A Coupled Adjoint-Based Method for Aeroelastic Design in the Open-Source SU2 Suite

Transcription

1 A Coupled Adjoint-Based Method for Aeroelastic Design in the Open-Source SU2 Suite Rubén Sánchez*, Rafael Palacios imperial.ac.uk/aeroelastics Airbus DiPaRT 2017, 22 November 2017, Bristol, UK

2 Optimal Aeroelastic Design Advances in computational methods and tools have made it feasible to solve fully-coupled aerodynamic and structural problems during analysis stages. Joaquim R.R.A. Eurogen Conference (2017) However, we want to take this further. We want to design. Hesse and Palacios (2014) 2

3 1. Outline A brief summary of my talk: SU2 Suite A high-fidelity solver for multi-disciplinary analysis and design. Aeroelastic Design with SU2 Our work at Imperial College. What s next? Bringing SU2 Aeroelastic capabilities to the next level. 3

4 Why SU2? A state-of-the-art Solver Formulated for Optimal Design Under Active Development Big Open-Source Community 4

5 Sliding Meshes A state-of-the-art Solver C++/MPI core, HPC ready Finite Volume, Unstructured, Multigrid Compressible RANS Aeroacoustics Chemically reacting flows Turbomachinery Compressible RANS High-Order Discontinuous Galerkin Rotating Frame Detached Delayed Eddy Simulation (DDES) Solid mechanics FEA & More: High-Order, DDES, Incompressible flows, Heat transfer... Vortical Flows Acknowledgements to the SU2 community for the images 5

6

7 Under Active Development International development team in academia and industry 100+ branches ~300 active forks ~200k lines of code commits since January Annual Developers Meeting December 18-19, 2017, Stanford, CA Register Online! - 7

8 Big Open-Source Community ~ s on user list ~600 repo clones / month ~2000 repo visits / month Released on GitHub under LGPL Flexible and customizable Managed repository, guarantees the quality of the code Web hits at SU2 homepage by city. 8

9 Aeroelastics Team at Imperial Led by Dr. Rafael Palacios Our focus is in computational methods for the multidisciplinary analysis, design and optimisation of next-generation air vehicles and wind turbines We target the integration of aero-structural analysis, rigid body dynamics and flight control systems 9

10 Involvement in SU2 Oct 2014 June 2016 Collaboration with Stanford team starts 2014 We lead multi-team paper at ECCOMAS 2015 Dec 2017 July 2016 Collaboration with TU Kaiserslautern team starts 2016 Full code release 2017 July 2015 Oct 2015 June 2016 Oct 2016 Sept 2017 Implemented Non-linear FEA Solver Implemented Coupled FSI Solver FSI code official release Implemented Coupled Adjoints Paper on Coupled IJNME 10

11 Coupled FSI Solver Fluid Solver Mesh Solver Original SU2 Solver I N T E R F A C E Non-Linear Structural Solver SU2 with FSI Solver 11

12 Coupled FSI Solver Fully Implicit, coupled solver ALE Formulation Geometrical Non-Linearities Complex material model Consistent interpolation on domain-filling discretizations Elastic solver for mesh movement Fully differentiated using AD 12

13 Aeroelastic Analysis x1 Fluid Solver x2... Mesh Solver I N T E R F A C E Structural Solver J xn Define Input Parameters Coupled FSI Solver Evaluate Performance 13

14 Gradient-Based Optimal Aeroelastic Design dj / dx1 x1,1 dj / dx2 x2,1 J,0 J,1... dj / dxn Design Performance Optimal Design Compute Gradients J... xn,1 New Design s Performance New Design 14

15 Gradient-Based Optimal Aeroelastic Design dj / dx1 x3 x2 x1 dj / dx2 J,0... xi dj / dxn Design Performance Compute Gradients xn Large Number of Design Variables

16 Gradient-Based Optimal Aeroelastic Design dj / dx1 dj / dx2 J, dj / dxn Design Performance Compute Gradients Finite Differences, 1st order 16

17 Gradient-Based Optimal Aeroelastic Design dj / dx1 dj / dx2 J,0 2x 2x... dj / dxn Design Performance Compute Gradients... 2x Finite Differences, 2nd order 17

18 Gradient-Based Optimal Aeroelastic Design dj / dx1 dj / dx2 Fluid Adjoint Structural Adjoint J,0... Mesh Adjoint dj / dxn Design Performance Compute Gradients Coupled Adjoint Method 18

19 How do we compute Coupled Adjoints? 19

20 Computation of Coupled Adjoints... Adjoint Variables

21 Computation of Coupled Adjoints...

22 ...based on Algorithmic Differentiation (AD)

23 Twitter [CC BY 4.0], via Wikimedia Commons What do we mean by RECORDING? Variable Variable x1 x2 double double Operator cos cos(x1) Operator * cos(double) x2 * cos(x1) product<double,cos(double)> Solution y = x2*cos(x1) 23

24 Twitter [CC BY 4.0], via Wikimedia Commons What do we mean by RECORDING? -sin(x1)x2δx1 cos(x1)δx2 Variable Variable x1 x2 -sin(x1)x2 cos(x1) Operator cos cos(x1) Operator * x2 x2 * cos(x1) Solution y = x2*cos(x1) 24

25 Reverse Iteration

26 Comparative of Coupled Adjoints Conventional Adjoints AD-based Adjoints Require exact linearisation of the primal FSI solver and sub-solvers Exact linearisation no longer needed, approximate Jacobians are acceptable (non-linear problems!) Require analytical derivation of all the cross-discipline terms Automatically accommodates crossed terms and new features Require constant updates due to extensions of the code Require native FSI implementation Implemented in SU2! Efficient AD tools using Expression Templates 26

27 Further details... This method is fully described in a recent paper, published in the Int J Numer Meth Engng doi.org/ /nme.5700 Source code to be released in December 27

28 Applications of AD-based Coupled Adjoints Electro-mechanically Actuated Membrane Wings Optimal Actuation of Dielectric Membrane Wings using High-Fidelity Fluid-Structure Modelling AIAA paper , presented at Scitech 2017 RECORDING FSI Solution Input Non-linear Solver Output - Performance Large Deformations Non-linear Material Model Electric effects Viscous flow 28

29 Optimal Material Distribution in Very Flexible Structures - Large Deformations Non-linear Material Model Large number of Design Variables Viscous flow 29

30 Application to Wing Aeroelasticity Imperial College Team Supervision Dr. Rafael Palacios - r.palacios@imperial.ac.uk Xabier Romero Sáenz, MSc Thesis, 2016 SU2 Lead Developer Rubén Sánchez - r.sanchez-fernandez14@imperial.ac.uk PhD Researchers involved in SU2 Charanya Venkatesan-Crome - charanya.crome09@imperial.ac.uk Alfonso del Carre - alfonso.del-carre14@imperial.ac.uk Joel Ho Munn Onn, MSc Thesis,

31 A Coupled Adjoint-Based Method for Aeroelastic Design in the Open-Source SU2 Suite Rubén Sánchez*, Rafael Palacios imperial.ac.uk/aeroelastics Further questions? Contact: