Pressure Drop Evaluation in a Pilot Plant Hydrocyclone Fabio Kasper, M.Sc. Emilio Paladino, D.Sc. Marcus Reis, M.Sc. ESSS Carlos A. Capela Moraes, D.Sc. Dárley C. Melo, M.Sc. Petrobras Research Center
Outline Introduction Objectives Physical Model Numerical Model Computational Details Results Discussions Main Conclusions Future Work
Introduction Hydrocyclones Extensively used in the oil industry to separate oil-water mixtures High efficiency High flow rates Compact equipment Challenging CFD test case High Re number Known turbulence modeling issues Courtesy of PETROBRAS S.A.
Objectives Compare the numerical results with field data obtained in a pilot offshore plant PDC (Pre-de-oiler cyclone) Kvaerner hydrocyclone Evaluate the overall pressure drop given by different modeling approaches PDC Kvaerner hydrocyclone Pilot plant facility Courtesy of PETROBRAS S.A.
Objectives Understand the flow pattern Analysis of velocity profiles and its influence on the pressure drop and efficiency Develop a methodology to perform parametric analysis Model needs to be robust and fast Help at pre-design stages
Physical Model PDC Kvaerner hydrocyclone Geometric parameters D cylinder = 70 mm Total height 1.8 m Operating Conditions Fluid: Water Inlet mass flow rate 93.1 kg / min Overflow split 41% Re 2.10 5 Overflow Inlets Underflow
Numerical Model Numeric details Strong gradients Numerical dissipation Highly swirling flow Physical instabilities Transient simulations General case: Multiphase with different oil concentrations On this study we will focus on the single phase (water) problem
Numerical Model Meshing Hexa elements Use of 2 GGI s to save on the total number of nodes Overflow and Underflow regions Pressure tap points Grids built in ANSYS ICEM HEXA Grid 1: 1.3mi nodes Grid 2: 2.5mi nodes Grid 3: 4.2mi nodes Maximum aspect ratio < 160 (axial refinement needed)
Numerical Model Grid 1 details Overflow region GGI PDC body Underflow region GGI
Numerical Model All runs were carried out with the ANSYS CFX-10 solver Boundary conditions 1 inlet and 2 outlets Specified mass flow boundaries Smooth walls Modeling approach Laminar RSTM SSG LES Smagorinsky (C S = 0.1)
Numerical Model Numerical details 2nd order transient scheme Convergence criteria 5.10-5 RMS within each time step Time step size Adaptive time step based on the maximum CFL number Usually between 2-4 internal loops required Higher order advection schemes All runs with β=1, except LES where CDS was used Transient statistics Time averaged variables were calculated after flow was established
Computational Details Performed analysis CPU: Intel Xeon 3.20 GHz Parallel runs with MPI Turbulence modeling approach Grid 1 Grid 2 Grid 3 Number of CPU s 8 10 16 CPU cost (Days) CFL range * SSG 5 5 4 10-20 LES 15 21 20 < 1 Laminar 11 13 12 20-30 * set to converge within 2-4 loops per time step
Results Pressure drop comparison Targets
Results Grid sensitivity analysis 0.120m
Results Tangential velocity profiles on Grid 3 0.160m 0.150m 0.140m 0.130m
Results Tangential velocity profiles on Grid 3 0.120m 0.110m 0.100m 0.090m
Results Axial velocity profiles on Grid 3 0.160m 0.150m 0.140m 0.130m
Results Axial velocity profiles on Grid 3 0.120m 0.110m 0.100m 0.090m
Results Eddy viscosity ratio comparison (µ t / µ) More than 100 times bigger SSG LES
Results Turbulence model smearing effect Starting from a Laminar result as initial guess
Results Tangential velocity on Grid 2 ( Laminar )
Results Pressure on Grid 2 ( Laminar )
Results Axial velocity on Grid 2 ( Laminar )
Discussions Runs were long enough to get statistical established flow conditions *LES runs were likely stopped prematurely Discrepancies still found for the underflow pressure drops even in the fine mesh (Grid 3) Apparently because of not enough axial mesh refinement Grid independent solution still not reached in the LES and Laminar models Might be already reached for the SSG model
Main Conclusions Turbulence modeling for this specific Hi-Re hydrocyclone RSTM-SSG seems to be too diffusive Smears the velocity profiles under predicting the experimental pressure drop LES is too expensive (high Re number) Fine mesh resolution requirement CDS scheme used Requires very low time steps CFL < 1
Main Conclusions The Laminar approach was the one who predicted best results whem compared to operational conditions Conceptually inconsistent It seems that numerical diffusion stabilizes the solution preventing the solver to diverge Cheaper when compared to the other turbulent approaches
Future Work Research is still going on Search for some systematic and error and sensibility (e.g. BC, fluid properties) analysis is now being done. THANK YOU!