Stacked Filtration Research

Introduction and Objectives

A vertically stacked filtration system meets many of the AguaClara Project constraints. First of all, both normal filtration and backwash operations are gravity driven and requires no electricity. It is an open system. The construction material, PVC pipes, sand, concrete, and rebar are relatively cheap and abundant in Honduras. Most importantly, sand filtration gives us the ability to consistently lower the effluent NTU standard to below 1 NTU.

Theory

The basic premise of the stacked filtration system is that the flow of filtration is equal to the flow of backwash so that we can use normal plant flow to backwash the filter. A conservative estimate of backwash velocity requires that it be 10 times the normal filtration velocity. We achieve this requirement by stacking layers of filtration on top of each other. Each layer, or plane, consists of a set of inlet tubes that introduce water to a layer of 20 cm of sand. The water once filtered is then collected by a set of outlet tubes. Each layer is essentially its own filtration system. When you stack them on top of each other, area for backwash stays the same and you can technically backwash all of them with the same backwash water. Figure 2 Basic Concept of Stacked Filtration Operation and the mathematical derivations demonstrate this relationship.

Unable to render embedded object: File (Clear Well Basic Concept.png) not found.
Figure 1: Clear Well Basic Concept

Method

1) Review of existing filtration/backwash technology and research
We conducted a literature and online review. We determined the flow rate needed to sufficiently expand and clean the sand filter bed. This will help us determine how high the clear well needs to be above the filter, how large the flow pipes should be, and how much water should be in the clear well.
Research of Existing Work.

2) Develop a MATHCAD file that generates backwash and filtration design parameters
We needed this for both an actual AguaClara plant and a bench-scale or pilot plant model of the plant for testing.
[MATHCAD File and description.]

3) Experiments of bench-scale model to confirm design success
Bench scale modeling tests the effectiveness of a filtration design by shrinking the design parameters of the system (filter bed depth, filter bed surface area, and etc) to a smaller scale that is easier to test. For example, we would simulated a filter bed of 50 cm of sand with 5 cm of sand with the porosity and specific gravity of the sand being constant. This would also enable us to test the validity of the empirical equations that are behind our design.

Our Mathcad design created two designs; one which was a conservative approach (most commonly used), based on the simple hydraulics that the necessary velocity of the backwash is 10 times the velocity of filtration. The second design was based upon empirical equations, called the Weber Equation. The accuracy of the empirical fluidization velocity equations needed to be tested so we developed a bench-scale model of our filtration system and conducted an experiment measuring the expansion of a filter bed as backwash velocity is varied. We then compared the empirically calculated fluidization velocities with the actual fluidization velocities required.
Fluidization Velocity Experiment.

Results and Discussion

MathCad Results: Empirical vs. Simple Hydraulics (Conservative) Approach

Unable to render embedded object: File (ClearWellTopView.png) not found. Unable to render embedded object: File (ClearWellSideView.png) not found.
Figure 2: Agalteca Plant with Filter Designed from the Conservative Approach

1) Our design based on simple hydraulics will work. However, it is a very large filter (see exact dimensions in Figure 2) and will not be sustainable economically. The material cost for construction will be too high.
2) The design based on the empirical Weber equation is smaller and less expensive. However, the validity of the empirical equations is not yet certain, in spite of our Fluidization Velocity Experiment. Therefore more testing needs to be done in pilot scale models.
3) If the empirical equations are valid, then we can change parts of the design, by changing the sand parameters. For example, lower the dimensions of the clear well by lowering the backwash velocity by decreasing the d60 and specific weight of the media.
Unable to render embedded object: File (Orgd60.png) not found. Unable to render embedded object: File (Changed60.png) not found.
Figure 3: Small Change in D60 can fix the error at 30% Expansion by over 100%

4)An additional advantage to Clear Wells is that the distribution tank does not have to be below the filtration tank, and in fact, it could be the clear well as well.
Unable to render embedded object: File (ClearWellIsDistrTank.png) not found.
Figure 4: Distribution Tank can be the Clear Well Tank

Experiment Results
We had mixed results with regards to Weber's equation for filter bed expansion. At low levels of filter bed expansion, the Weber equation accurately predicted the fluidization velocity required to achieve the targeted bed expansion. As the target bed expansion increased, so did the degree of error. At 9% expansion, the degree of error was at 14%. At 30% expansion (our target expansion), the degree of error was 110%.
Unable to render embedded object: File (Error.png) not found.
Figure 5: Error Between Calculated and Actual Expansions

Sources of Error

Human error:
Despite our best attempt at being consistent (by measuring and marking heights on the test tube, while also holding a ruler on the test tube wall), there will always be human error in observing the bed expansion visually.
Fix: The next expansion experiment should use a camera so there is record of the heights at each flow rate, and also tape a ruler to the filtration bed wall, rather than holding the ruler or drawing it on.

Wall Friction:

We can attribute the increase in error as flow rate increased due to the increase in wall friction on the test vial.
Fix: We can minimize the wall and tube friction by increasing the size of our bench scale experiments.

Sand Properties & Parameters:

We might have used an incorrect D60 and porosity for the filter bed in our equations. We show on the Fluidization Velocity Experiment page how small changes in these values could easily account for the error.
Fix: For the next experiment those parameters should be tested for the sand or material before conducting experiments.

Preferential flow:

Despite our best attempt to keep the test tube as level as possible, we might have introduced preferential flow in our experiment causing an unbalanced backwash flow.
Fix: In the future, this hypothesis can be tested using dye. In addition, precautions can be taken, such as two levels could be clamped on the sides of the filter walls to ensure it is level or also use two clamps, rather than one.

Expansion Headloss:

The accuracy of our model also required the headloss occurring through the expanded bed to be more or less constant, which we did not have time to test (but is part of Recommended Future Research below).
Fix: Testing it would involve putting a pressure sensor into the system, connected on either end of the filter.

Recommended Future Research

Future Research should be devoted to the following objectives:

• Repeating the Weber Fluidization Velocity experiment with a larger scale bench model to see if error decreases.
• Repeating the Weber Fluidization Velocity experiment with multiple layer filter media.
• Testing the headloss in the system through the expanded bed to ensure it is constant
• Create pilot scale model to determine any remaining error in the design before creating a plant scale