1 st Automotive CFD Prediction Workshop Zhenyu Wang 1 and Lian Duan 2 - PowerPoint PPT Presentation

1 st Automotive CFD Prediction Workshop Zhenyu Wang 1 and Lian Duan 2 1: Research Scientist, Simulation Innovation and Modeling Center 2: Associate Professor & Honda Endowed Chair in Transportation Computational resources by 1

Summary of Submission Ø Case 1: SAE Notchback 20 deg Code Method Spatial Scheme Temporal Scheme Criteria 2 nd order STAR-CCM+ URANS 138 convective flow units 2 nd -order upwind (v 13.02.011) (realizable k-e) ( ΔtU/L = 1.43x10 -3 ) (28.4 for time-averaged statistics) 2 nd order ( ΔtU/L = STAR-CCM+ URANS 138 convective flow units 2 nd -order upwind (v 13.02.011) (k-omega SST) 1.43x10 -3 , 2.86x10 -3 ) (28.4 for time-averaged statistics) 2 nd order STAR-CCM+ Hybrid bounded 104 convective flow units SST-DDES (v 13.02.011) central difference (ΔtU/L = 2.86x10 -3 ) (28.4 for time-averaged statistics) 2 nd order STAR-CCM+ Hybrid bounded 104 convective flow units SST-IDDES (v 13.02.011) central difference (ΔtU/L = 2.86x10 -3 ) (28.4 for time-averaged statistics) • Simulations done with Cartesian trimmer mesh for RANS and DES provided by workshop committee RANS DES Ø Case 2a: DrivAer Fastback (Total Cells: 4.1 M) (Total Cells: 29.1 M) Code Method Spatial Scheme Temporal Scheme Criteria 2 nd order STAR-CCM+ Hybrid bounded 10.4 convective flow units SST-DDES (ΔtU/L = 1.74x10 -4 ) (v 13.02.011) central difference (1.74 for time-averaged statistics) • Simulations done with Coarse , Medium and Fine grids provided by workshop committee Coarse Medium Medium (Total Cells: 93 M) (Total Cells: 165 M) (Total Cells: 258 M) 2 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Computational Setup Outlet: Free Outlet Computational Domain & Boundary Conditions Roof: Symmetry Plane Model Name: SAE Notchback 20 deg Right Wall: Symmetry Plane Left Wall: Symmetry Plane Vehicle Dimension: 840mm (L) x 320mm (W) x 240mm (H) Vehicle Frontal Area: 0.076m 2 Floor: No-slip Wall Inlet: Velocity URANS: DES: Ø Solver: Star-ccm+ 13.02.011 Ø Solver: Star-ccm+ 13.02.011 Ø Inlet freestream velocity: 40m/s Ø Inlet freestream velocity: 40m/s Ø Inlet turbulence Intensity: 0.2% Ø Inlet turbulence Intensity: 0.2% Ø Wall Treatment: All y+ wall treatment Wall Treatment: All y+ wall treatment Ø Time step (dt): 3×10 -5 s and 6×10 -5 s Ø Ø Time step (dt): 6×10 -5 s Ø Inner iterations: 15 and 10 Ø Inner iterations: 10 Ø Turbulence model: SST (Menter) k-omega Ø Turbulence model: SST (Menter) k-omega and k-epsilon realizable IDDES and DDES Wall Y+ Wall Y+ Min Wall Y+: 0.06 Min Wall Y+: 0.09 Max Wall Y+: 127.5 Max Wall Y+: 83.1 3 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Force Prediction Zero Yaw SST URANS SST URANS RKE URANS Exp. IDDES_SST DDES_SST Angle (dt = 6e-5 s) (dt = 3e-5 s) (dt = 3e-5 s) C l -0.035 -0.07 -0.07 -0.075 -0.076 -0.0785 C d 0.207 0.194 0.193 0.197 0.226 0.198 C m -0.075 -0.066 -0.067 -0.065 -0.138 -0.124 Drag Coefficient Moment Coefficient Lift Coefficient Ø Force prediction of URANS is insensitive to time step Ø Good predictions in C d for all modeling techniques Ø The origin for moment calculation is at X=-10mm, good predictions in C m for all URANS Models Ø All models give erroneous predictions of C l 4 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Pressure Prediction Time-Averaged Pressure Coefficient Experiment URANS_SST k-w_dt=6e-5s URANS_SST k-w_dt=3e-5s Time step change has little influence on prediction of pressure distribution Ø URANS and DES give similar predictions in Cp IDDES_SST k-w_dt=6e-5s DDES_ SST k-w_dt=6e-5 Ø Apparently good correlation between URANS/DES and experiment Ø DDES seems to overpredict the strength of trailing pillar vortices 5 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Pressure Prediction Time-Averaged Pressure Coefficient at Mid-Plane (Y=0) 840mm 20° 30° Ø Predictions of URANS 240mm and DDES correlate well with the experiment Ø C p predicted by IDDES X deviates appreciably Z from those of other models and the experiment in regions close to the notchback 6 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Pressure Prediction Time-Averaged Pressure Coefficient at the Base Surface (x = 420mm) Z = -180mm Z = -150mm Z = -120mm Z = -90mm Z = -60mm Y Z = -120mm Z = -180mm Z Ø URANS provides equally well or slightly better predictions than the DDES Ø IDDES shows largest significant deviation from the experiment Z = -60mm 7 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Velocity Prediction Inspection Locations on the Mid-plane (Y = 0) X=0mm X=-150mm X=-300mm X=-450mm X = 0mm X = -150mm X Z Ø All the turbulence models have similar results over the roof (x = 0mm) Ø URANS gives similar velocity profiles to DDES at all locations X = -300mm X = -450mm Ø IDDES deviates from the other three models in the backlight/boot- deck/base regions 8 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Velocity Prediction Time-Averaged Velocity at Mid-Plane (Y=0 mm) Experiment URANS_SST k-w URANS_ke-realizable Ø URANS underpredicts flow separation over the backlight surface, while IDDES significantly overpredicts flow separation Ø DDES best predicts the extent DDES SST k-w IDDES_SST k-w of flow separation 9 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Q-Criterion URANS_SST k-w URANS_ke-realizable IDDES_SST k-w DDES_SST k-w 10 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – RMS of Pressure Coefficient Experiment URANS ke-realizable URANS_SST k-w Ø All models fail to predict DDES_SST k-w IDDES_SST k-w values of RMS Cp • Significant underprediction by URANS • Significant overprediction by IDDES • DDES model seems to give the right order of magnitude over the base surface 11 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 1 – Summary Ø URANS (URANS_SST & URANS_RKE) and DES (DDES_SST & IDDES_SST) were conducted for the SAE Notchback 20 deg model Ø Timestep insensitivity was achieved for URANS at Δt = 6×10 -5 s (ΔtU/L = 2.86x10 -3 ) Ø URANS results are largely insensitivity to turbulence models, although SST k-omega showed slightly more accurate prediction of flow separation over the backlight and bootdeck surfaces Ø URANS and DDES provided good predictions of mean pressure and velocity Ø All models failed to predict values of RMS Cp Ø IDDES significantly overpredicted flow separation over the backlight surface, leading to the large deviation in the prediction of mean pressure and velocity over the backlight, bootdeck, and base surfaces from the experiment Ø DDES performed as the best option in accuracy among all the models, although the RMS of Cp shows significant discrepancy from the experimental data 12 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 2a - Computational Domain Computational Domain & Boundary Conditions Outlet: Free Outlet Left Wall: Symmetry Plane Roof: Symmetry Plane d n u o r G y r a Right Wall: Symmetry Plane n o i t a t S : r a f _ d a o R Inlet: Velocity Road_near: Moving Ground Model Name: DrivAer Fastback Vehicle Dimension: 4.6 m (L) x 1.778 m (W) x 1.408 m (H) Vehicle Frontal Area: 2.16 m 2 13 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 2a – Numerical Settings Coarse (Total Cells: 93M) Numerical Settings: Ø Solver: Star-ccm+ 13.02.011 Max Wall Y+: 1.54 Ø Inlet freestream velocity: 16m/s Ø Inlet turbulence Intensity: 0.1% Ø Wall Treatment: All y+ wall treatment Ø Time step: 5×10 -5 s Ø Inner iterations: 10 Ø Total physical time: 3.0s Medium (Total Cells: 165M) Ø Turbulence model: DDES-SST Max Wall Y+: 1.63 HPC Info: Ø Processor: Intel E5-2680 v4 Ø Cores Per Node: 28 Ø Clock speed: 2.4GHZ Medium (Total Cells: 258M) Ø Memory Per Node: 128GB Ø Cores Used: 840 Max Wall Y+: 1.69 Ø Total Run Time(s): 298,800 (Medium mesh) Ø Total Iterations: 600,000 (Medium mesh) Ø Time per Iteration Avg:0.498 14 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 2a – Force Prediction Mean Lift & Drag Coefficients C l _body (without C d _body (without C l _total C d _total tires and wheels) tires and wheels) Coarse Mesh 0.0453 0.0445 0.226 0.185 (93m) Medium Mesh -0.00706 -0.00798 0.242 0.190 (165m) Fine Mesh -0.0059 0.002 0.229 0.170 (258m) Ø Large differences observed in lift coefficient among coarse, medium, and fine mesh cases Ø Drag coefficient relatively insensitivity to mesh density 15 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 2a – Pressure Distribution Pressure on Vehicle Surface Coarse_Time-averaged Medium_Time-averaged Fine_Time-averaged Fine_Instantaneous Coarse_Instantaneous Medium_Instantaneous 16 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK

Case 2a – Wall Shear Stress Distribution Instantaneous WSS on the Vehicle Surface Coarse Fine Medium • Compared with the coarse mesh case, small WSS variations can be found between the medium and fine mesh cases. 17 1 st Automotive CFD Prediction Workshop, Dec 11-12, Oxford, UK