Cyclone Separator Challenge

Cyclone separator

Aim:

The main aim of this study is to check the check the behavior of the cyclone separator CFD model on variation of the Discrete phase model boundary conditions. It would answer the How the different types of boundary condition influences the physical behavior of the computationally modeled cyclone separator. Also the behavior of the system is Analysed when we increase the number abrasive particles injected.

Introduction:

It is a device to separate the high density materials from low density fluid stream by creation of vortex to the flow mixture. The separation process purely depends on vortex formation and gravity. Due to its combined rotational and gravitational force particulates gets separated from the mixture. It would be designed in such a manner that helical movement of the stream from the top end to bottom end redirects to the outlet at the top, taking the central path parallel to the axis of the cylindrical/conical drum. The bottom outlet makes the pathway of the high density particulates which is separated from the less dense fluid stream.

 Fig 1: Geometrical model of cyclone separator

The above figure shows the geometrical model of the cyclone separator which we are considering for our study. Few parts of it were highlighted in this figure

Solving & Modelling approach:

In this study, we are going to work on the modelling of this physical happening in cyclone separtor using CFD tool and going to check the behavior of the CFD model of cyclone separator for various Discrete Phase Model (DPM) boundary condition as a phase I study. As the stream have low density fluid and relatively high density discrete particulates, we have to incorporate discrete phase model in the CFD model to capture the physics of cyclone separator with particle injection as discrete phase.

To check the influence of inlet injection boundary condition a mesh model is developed and 4 type of CFD models were developed over it. Each CFD model developed with the each of the BC’s present under DPM. So the output of the each CFD models can be compared and the influence of the DPM BC’s can be summarized.

As a Phase II study the additionally two mesh models other than base model(eg., Case 1) is developed in the increasing order of mesh as MM Case 2 and MM Case 3. By increasing the mesh count the inlet surface also gains more number of cell surface which in turn increases the number of particles injected in the domain. No of cell surface at inlet surface = No of particles injected. In this way we get 3 types of mesh model including Case 1 as one of the mesh model.

Pre-processing and solver setting:

As the geometry is not that much complicated, the fluid wet domain is easily extracted from the geometrical model in spaceclaim using volume extract-edges. The volume extracted is taken to the mesh modeller and the domain is meshed with tetrahedral volume mesh. In this study we are not much concentrated on the advanced mesh options. The mesh in an acceptable way is enough for our study so no any specific options were used and the entire domain was meshed uniformly with same meshing parameters. Only for Phase II study mesh count improvement done by decreasing the mesh size value to get 3 differernt mesh models in total.The below Table 1 shows the meshing details of mesh models.

 Table 1: Mesh model details

 Fig 2: Mesh metrics - Element quality of Case 1.

 Fig 3: Mesh metrics - Element quality of MM Case 2.

Fig 4: Mesh metrics - Element quality of MM Case 3.

The element quality of meshing of 3 mesh models proposed is shown above in Fig 2, 3 & 4. On examining the Mesh metrics plot the quality of the elements in all the 3 mesh models are fairly good enough to solve the problems on those meshes. In all 3 cases the maximum number of meshes element quality is nearby 0.88 and very less number of elements are below 0.63 and the influence of low quality elements are very minimal in the accuracy of solution.

Thus developed mesh model is loaded in fluent. The fluid domain is modelled in to a computationally modelled cyclone separator by enabling certain solvers and models with relevant materials assigned to the corresponding domains and particles injected. Pressure based, absolute velocity formulation solver is enabled with gravity enabled in negative y direction for 9.81 m/s (acceleration due to gravity). The model is solved in steady state with swirl dominated RNG k-e (2-equ) turbulence model with default parameter settings. As the abrasive particles are smaller in size which is coming as a mixture with air the discrete phase model is also enabled with injection settings. Discrete phase interaction with continuous phase is enabled, step length factor is assigned to a value 5. The particle injection velocity is set to 3 m/s, particle diameter is set to 5e^-6m.

Results:

Fig 5: Scaled residuals plot of Case 1

Fig 6: Scaled residuals plot of Case 2

Fig 7: Scaled residuals plot of Case 3

Fig 8: Scaled residuals plot of Case 4

Fig 9: Scaled residuals plot of MM Case 2

Fig 10: Scaled residuals plot of MM Case 3

The above figures are the scaled residuals of the models in our study. In all the cases the convergence criteria is set to 10 e^-3, but in all the cases the convergence not met. Even though the cases dint met the convergence criteria, all the residuals reached a steady pattern in the variation. This behavior of residuals can be considered as a convergence behavior. Not all the time we can make this type of convergence because getting a steady pattern above the 10e^-1 should not be treated as converged solution. That would mislead us in attaining the valid solution. So considering the steady variation pattern near the convergence range is advisable for the steady state analysis.

Each and every plots differs from each other especially the number of iteration. The reason is that the variation which we made in each model will influence the solution, so each models convergence reached at different iteration numbers. In some cases in our study we intentionally made the solution to run more iteration even though after reaching the steady variation pattern at the lowest range, this is to check whether the residuals getting shoot up or drop down after iterating a huge number of iteration. The Fig 7 which is the scaled residuals plot of MM Case 3 model, where we run the iteration up to 11000 numbers, but the change we observed here is the slight change in the pattern of variation. No shoot up or drop down is observed, and also the pattern travels nearby 10 e^-3.

 Table 2: Particle Injection data for varrying DPM boundary conditions.

The above Table 2 shows the behavior of the particle injection and its track in the domain of the cyclone separator with 4 different inlet DPM boundary conditions available. From the data’s in the table we can say that for this particular case of cyclone separator the dependency of DPM BC is not up to a greater extent. Even though reflect BC is having high count of particles which completes the cycle in the domain but other BC’s like trap, escape and wall jet also gives a considerable amount of particles completing the cycle. And also number of particles injected, number of particles escaped and incomplete number of particles don’t show any appreciable range of variation between these three models. The reason for this type of non-reactive BC is due to the working parameter of this particular type this physical model. The variation may be seen for some other cases like high velocity flows and high back pressure creating flows especially while giving “escape” for inlet injection BC. If particle enters in to the domain of high pressure stream there is possibility of particles to reverse the direction of flow marching towards the direction of inlet surface, which makes the particle escape through inlet surface itself. Similarly for “trap” BC also, but the only difference is it won’t escape through the inlet surface but it would lost its particle tracking in the domain and the particle is recorded as trapped particle.

As a conclusion of phase 1 study the available physical model with the working parameters is non-reactive to the inlet DPM BC. In some other physical models the influence of the DPM BC may be evident.

In the motive of investigating this type of non-reactive characteristic, particle track residence time profile on each model is considered here.

Fig 11: Particle track for Case 1 Model.

Fig 12: Particle track for Case 2 Model.

Fig 13: Particle track for Case 3 Model

Fig 14: Particle Track coloured with particle residence time in Case 4.

 The above figures Fig 8, 9, 10, 11 shows the particle track profile of each DPM BC models. In each plot the residence time was clipped between range of 0 to 10secs. With this type of common comparison platform we can also compare the particle flow track behavior between each models. While going through these figures we can say that residence time of those particles in the domain have a very good similarity. Influence of inlet DPM BC is not up to the considerable level. One thing is clear, in this particular model the each and every particles exit through outlet 1 only and some particles ended incomplete in the domain itself. We cannot say this type of happening may happen in actual system, this incomplete particle track status is because of the situation that the particle struck somewhere in the domain and the solver could not bring that back in to the track in the stipulated time period.

 Table 4: Particle details in the domain for 3 Mesh Models.

In the next phase 2 of study the three types of mesh model of three different mesh count is taken in to study the particle injection based on the mesh count. The above Table 3 shows the particle injection details of those three mesh models. From the table, it is evident that the number of particles injected depends on the mesh count purely. As the mesh count increases number of particles also gets increased, in other words we can say that number particles injected inside the domain is same as that of the number of cell surfaces at the inlet surface.

Fig 15: Particle Track coloured with particle residence time in MM case 2.

Fig 16: Particle Track coloured with particle residence time in MM case 3.

 Particle track of Case 2 & Case 3 mesh model coloured with particle residence time between 0 to 10 secs clipped range plot is shown in Fig 15 & 16. The same type of plot of mesh model Case 1 is shown in Fig 11. On comparing those figures and Table 3 increasing particles or in other words increasing gas particle mixture, flow choke arises which makes the particles movement slower which in turn increases particle residence time especially in MM Case 3 lot of particles lost in domain which is untracked after stipulated time step. In 920 injected particles 509 particles lost in the domain, but on comparing MM case 1 and MM Case 2 particles lost in domain is 98 and 96 respectively which is more or less same. So we cannot say when particles increases particle lost in domain also increases. This may be due to the error in the solving solution with this type of discrete phase model with swirl dominated flows. To investigate that another one mesh model with 8 e^-3 m mesh size is developed with 708041 elements is developed and the solution is solved for same problem. The particles data is monitored for every 100 iterations which is shown in Table 4. From that table we can say that there exists an abnormal flutuations existing in particle data iteration by iteration. So the solution convergence becomes questionable here. This is the problem of convergence only, so we need proper convergence criteria, if it dint reach the criterian we should have acceptable manual convergence characteristic to conclude the solution is convereged or not.

Table 4: Fluctuation of particle details in the domain during iteration.

Conclusion:

In this study we can summarize totally 3 number solution. They are as follows

1. Discrete phase boundary condition in the inlet surface is not going to affect the solution in the considerable range.

2. Increasing the particle in the domain beyond a limit will reduce the Cyclone separator efficiency.

3. Convergence characteristics for manual convergence should be framed more conciously.