M. Cross, T.N. Croft, D. McBride, A.K. Slone, and A.J. Williams Centre for Civil and Computational Engineering School of Engineering University of

Transcription

1 M. Cross, T.N. Croft, D. McBride, A.K. Slone, and A.J. Williams Centre for Civil and Computational Engineering School of Engineering University of Wales, Swansea 1

2 !"# $% #"&" " ' (" $ #") * ) $% $ % % " " " $% "$+, &% &% - $../.,0,& %' $ " # %1 1

3 Ka = f K = K f K fs K ft K sf K s K st K tf K ts K t a = a f a s f = f f f s A single code for all phenomena coupling direct a t f t! K f n a f n = f f n - g 1 (a s n-1,a t n-1 ) K s n a s n = f s n - g 2 (a f n,a t n-1 ) K t n a t n = f t n - g 3 (a s n,a t n ) Impact of using distinct solvers for each phenomenon ( " 2

4 ! " # " $ %&&! ' " $ $(!! )$ $ %&& ' " (! %&&! ' Interpolation from one set of variables to another => compatibility of mesh Virtual single database of mesh data & simulation variables Solver strategy - direct vs iterative - Eulerian vs Lagrangian Phys-A Phys-B Is coupling strategy compatible with scalable parallelism, EVEN if software components are parallel 1

5 (* +!$"!! #, (!! # $! -!!! ".)./! / / # 0! 1 #! #! 2 3" # #!/!, ! 9/ ! 9/ /! " ( #!)#&& 2

6 !"#$ %%&& #'#"($%%& & )*("! %%&& + *!("!'*%%&& ("!'*, %%&-&. %%&& :../*0)%%& & ( %%&& #* - $ ANSYS/Multi-physics ABAQUS ADINA ALGOR AUTODYN CFD-ACE DYNA COMSOL LMS software MSC- NASTRAN PHYSICA+ STAR-CCM

7 !" #$! % &$! $! $' () % % *+$,# "! $( -) (" - ANSYS/Multi-physics ABAQUS ADINA ALGOR AUTODYN CFD-ACE DYNA COMSOL LMS software MSC- NASTRAN PHYSICA+ STAR-CCM

8 Alternative approach: Single Software Framework.% % (/ 0 (! "# $ ( ( "%" &% $ - % && % % "%&% $ ("%-1 % "'" ()& $" (2(" " ( *++( ", $ (- (" &% --. ("3 $ % )% &% &" $ / 01$2*'01$3, $ % 4& % $ 2()% % 567 $ 8*" " )4,' *", 2

9 9%" ): $&% 2 %! & % &%! # $5 $" $ % (" %1! ) 5" " ;&%% $ < &8. =.07$567 =01=0= !#- 3

10 Mixture of liquid steel and argon injected into rectangular mould Liquid metal flux sits on top of mould Water cooled mould extracts energy forming a solid steel shell Continuous withdrawal B.G. Thomas 1

11 φ + u. φ = 0 t!" "! #"! $ #" #!!" "!!"%"! # "# & &"!! #'!"! & &" Top view End View z Solid regions appear in blue 2

12 '''!("! &$)&& Eval integrated path L particles Embed with mass flux effects Incorp in cont CFD code 3

13 Computations were also performed to estimate the effects of EMB on the free surface. For this the Maxwell equations were solved, which with the usual MHD assumptions, lead to: Continuity of magnetic flux: Ohm's Law for conducting metals Magnetic Transport, or Induction equation Lorentz force:.b=0 J = σ( E +U B ), where E = - φ = (U B )+ η B t where, η = σ µ 2 1 m Note: Terms containing the velocity U, are only important when Rm (=LU/h)> 1 1

14 Fluid behaviour under EMB conditions Flow suppressed here B=0.4T B=0T Coupled EM-flow calculations For most practical calculations in metals processing: The EM field influences the flow and thermal fields BUT the thermo-fluid phenomena has little influence of the EM fields Hence, essentially one way coupling So calculate the EM field and calculate the thermal and flow loads in the CFD calculation Can implement above model in any good CFD code! 1

15 Welding processes simulation - natural multi-physics Processes involve: free surface flow electromagnetic forces heat transfer with solidification/melting development of non-linear stress Ideal candidate for multi-physics modelling T-Junction arc weld simulation 2

16 I I Experiment and simulation FEMGV Greenwich University 28 FEB 2000 Model: T_J CASE1: PHYSICA Results Step: 1 TIME: 0 Nodal LFN Max = 1 Min = 0 Y Z X E-1 T-junction section, highlighting HAZ region R S V E T Y I of N G U R E the E N W C H the UNIVERSITY of GREENWICH Distortion of T-junction due to heat source Heat source 1

17 Weld pool dynamics Velocity vectors in crossection Lorentz force distribution in the weld-pool Distortion of T-junction due to heat source Distortion 2

18 Welding multi-physics BUT.. Welding involves: free surface fluid flow heat transfer and solidification/melting electro-magnetic fields non-linear stress BUT.. no coupling back: from thermo-fluids to EM field from stress calculation to thermo-fluids SO.. reasonably loosely coupled Generic Dynamic Fluid Structure Interaction Closely coupled multi-disciplinary problem Time & space accurate Very challenging in every respect. Issue of GCL CFD Mesh adaptation Traction boundary condition CSM Deformation Implementation of boundary conditions. Features of single software framework: Consistency of mesh. Single database & memory map. Compatibility in the solution approaches FV-UM. 3

19 Three Phase Approach CSD M d + Cd + Kd = d F s ( t) on Γ fs tn tn+1 CMD K d = F m m m u m on Γ fs t p = on fs Γ fs CFD PHYSICA Spatial Discretisation for closely coupled multi-physics Finite Element Unstructured mesh CFD Cell centred Or mixed CC- VB FV CSM Vertex based FV/FE Gauss Point Mesh Element x x x Node x Finite Volume Vertex Based x x x x x x x x x Finite Volume Cell Centred Control Volume Integration Point Control Volume 1

20 Dynamic fluid-structure interaction Targeted at problems involving flow induced vibrations Use dynamic structural equations and Navier-Stokes flow equations Wind direction Flow induced vibrations Dynamic response of structure without flow 2

21 Fluid Velocity and Pressure Movies At tip of wing Shear stress σ xy in wing 1

22 Bio-medical multi-physics modelling: the heart! Heart a multi-physics system, featuring interactions between: - electro-chemical system - fluid - structure Image from Geometry 0.045m 0.01m 0.01m 2D model Right ventricle Inlet Outlet 0.08m Wall 0.005m Blood 2

23 Electro-fluid-structure interaction Electrical Structure Fluid flow Mesh dynamics Multi-potential heart electrical activity model: comparison of results Clayton and Holden 02 PHYSICA Primary and secondary potentials Currents 3

24 Coupling electrical field to structural mechanics Assumption made that small strain model can capture behaviour of heart wall Dominant behaviour of wall is contraction/expansion Shearing effects negligible Elastic model Change in potential results in a change in tension in the heart wall To model tension we introduce an electric strain into structural mechanics equations Electric potential, deformation & flow patterns Results at various stages through a heart beat cycle 1

25 Parallel Multi-Physics Modelling Exploiting parallel cluster technology: the challenge MpCCI and other filter technologies Upside enables interaction at the code d base level Downside all data exchange must go via filter and is a compute bottleneck wrt scalability on parallel clusters Filter tech: Code A (CFD) Code B (FEA) Map onto Parallel cluster BOTTLENECK Map onto Parallel cluster 2

26 Parallel Multi-Physics Framework Simulations very Time Consuming need Parallel capability #$% & '"!" ()) Parallelisation approach Partition of 3D unstructured mesh by JOSTLE Uses mesh partitioning SPMD strategy with non-uniform workload Assumes a homogeneous load balance across the mesh: load balanced ( even no of cells per node) minimises sub-domain interface elements sub-domain connectivity matches processor topology of the parallel system 3

27 Multi-physics Simulation parallel issues Sub-domains have specific physics so partition must reflect this: non-uniform load/node Distinct physics uses distinct discretisation procedures: secondary partitions Sub-domains may change as problem develops: dynamic load balance Heat transfer Fluid flow Solid mechanics Strategy needs to address all the above issues Primary & secondary partitions Primary & secondary meshes Good primary & poor Secondary partition Good primary & Secondary partitions from JOSTLE 4

28 Parallel multi-physics: two level approach Implement a generic parallel version of Multi-physics code/ MDA codes without regard to in-homogeneity of the computational work over the mesh(es) defining the analysis domain #$% & Dump the load balancing into the mesh (re)partitioning task - JOSTLE_DLB Process as straightforward as possible!" Mesh Partition Involves large scale deformation of metal workpiece through interaction with one or more dies Multi-physics problem Flow/deformation of work-piece Metal Forming - Extrusion Heat transfer generated by internal friction Stress/strain in die(s) 5

29 Mixed Eulerian-Lagrangian Approach Workpiece Eulerian mesh Free-surface algorithm to track deformation Non-Newtonian material model Heat transfer plus energy generated by internal friction Die Lagrangian mesh Mechanical behaviour coupled with: Thermal behaviour in workpiece Fluid traction load from workpiece Governing Equations - Extrusion Coupled Thermo - mechanical problem CFD Heat transfer significant factor in deformation process Non-Newtonian viscosity model Plastic Norton Hoff law Heat Transfer - Friction between die and workpiece. Free Surface - Van Leer method CSM Static equilibrium equation linear elastic solid. Coupling at the workpiece/die boundary: Die subject to fluid traction boundary condition. Workpiece subject to a die velocity boundary condition. Dynamic meshes GCL. Fluid velocity relative to mesh movement. 6

30 Extrusion through U-shaped die mm Initial diameter = 200mm Bearing length = 2.5mm Punch speed = 5.85E-3m/s 4.76 mm mm Workpiece = 470 C Die = 450 C Air = 30 C elements nodes Temperature contours in extruding work-piece 7

31 Effective stress contours and deformation of die Parallel results Processors Run time (hours) Speed-up Single phase mesh partitions on 16 processors Itanium IA 64 cluster running Linux OS Eight nodes, two 733MHz processors per node Each node with 2 Gb memory & 2Gb swap space 8

32 Challenges Multi-physics OK where discretisation methods are coherent What about distinct methods FE-BE can be OK Continuum-particulate interaction fracturing What happens when it involves combustion, heat transfer, fluid flows, etc Can I do this scalably in parallel? ) Conclusion Multi-physics simulation is emerging in a commercially supported manner Most successful multi-physics is based upon loose or oneway coupling even then, heroic computing Close coupling in time and space another ball game - Key here are procedures for time & space accurate simulations; DFSI a key exemplar Multi-physics essentially compute intensive leads to challenge of parallel scalability for multi-physics simulation tools Can do for bespoke single software solutions, but for multi-code components, not so clear! Challenges for the future integrating components using essentially distinctive model paradigms & solver strategies 9