Application of V&V20 to Quantify Validation Uncertainty and Model Comparison Error for Industrial CFD Applications: Multi-Phase Liquid-Solids Mixing in Impeller-Driven Mixing Vessels Michael Kinzel Leonard Peltier Andri Rizhakov Jonathan Berkoe Brigette Rosendall Mallory Elbert Kellie Evon Kelly Knight Bechtel National, Inc. Advanced Simulation & Analysis ASME V&V Symposium, May 22 nd, 2013 V&V2013-2266
Outline Introduction and Background Validation Case CFD Model Description Methods and Results Conclusions page-2
Introduction and Background Growing interest in CFD modeling of granular flows, i.e., Bechtel involved in the following Waste water treatment Slurry flows (mining operations) Scouring (offshore platforms, civil) Dust/sand control (renewable energies) Many other applications Granular multiphase CFD models are maturing Newer branch of CFD (rapidly growing) Limited literature of successful validation Opportunity to qualify granular, multiphase CFD for real application Introduces needs for applying V&V 20 Standard Enables qualification of CFD models for complex, multiphase flows Goal is to expand usability of CFD in engineering page-3
Validation Case 4 Baffles Experiments based on bladed mixing vessel (Montante et al, 2002) filled with solids Tall vertical vessel 3-bladed impeller to mix solids with baffles 0.48m 2 different particle sizes each loaded at a concentration of 100 g/l. Mean sizes are: (1) 675 µm (2) 327 µm Vertical concentration profiles measured and used as V&V data (little reporting of experimental data uncertainties) H v =1.44m D i =0.195m D v =0.48m 0.48m 0.24m Montante, Giuseppina, Donato Rondini, André Bakker, and Franco Magelli. "CFD predictions of solid concentration distributions in a baffled stirred vessel agitated with multiple PBT impellers." In Proceedings of the CHISA Conference. 2002. Montante, G., G. Micale, F. Magelli, and A. Brucato. "Experiments and CFD predictions of solid particle distribution in a vessel agitated with four pitched blade turbines." Chemical engineering research and design 79, no. 8 (2001): 1005-1010. page-4
CFD Model Model Based on ANSYS Fluent 14.0 Mesh/model setup provided by Fluent (Montante, 2004) Base mesh cell count: 116,000 cells ¼ domain w/periodic BCs Bechtel UDF for granular, multiphase drag law Mesh refinement Refinement 1: Cell-size halving Refinement 2: Cell-size halving (refined only in regions with high gradients in solids content to reduce cell count) Montante, G., Bakker, A., Solid-Liquid Multiphase Flow Validation in Tall Stirred Vessels with Multiple Impeller Systems, Fluent Inc. Technical Note 253, Aug. 2004. page-5
Methodology ASME V&V20 Quantification of model comparison error (E) E = S D validation uncertainty (u val ) u vvv = 2 2 2 u nnn + u iiiii + u D S D model solution experimental result u nnn uncertainty from numerics u iiiii uncertainty from uncertain input parameters u D uncertainty from uncertain experimental data Probability Plot for δ model The true model error (δ mmmmm ) is characterized E - u val E + u val δ mmmmm [E u vvv, E + u vvv ] E -3.00-2.00-1.00 0.00 1.00 2.00 3.00 page-6
Baseline V&V Approaches: 327 micron u num Determination u num =1.15 (CBa se CEx trap )2 Apply Richardson s extrapolation for all elevations => C Extrap Base Resolution Refinement 1 Refinement 2 Cells 116640 933120 1691493 Used in rest of cells 1/2 size Cells 1/2 size of Refinement 2 Comment paper of Base in regions with large gradients page-7
Baseline V&V Approaches: 327 micron u input Determination u input = (CBa se C D110 )2 +(C Base C Rho110)2 +(C Base CRP M110 ) 2 Perturb multiple input quantities 10% (one sided, all positive) Perturbed the following input quantities: Particle diameter RPM Particle density Directly computation of perturbations page-8
Baseline V&V Approaches: 327 micron Final Result (327 micron) u input determined earlier u num determined earlier u D ~ 10 (g/l) (expert opinion) 84% of data fall within characteristic uncertainty interval Evidence of issues with experimental data. Average C is 87 g/l which is less than the reported loading implies bottom solids, experimental error, or wrong CFD initial condition. page-9
Baseline V&V Approaches: 675 micron Final Result (675 micron) u input determined u num determined u D ~ 10 (g/l) (expert opinion) 41% of data fall within characteristic uncertainty interval Evidence of issues with experimental data. Average C is 68 g/l which is less than the reported loading implies bottom solids, experimental error, or wrong CFD initial condition. page-11
Baseline V&V Approaches: Final E & U Assessments 327 micron: uncertainty embraces the model comparison error implies no significant model bias 675 micron: uncertainty is often less than the model comparison error implies possible significant model bias Normalized Elevation, z/h page-12
Conclusions Applied V&V 20 to an impeller-stirred mixing vessel Quantification of model comparison error and uncertainty for two conditions. Results suggest that model fidelity changes with particle size, i.e., lower comparison error for 327 micron than for the 675 micron. Interpretation should be viewed with caution Reported 675 micron experimental results are less than reported loading, indicating bottom solids, experimental error, or wrong CFD initial condition. It is possibly (and probably likely) that u D is substantially different between the varied particle sizes. This result underscores the need for more complete collection of validation data for granular, multiphase flows and/or better documentation of the experimental results. page-13