was carried out to see if it was possible to achieve a significant increase in capacity without extensive redesign. Unfortunately this exercise was inconclusive.  Promising results with were initially obtained later proved to be a result of incorrect flow rate data supplied by the Metro.  When these were corrected, the improvements that could be achieved using simple baffles were found to be inadequate for the requirements of the treatment works.  In view of the lack of a conclusive outcome, this investigation has not been included in the report.

At the beginning of 2001 Tzu-Hua Huang, a chemical engineering graduate of the University of Natal, enrolled for a MScEng, taking the study of the ozone contactor as her thesis topic. Chapter 3 of this report is based on her work. Additionally, Emilie Pastre, a chemical engineering graduate from ENSIGC in Toulouse, undertook her MScEng studies in Durban, in terms of an exchange agreement between ENSIGC and the University of Natal. The topic that she chose involved modelling the final product water reservoirs at the Wiggins Water Treatment Plant, in order to develop an effective control strategy for chlorine dosing. The primary focus of her work was the control, with the CFD model contributing only to the conceptual development of a control model.  This work will be reported in another WRC report, so has not been included here. The During 2001 the pilot plant Anaerobic Baffled Reactor (ABR) was installed at the Umbilo Wastewater Treatment Works and was operated to gather data for the design of system for treating wastewater from informal settlements. The CFD support that was provided to this investigation was never a central issue; rather the CFD modelling was used to help deciding on structural details such as the shape and placement of baffles, by allowing visualisation of the consequences of different options. Operating experience led to the conclusion that the concerns which motivated the CFD modelling were not of crucial importance, and that the CFD work had not made a significant contribution to the investigation. This work has also been omitted from this report as it will be part of the ABR report.

In July of 2001 Ms. Huang visited Institut National des Sciences et Appliquée (INSA) de Toulouse, for an month.  There, with the help of Prof. A. Liné, she undertook some two phase modelling of the ozone contactor. The investigation formed the basis of a paper presented at the IWA Conference on Water and Wastewater Management for Developing Countries, Kuala Lumpur, Malaysia, 29-31 October 2001, which was subsequently accepted for publication in Water Science and Technology. Ms. Huang subsequently applied successfully to have her MScEng registration changed to PhD.
Also during 2001, an investigation into the backwashing of sand filters at the Faure Water Treatment Plant operated by the Cape Metropolitan Council. The motivation for the investigation was that the efficiency of back washing the sand filters was not uniform over their whole area, and that excessive quantities of backwash water were required to get some areas of the filter clean. A number of other treatment works under the control of Cape Metro have filters of similar design, with similar problems, so a solution to the problem would be widely applicable. This study is the basis of chapter 4.
In 2002 an extension to the project was granted, in order to complete the study on the ozone contactor at the Wiggins Water Treatment Plant. A series of studies was carried out to monitor the contactor at the Wiggins Waterworks, to obtain data to be used in partial validation of the computational fluid dynamic model.

The studies reported in chapters 5 and 6 resulted from suggestions made by members of the steering committee.  During 2002 CFD modelling was undertaken to support the design of modifications to a potable water clarifier at the Hazelmere Water Treatment Works north of Durban, operated by Umgeni Water.  This was the same clarifier that had been reported in WRC Report No 648/1/02 The Application of Computation Fluid Dynamics to Water and Wastewater Treatment Plants, and the upgrade was a direct sequel to the previous investigation. The final investigation was prompted by a suggestion that a CFD analysis of batch settling tests might be useful for developing a strategy to advance the CFD modelling of secondary settling tanks.

3 Literature Survey: the use of CFD in water treatment

A survey was carried out on the literature relating to the use of CFD modelling in water and wastewater treatment. This covered recent papers on CFD in  water research, CFD in the water and wastewater industry, and CFD in environmental studies. It was found that, although CFD has a very extensive literature, very little of this is related to water treatment. In a search conducted through the ISI Web of Knowledge site, the keyword CFD found 2196 references in the 12 months preceding September 2003, but CFD water treatment found only 8, and of these only 3 referred to water treatment as understood in this report.

4 Case Studies

The case studies carried involved a range of aspects that covered many of the broad issues found in the literature.
4.1 The Ozone Contactor at the Wiggins water treatment works
Ozonation is used in drinking water treatment primarily to oxidise iron and manganese, to remove odour- or taste-causing compounds, and to destroy micro-organisms.  The peripheral benefits include possible reduction in coagulant demand, enhancement of algae removal and the colour removal.  Ozonation of water is typically carried out by dispersing gas containing ozone into the liquid phase. The contact between the two phases accompanied by an ozone mass transfer takes place in ozone contactors.

The pre-ozonation system at Wiggins Waterworks, operated by Umgeni Water in Durban, consists of four contactors. Each of the contactors is preceded by a static mixer such that every chamber can operate individually or in parallel with another contactor. An ozone-oxygen gas mixture is injected as a side-stream through the static mixer which is employed to achieve high mass transfer of ozone to water.  The Wiggins pre-ozonation system has an unusual configuration, as it was adapted from an existing structure, which had originally been designed for a different purpose.  Water enters from the static mixer at the bottom, and passes through three horizontal compartments before it exits over the weir at the top.

The objectives of the investigation were:
In outline, the phases of the investigation were approximately as follows (various mistaken or misguided excursions have been excised from the sequence):
  1. A single phase (water only) CFD model was set up to provide an initial understanding of the flow patterns in the contactor.
  2. Tracer tests using lithium chloride were carried out to compare with the model.  These were conducted with and without gas injection into the static mixer. Although the gas injection did cause some noticeable difference in the measured outlet concentrations, the effect on the overall residence time distribution was very small.  After some adjustment to the model, it was concluded that a single phase (i.e. water only) model would give an adequate representation of the RTD for modelling the ozone reactions.
  3. An ozone reaction scheme was added to the model, using kinetic data obtained from the literature. From this it was evident that the ozone consumption is very dependent on the local characteristics of the water, which need to be determined experimentally.
  4. Sampling lines were installed on the contactor which allowed ozone concentrations to be measured at various points.  The positions of these were chosen with reference to the CFD model results. Consideration of both the model results and the measurements suggested that the best point for monitoring the ozone concentration for control purposes was located between the 2nd and 3rd compartments, rather than at the outlet of the 3rd compartment as at present.
  5. A laboratory study was initiated to obtain rate constants for the reaction scheme. At the time of writing, this was in progress.
  6. To check the validity of neglecting the effect of gas injection, some two phase modelling (gas bubbles in liquid) was carried out. First a two dimensional model was tried, and when this proved successful, a full three dimensional model was implemented. However satisfactory results were not achieved, due to grid resolution and convergence difficulties. These might well have been resolved with more powerful computers than those available.
The main conclusions related to the objectives of the case study were:
The investigation was aimed at improving the operating rules for the contactor rather than changing any aspects of its design, however the  mass of detailed information provided by the models did indicate aspects of the design which could be improved. Since the study was by far the most comprehensive undertaken during the project, it provided the broadest illustration of the use of CFD in research into water treatment processes, and some of its strengths and weaknesses. These conclusions can be generalised to an extent by noting that CFD modelling is very successful where the underlying physics of the process are very well understood, but becomes less useful and reliable when sub-models are added which involve approximations and uncertainties.

4.2 The sand filter backwash system at the Faure Water Treatment Works
A CFD model was set up to model the water-only phase of  the backwash cycle one of the sand filters operating at the Faure treatment works.  The objective of the investigation was to determine why parts of the filter take much longer to clean than others, and to propose modifications that would lead to improved operation. The modelling was accordingly divided into two phases: modelling of the existing configuration and modelling of the proposed improvement.

The model of the existing configuration showed that the pressure in the underdrain tends to increase towards the far end from the feed, due the general deceleration of the flow. Because of this increase in pressure, the flow through the nozzle slabs also tends to increase towards the far end, where the model predicted that the flow would be about 30% higher than the average for the filter.  This means that the parts of the filter close to the feed end get less than their fair share of the flow, which explains why they take longer to clean.

Having satisfactorily explained the reason for the operational problem, a CFD model of a proposed solution to the problem was set up. This was to install a flow distributor down the centre of the under drain, which ensures an even supply of water to each section.  This would take the form of a pipe laid along the length of the under drain, with holes on each side.  The diameters and spacing of these holes would need to be carefully gradated down the length of the pipe to deliver a uniform volumetric flow per unit length in spite of the pressure rise.

The model was constructed in two parts, one for the flow inside the distributor, and one for outside the distributor. The model predicted that the variation over the filter surface should be reduced to less than 1%.

Although the distributor appeared to be relatively inexpensive to install, at the time of  writing it had not been installed so its efficacy had not been verified.

4.3 The clarifier upgrade at the Hazelmere Water Treatment Works
In this case study, a series of CFD models were generated to support the design work for modifications to a clarifier which needed to have its performance upgraded. The peripheral feed arrangement for this clarifier was particularly unusual, and caused it to be plagued by poor feed distribution resulting in severe short-circuiting. An investigation into the maldistribution of flow occurring in this clarifier using tracer testing and a CFD model was reported in WRC Report No 648/1/02 The Application of Computation Fluid Dynamics to Water and Wastewater Treatment Plants. The conclusion of that investigation had been that converting the clarifier to a central feed arrangement was the only way to obtain a significant improvement in its performance.

In June 2000 Umgeni Water reviewed the existing design and made recommendations on proposed improvements to Clarifier 1 and Clarifier 2. The working group tabled the following design proposal:
It was expected that these modifications would increase the clarifier capacity from about 9 ML/d to 15 ML/day at an up flow velocity of 1.2 m/h (within the design guideline value of
1.5 m/h overflow rate for this type of clarifier). While the design work was being carried out, CFD modelling was undertaken to help evaluate various design options.  This interaction led to a number of changes to the design:

The modelling was based on data supplied by the Umgeni Water design team as the design work was proceeding, and the configuration was continually being changed while the modelling exercise was in progress.

The design modifications were implemented on the No. 2 clarifier at Hazelmere, which was re-commissioned in August 2002.  A comparative performance test between clarifiers No. 2 and No. 3, which also has a central feed configuration but none of the other CFD-designed features, was carried out during August 2003.

                      Clarifier turbidities                

% Cases where turbidity is exceeded

                 Turbidities measured at each clarifier during the comparative test

The graph shows the inlet and outlet turbidities measured during the test, plotted first against the elapsed time of the test, and also on a percentile basis.  During the test period the feed water turbidity was extremely low, so that flocculant dosage increased the turbidity significantly, which is how the turbidity from No. 3 comes to be higher than the water feeding it at times.  The superior performance of the No. 2 is clearly evident, which vindicates the use of CFD in its design.

4.4 Batch settling of secondary sewage sludge

The modelling of solids settleability is essential for modelling settling tanks in water and wastewater treatment.  Until the advent of hydrodynamic models, the focus of modelling solids settleability was on describing the behaviour of the solids in the water while the water itself was considered a stationary or ideally moving medium in which the solids settled. Hydrodynamic models now allow the behaviour of the water in the settling tank to be modelled.  While the modelling of the water flow has made extraordinary advances in the past 20 years, modelling the settleability of the solids has not improved much over the this time.  In fact, the weakest part of hydrodynamic models of settling tanks may be the modelling of settleability of the solids.  This investigation explored methods for measuring and modelling solids settleability with the view of improving these for hydrodynamic models of settling tanks.

The design and operation of secondary clarifiers is commonly based on the solid flux theory. The basic data required for the application of this theory can be obtained from multiple batch tests by which the stirred zone settling velocities over a range of sludge concentrations are measured (dilution experiments).

Many CFD modellers of settling tanks have used the Takács equation to describe the settling velocity of the solids, however the equation is not well formulated for experimental calibration. It contains 4 constants that require measurement to calibrate it.  Only 2 of these constants are readily measurable from laboratory scale tests, the remaining 2 usually have to be inferred from measured values of the suspended solids in the effluent from the full-scale clarifier. This is unsatisfactory, in that the clarifier cannot be properly modelled without using its own operating data.
 
The strategy of this investigation was to incorporate the Takács settling model into the simulation of batch settling tests, in an attempt to identify characteristics which might be amenable to experimental measurement, and which might allow the complete set of Takács parameters to be estimated.

From the batch settling simulations, two characteristic settling behaviours were identified, dependent on the initial concentration of sludge in the settling test. The features of Type I settling, which occurred at higher sludge concentrations were:

The notable features of this result are:
Type I settling occurs for initial sludge concentrations higher than a critical value Cm for which the sludge settling velocity is a maximum.  For tests starting from concentrations below Cm the a qualitatively different Type II settling behaviour takes over. Under these conditions, the interface with the "clear" liquid is diffuse, whereas the interface with the
settled sludge at the bottom of the column is sharp.  These qualitative features correspond well to experimental observations.

Although the simulation results indicated there was no direct way to determine all the Takács equation parameters from a single batch settling test (which confirms practical experience), a series of experiments with different starting concentrations could be conducted to determine the value of the critical concentration Cm at which the settling behaviour switches between Type I and Type II. This value, together with the settling velocity vm of the sludge at this concentration, could then be used to infer the remaining Takács equation parameters.

The character of this case-study was somewhat different to the others undertaken during the project, in that the CFD model was used to suggest a direction for further research, rather than to interpret or extrapolate research results. An experimental investigation needs to be undertaken to verify the suggested protocol.

4.5 Conclusions and recommendations

This section presents the more general conclusions which arise from considering the project as a whole.

4.5.1 The scope for application of CFD modelling in water treatment
It is interesting that most of the broad issues identified the literature were touched on in one form or another during the  investigations undertaken during this project. The Wiggins ozone contactor started with simple hydraulic modelling and prediction of the residence time distribution, and progressed to more complex physical modelling of reaction kinetics, disinfection performance and 2-phase flow. The Faure filter backwashing investigation looked at simple hydraulic modelling of flow distribution in the context of an equipment re-design exercise, concentrating entirely on the one specific issue for the design, and ignoring or approximating all other aspects of the system. The Hazelmere clarifier investigation was similarly a re-design exercise, but this time it involved two-phase modelling. It also provided experience of working interactively in the design team, with the concomitant time and budget constraints, requiring strict focus on the specific design objectives, at the expense of realism and unnecessary detail.  Finally the batch settling investigation again involved two phase modelling, but this time addressed a purely theoretical question. Thus the experience gained allows a reasonably comprehensive assessment of the role that CFD can play in water and wastewater treatment.

The very fundamental nature of the CFD approach has the advantage of being able to represent appropriate systems (see below) in great detail with minimal requirements for empirical data, but the disadvantages of complexity and difficulty in solving the resulting systems of equations.  These practical difficulties prevent CFD from being a universally appropriate approach to all problems involving fluid flow, in spite of its fundamental basis. Generally, CFD is most useful for systems which are well-connected, that is, where all the boundary conditions have relatively strong influences on all parts of the flow field. This applies to many systems found in water treatment, such as reservoirs, contact chambers, sedimentation basins, ponds and even lakes and lagoons.  However there are also many systems in water treatment where CFD does not provide an effective approach, for example a set of equipment connected by a pipe network, or a long reach of a river. In such cases the CFD model would expend enormous computational effort on calculating the practically negligible effects of remote boundary conditions.

The simplest CFD models consider only the hydraulic aspects of a system.  Frequently these models are used to predict the residence time distribution (RTD), which often provides a link to more direct performance indicators through empirical rules based on experience (e.g. the disinfection CT rule). As CFD modelling has become more established, more detailed models are appearing which attempt direct representations the physical and chemical processes taking place in the treatment processes, such as sedimentation, flocculation, inter-phase mass transfer and chemical reaction.  In all cases these more complex models need to be supported by experimental studies to establish the parameters for the physical and chemical parts of the models.  The CFD modelling thus has a role in both the interpretation of results from experimental apparatus, and in extrapolating research results to the design of full-scale processes.

The relationship between tracer testing and CFD modelling to determine the RTD of a system is worth mentioning.  The ozone contactor study demonstrated the use of tracer testing to verify the CFD model, and concluded that CFD modelling is often able to predict the RTD very accurately. However, if the RTD is all that is required, the tracer test may be quicker and less expensive to perform than to develop a CFD model.  However this depends on the size of the system: for many water treatment systems the size is such that a very large dose of tracer is required, together with an elaborate and expensive sampling and chemical analysis programme, and the time required to complete the test is so long that it is not feasible to maintain conditions steady for long enough..  Nevertheless, tracer testing should always be considered as a possible alternative to a CFD study, as long as the RTD is adequate to address the required purpose. A less tangible factor that should be borne in mind is the extra insight that the CFD model is able to bring to the investigator.

4.5.2 The costs involved in CFD modelling

The literature does not reflect a widespread acceptance of CFD modelling in water and wastewater treatment, and this is mirrored in the South African water industry. The cost involved in undertaking such modelling is undoubtedly one of the factors contributing to this situation. To some extent this is a matter of perception, but the reality is that the overall cost of a CFD investigation is likely to be fairly high.

To start with, the skills required are relatively rare, and take some time to develop. CFD has not yet found a place in undergraduate curricula, so postgraduate training is involved.  The underlying mathematics is complex and not easy for practising engineers to master on their own. It is true that the CFD software now available takes care of almost all the mathematical complexities, but paradoxically this may make the problem worse rather than better, because it makes it so easy to obtain plausible results which one does not really understand, increasing the potential for making serious errors.  The large international water treatment firms such as Veolia and Thames Water have a small number of CFD specialists who act as internal consultants for equipment design. During this project, the example of the Hazelmere clarifier illustrated how such an arrangement might work..
 
CFD modelling does require more than usually powerful computing hardware, and the requirements escalate rapidly when modelling the more complex physical processes, as illustrated by the difficulties encountered with the gas phase in the ozone contactor. However the hardware cost is less and less significant as a cost factor, as it continues to decline steadily, and since the skilled personnel costs involved  in such advanced modelling are likely to be very significant.

The cost of CFD software has come down steadily during the duration of the project, but is still high, even with a discounted academic licence. A non-academic licence would have been much more expensive.  The development of the software seems to be driven by much higher cost applications that water treatment - aerospace, chemical manufacture, power generation, automotive design etc., and the pricing appears to reflect this kind of market.  Many of the models, for instance combustion or solidification, that are available in the software are not relevant to water treatment - it could be that more limited package could be marketed to the water and wastewater industry.
 
The wide range of problems which might be tackled with CFD makes it very difficult to make any general statement about the cost of undertaking CFD modelling. However, some idea can be obtained by considering a relatively straightforward investigation, such as the Hazelmere clarifier (chapter 5).  The time involved was about 40 h (excluding report-writing), so the personnel cost would be of the order of R6 000 (2003 rand values). The software licence cost for commercial use of the Fluent software was about R16 000 per month.  Since the minimum period made available by Fluent was 1 month, it depended whether other jobs were available to share the cost, so the software cost would be between R4 000 and R16 000, and the overall cost would be between R10 000 and R22 000, to which would be added the costs for computer time: if done with a suitable personal computer (1GHz pentium with 500Mb RAM) this would amount to about R100.

This cost might be considered reasonable for a large clarifier, but appear excessive for a small unit. The problem is that the cost would probably be about the same irrespective of the size of the installation. For small units the costs could be effectively reduced by developing standard designs which could be reused a number of times.  The extra design cost should be recoverable through lower operating cost, however this will probably be difficult to quantify beforehand.

4.6 Recommendations
The recommendations are divided into those that relate to the individual case studies, and the more general ones that relate to the application of CFD to water and wastewater treatment.

4.6.1 The Wiggins ozone contactor

  1. The position of the ozone sensor monitoring the residual ozone should be moved to the position identified in the investigation.
  2. A new strategy for the control of the ozonation should be developed that takes into account the ozone demand of the raw water and the disinfection efficiency of the contactor. 
  3. The cost-benefit balance of the ozonation in the overall water purification process is not easy to quantify; it could even be the case that the benefits do not justify the cost.  The model that has been developed for the contactor could be very useful as a component in a wider investigation of the role of ozonation in the overall treatment process.
4.6.2 The Faure filter backwash system

A backwash distributor should be installed on a trial basis on one of the filters at the  Faure Water Treatment Works.  If this proves as successful as predicted, the system could then be installed on the other filters at the works, and on other filters of similar design.

4.6.3 The Hazelmere clarifier

The design modifications which proved so successful should be implemented on the remaining peripherally fed clarifier at the Hazelmere Water Treatment Works when appropriate. The design should also be considered for new clarifiers of a similar size.

4.6.4 The batch settling test for sewage sludge

An experimental project should be undertaken to test the protocol suggested by the modelling results. This would involve carrying out laboratory settling tests to determine the sludge settling parameters, using these parameters in a CFD model of a full-scale clarifier, and testing the model predictions experimentally on the full-scale unit.

4.7 General recommendations

The South African water industry still needs to develop an adequate pool of CFD expertise that can be called upon when appropriate. This might involve: