CFD Optimisation case studies with STAR-CD and STAR-CCM+

Transcription

1 CFD Optimisation case studies with STAR-CD and STAR-CCM+ Summary David J. Eby, Preetham Rao, Advanced Methods Group, Plymouth, MI USA Presented by Fred Mendonça, CD-adapco London, UK

2 Outline Introduction (Design) Optimisation & CFD CAD embedded verses CAD integration Examples: Static mixer Water jacket Manifold Fuel sloshing in a tank Summary I ll try to put emphasis on: Best practices Breakthrough Technologies - enablers

3 Design Optimization with CFD? If you currently use CFD in your design process, you are already performing optimization. Can design optimization be combined with CFD simulations? Current engineering design cycle is manual optimization with CFD. If you believe in manual optimization, develop a methodical design optimization approach. Ingredients required to develop a methodical design optimization approach: Accurate models (confidence). Efficient models. Automation of model construction (integration). Automation of solution post-processing (integration). Application of theoretical optimization framework (integration).

4 Introduction Design Optimization can be defined as: The search for Design Variable values that improve a set of Design Criteria. Design Variable examples: The radius of a duct, or Shape control vertices of a turbine blade. Design Criterion examples: The efficiency of a turbine, or The crashworthiness of a vehicle. Design Criteria are functions that are dependent upon the Design Variables.

5 Introduction Optimization algorithms: Gradient-based Strengths: quick convergence. Weaknesses: cannot escape local minima, requires convexity. Design of Experiment / Response Surface Method Strengths: quickly evaluate behaviour of design space, similar to typical design processes. Weaknesses: small number of design variables, assumes functional form of design space. Evolutionary Strengths: design space can be multi-modal, non-convex. Weaknesses: requires a large number of evaluations. Hybrid Combines multiple optimization approaches.

6 Introduction Optimisation algorithms: Requires multiple design evaluations. Performance of an optimisation algorithm. Number of required design evaluations. Choice of design variables. Computational resources for multiple CFD analyses. Common Day CFD simulations (6-12 hours, regardless of time period).

7 Introduction (enabler): CAD-integration CAD CFD CFD CAD-translation CAD CAD-embedding

8 Polyhedral Meshing (enabler) Optimize more, for less with polyhedra: Improved accuracy over tetrahedra. Improved convergence over tetrahedra. Improved efficiency over tetrahedra. Fully automated construction for complex geometries.

9 Polyhedral vs. Tetrahedral Meshing (mesh dependencies best practice) Efficiency + Accuracy + Easy-of-Use = Optimisation 7 MESH DEPENDENCY POLY TET 6 Delta P (kpa) hours < 3% error 6.3 hours > 5% error 10 hours hours Number of Cells

10 Introduction (best practice procedure) Overall design optimization methodology for Common Day CFD simulations (6-12 hours per evaluation): Verify CFD baseline model (confidence). Use engineering intuition to select design variables. Define design criteria. Perform design of experiment. Fit response surfaces to STAR-CD outputs. Approximate design criteria with response surfaces. Verify accuracy of response surfaces (confidence). Explore design space evaluating approximate design criteria from response surface. Verify accuracy of results (confidence).

11 Introduction Process Integration requires the automation of: Model building tool(s), simulation tool(s), and post-processing tool(s). Automation reduces man effort, maximizes compute effort. Optimisation driver can be modefrontier, Optimus, HyperSTUDY, isight FD...

12 Process Integration with Multiple Load Cases

13 File Parsing with Teach isight to modify the input files for STAR-Design (CAD) for each load case

14 File Parsing with Teach isight to extract the mass flow rate from the STAR-CD output (mass flow rate)

15 Mixer Problem Design Variables A grid of rectangular bars is placed in a circular pipe near the inlet to enhance the mixing of the fluid (water) at the outlet. γ β α Design variables: Angles between the bars for three sets of crosses (three in each set) β Side view (from +x) γ α

16 Static Mixer Utilize symmetry plane. Additional inlet - passive scalar (tracer) released. Passive scalar follows the fluid but does not affect the flow. Uniform velocity profile at the inlet. Velocities are equal for both inlets. Monitored passive scalar concentration at a section near outlet.

17 Static Mixer! body,move,0, -5, 0, 19, 20, 16, 17, 1, 2 workplane,create,face,118 workplane,create,face,206 workplane,create,face,74 workplane,select,2 body,rotate, 14,45,1,1.0781e-17, e-18,2.5, , body,rotate, 2,45,1,1.0781e-17, e-18,2.5, , body,rotate, 11,45,1,1.0781e-17, e-18,2.5, , body,rotate, 13,-45,1,1.0781e-17, e-18,2.5, , body,rotate, 1,-45,1,1.0781e-17, e-18,2.5, , body,rotate, 10,-45,1,1.0781e-17, e-18,2.5, , workplane,select,3 tool,command,modelbroker tool,command,procad body,unite,30,31

18 Mesh and Design Variation Movie (mesh)

19 Objective and Constraint Objective: Minimize standard deviation of passive scalar concentration at the section. ρ φ = φ = A passive _ scalar ρ φda A water ( φ φ ) Constraint: Minimum pressure drop from inlet pressure <= 6000 pa. σ area = ρ water A A 2 da

20 Pressure Variation for Each Design (Animation)

21 Scalar Concentration for Each Design (Animation)

22 Velocity Magnitudes Initial Optimized

23 Scalar Concentrations Initial σ = dp = 5911 Pa. Optimized Flow separates from wall in optimized design and promotes mixing γ β α σ = dp = 5727 Pa. γ > β > α

24 Manifold Optimization Design Variables: 6 cylinder cross section radii Swirl inducer thickness Design criteria: Objective uniform mass flow through each cylinder (minimize variance between each cylinder) Constraint pressure drop less than prescribed value

25 Manifold Optimization isight optimization: Sequential response surface approximation 11 evaluations Pressure drop constraint satisfied Mass flow variance between cylinders decreased by 12% Center cylinder radii decreased Swirl inducer thickness important

26 Water Jacket Optimization Design Variables: Fuel rail to water jacket cross section size Design criteria: Variance of mass flow through each cylinder m ν = i = 1 = i i= 1 12 m ( m m) i m

27 Explore Response Surface Response Surfaces

28 Water Jacket Results Decreased variance by 44% compared to the base case using a DOE study Baseline. 1 9 Best 5 Worst

29 Variation of Velocity Magnitude at the Exit for each Design

30 Water Jacket Optimization Near uniform flow through each cylinder

31 Fuel Sloshing in a Tank Transient Volume of Fluid (VOF) free surface simulation Tank is decelerated from 22 to 0 MPH over 1 second. 4 CPU hours per evaluation on single 3 GHz processor. 14 gallon tank ¼ full. 150,304 polyhedral cells (~500,00 tetrahedral cells) Grid morphing

32 Fuel Sloshing in a Tank Design Variables 4 Design Variables: Each baffle has a constant number of cutouts and cutout radii Design Criteria: Minimize peak total force transmitted across tank to chassis

33 Fuel Sloshing in a Tank DOE Results Lowest peak force Very high peak force

34 Fuel Sloshing in a Tank Design 12 Free surface leap frogs over baffles Slapping of free surface causes extremely high peak force Surface tension included Free surface climbs the inlet, Formation of air bubbles, Buoyancy included

35 Fuel Sloshing in a Tank Response Surface Modelling The sloshing of liquids in tanks can cause complex peak discontinuous forces. The relationship between the design variables and design criteria for sloshing liquids can be non-convex and multimodal. Often, an over-specified least-square linear response surface will capture trends in the space. In addition, the outlier response of design 12 is due to sloshing, slapping, slamming and leap frogging. If fuel gently rolls over baffles, then the relationship between baffle radii and peak force should be better behaved. We will therefore ignore the response of design 12 and refit a linear least square model to the data.

36 Fuel Sloshing in a Tank Explore Response Surface Approximated peak forces Optimized with response surfaces Best design -1, 1, -1, -1 Approximated peak force: 131 N.

37 Fuel Sloshing in a Tank Verify Final Improved Design Improved design : Approximated force: 131 N, calculated force: 136 N. Reduced force by 43% with respect to design 12 (worst) Reduced force by 6% with respect to design 10 (best) Fuel gently rolls over baffles

38 Summary CAE trends survey (interim findings) CFD Design Optimisation Simulation automisation set-up costs 350% more than a single analysis Allows to analyse O[50] designs Return of Investment (RoI) = order 10 CFD Topology Optimisation? Optimal topology nominally comes from a SINGLE simulation Costs 50% more to set-up Return of Investment (RoI) = order 100