Preliminary 3D CFD Simulations of a Hydraulic Flocculation Tank with a Single Baffle


Abstract

To extend the current 2D CFD models of flocculation tank in to 3D will greatly enhance our understanding of the fluid dynamics  inside the flocculation tank. In this report, four sets of preliminary 3D computational simulations were presented and discussed: first on investigating the difficulty of 3D models in converging to satisfactory residual; and then on comparing  the validity of 2D and 3D models by running 3D simulations with the model on which 2D assumptions were imposed.

Introduction

Compared with 2D models, CFD simulations in 3D provide closer numerical approximation to the real fluid flows in hydraulic flocculators,  which enable better quantitatively investigation of performance parameters of flocculation and thus lead to well-studied guidelines for the design, construction and operation of hydraulic flocculators of AguaClara treatment plants.

To extend the current 2D model into 3D, preliminary simulations were essential for evaluate the scope of the project before more conclusive tests. The goals were: to get familiar with using Gambit and FLUENT in 3D settings; to come up with effective approaches for 3D simulations; and to evaluate the relative validity of 2D models vs 3D models. In this report, the 2D model was extended to 3D by minor modifications of its mesh and FLUENT parameters.

Procedures

Investigation of Convergence

Four simulations were attempted to investigate the effect of mesh and FLUENT parameters on the convergence of 3D models, aimed at smaller residual in the resultant numerical solution:

  • Case 1: 1st order solver with convergence criteria of 10^-3 ;
  • Case 2: 1st order solver with convergence criteria of 10^-6, and to investigate the effect of    inlet turbulence intensity and hydraulic radius:
    • turbulence intensity: 10%, hydraulic radius: 0.004;
    • turbulence intensity: 1%, hydraulic radius: 0.04;
  • Case 3: Coarsened mesh in x, y and z directions,  1st order solver with convergence criteria of 10^-9 and 2nd order solver with convergence criteria of (solution obtained from the 1st order solver was used as an initial guess for the 2nd order solver).

Comparison of 3D simulation results with 2D

2D models approximate the real 3D flows by assuming all variables of fluid flows are uniform along the 3rd dimension, i.e. the z direction. Thus it was expected that the results from 2D models could be duplicated if imposing this "uniform" condition on the 3D model. In Case 4, this imposition was incorporated in the 3D model by applying periodic boundary condition to the walls in the 2D plane, i.e. the x-y plane, assuming that periodic repetition along the z direction was equivalent to "uniform":

  •   Case 4: Periodic boundary condition was defined at the walls along the x-y plane, 1st order solver with convergence criteria of  10^-6 2nd order solver with convergence criteria of  10^-6 (solution obtained from the 1st order solver was used as an initial guess for the 2nd order solver).

Click here for more detailed procedures of mesh generation in Gambit and problem definition in FLUENT.

Results and Discussion

The solution data of FLUENT contains the numerical values of a complete set of fluid dynamics variables at all the nodes defined by the mesh. These values can be used to calculate all relevant characteristic parameters. Presented in the following discussions are the plots of residual as a function of iteration steps and contours of energy dissipation rate.

Investigation of Convergence

Results are shown in Figure 1 and Figure 2 below. Figure 3 indicates the plane where the contour in Figure 2 were drawn, in grey.

a. Case 1

b. Case 2a

c. Case 2b

d. Case 3
Figure 1 Plots of residuals as a function of iteration steps
a. Case 2a

b. Case 2b
Figure 2 Contours of energy dissipation rate showing the effects of turbulence intensity and hydraulic diameters, drawn on the same scale

Figure 3 The plan where the contours were drawn, in grey
Observations:

  • Case 1 converged to 10^-3 within 300 iteration steps;
  • Case 2a converged to 10^-3 within 300 iteration steps, stopped converging after 500 steps and started fluctuating after 1000 steps;
  • Case 2a converged to 10^-3 within 200 iteration steps, stopped converging after 1000 steps and started fluctuating after 1000 steps;
  • Case 3 converged to 10^-9 with 1st order solver and then to 10^-6 with second order solver within altogether 8000 iteration steps.
  • There was no observable significance difference in the results between different turbulence intensity and hydraulic diameters at the inlet, except for less turbulent flow had lower minimum energy dissipation  rate at the entrance region.

Discussion:

The failure of convergence was caused by ill-conditioning of the problem. In CFD simulations, numerical values are solved for from large systems of linear equations with the basic fluid variables at the mesh nodes as unknowns. In Case 1, 2a and 2b, the mesh density along the xy plane and the z direction differed by a factor of 2~5. This difference could make some numbers in the intermediate steps of iteration smaller/larger, and could be propagated through the process of matrix operations into several orders of magnitude in the 3D system of ~10^5 cells. The extremely small number thus appeared could be close to machine precision, and couldn't reduce to any smaller number through iteration, which led to the stagnation of residual plots observed in Figure 1. As was show in Case 3 Figure 1c, after the mesh in xy plane was coarsened to the same order as the mesh in z direction, the residual converged to 10^-9 with 1st order solver and then to 10^-6 with second order solver after enough number of steps.The fluctuation in the tails of the residual plots could either be caused by fluctuation of extremely small numbers, or the fluctuation term defined in the turbulence model. Further experiments should be designed to verify the above hypothesis.

 The results were not sensitive to different inlet turbulence intensity and hydraulic radius. And this minor difference were overwhelmed around the 180 degree turning.

To create a more stable model that converge more accurate solution, the mesh in z direction must be refined along with proper boundary layers. And to explain the fluctuating tails in residual plots, more knowledge was need about both the algorithms used by FLUENT numerical solvers and the turbulence model.

Comparison of 3D simulation results with 2D

The results from Case 4 are shown in the following Figures.

a. Contour of energy dissipation rate of the 2D model

b. Contour of energy dissipation rate of the 3D model with periodic boundary condition
Figure 4 Comparison of energy dissipation maps of 2D and 3D models,drawn on the same scale

a. Contour of energy dissipation rate along the plane indicated in Figure 3

b. Contour of energy dissipation rate along the plan indicated in Figure 7
Figure 5 Contours of energy dissipation rate of the 3D model with periodic boundary conditions, drawn on its max/min scale
a. Velocity profile along the plane and line indicated in Figure 7, drawn on the max/min scale of the z velocity in the indicated line and plane

b. Velocity profile along the plane and line indicated in Figure 7, drawn on the max/min scale of the x,y,z velocity in the whole domain.
Figure 6 z velocity profile

Figure 7 The plane and line referred to in Figure 5 and Figure 6, in red
Observations:


  • 3D and 2D models resulted in different predictions of the shape and size of energy dissipation region after the baffle turning; (Figure 4)
  • 3D model predicted a higher maximum energy dissipation rate and a smaller energy dissipation zone; (Figure 4)
  • Energy dissipation rate was uniform along the z direction, as expected;(Figure 5)
  • There were still non-zero components of velocity in z direction, though insignificant, and not uniform along the z direction.

Discussion:

The difference of the prediction from 3D and 2D models indicates the importance of z component in the flow field, which made the assumption of the equivalence of periodic 3D model and 2D model invalid. When we used 2D model to approximate 3D flows, other than only uniform condition, along z direction, we were actually assuming the all the fluid variables only have x,y components and no z components. When applying the periodic boundary conditions to the walls in xy plane, although the assumption that periodic repetition in z direction were equivalent  to "uniform" could be valid, uniformity in z direction alone wasn't equivalent to "no components" in z direction. Thus these two models could not be good approximation to each other, and could generate significantly different predictions. As shown in Figure 6, there were still non-zero components of velocity in z direction, though insignificant in magnitude, and not uniform along the z direction.

Furthermore, the importance of z components could put the validity of 2D model as an approximation in question: in Case 4,  even small components in z direction could make significant difference in results from 2D model, let alone in the real flow.

However, the above hypothesis must be further investigated, ruling out all other possible causes of differences. Particularly, the effect of the length of the period must be investigated by vary the width of the flocculator. Ideally, periodic repetition with infinitely small period length is equivalent to "uniform".

Conclusions

To build a well-conditioned model that converges to accurate numerical solution, the mesh density in all 3 dimensions must be in the same order of magnitude, and refined enough to resolve the region where fluid flows vary violently; "uniform" in z direction is not equivalent to no components in z direction, thus 3D model with periodic boundary condition may not be equivalent to 2D model, which also put  the validity of 2D models in doubts.

Future Simulation Experiments

Possible future simulation experiments and research topics are:

  • To create refined mesh with proper boundary layers, and run simulations on SGI server, check grid convergence;
  • To investigate the fluctuating tail of residual plots;
    • Design numerical experiments to observe the fluctuation of extremely small numbers;
    • Use difference turbulence models and compare the shape of the tails;
  • Simulation with periodic boundary conditions with various period length
  • Use backstep experimental data to validate 3D models with periodic boundary conditions;
  • Find and compare with experimental data of free/confine jets .
  • No labels