1. IHPC-IMS Program on Advances & Mathematical Issues in Large Scale Simulation (Dec 2002 - Mar 2003 & Oct - Nov 2003) Tutorial II: Modeling and Simulation Techniques for Granular Materials Alberto M. Cuitiño Mechanical and Aerospace Engineering Rutgers University Piscataway, New Jersey cuitino@jove.rutgers.edu Institute of High Performance Computing Institute for Mathematical Sciences, NUS

2. Singapore 2003 cuiti ñ o@rutgers Collaborators Gustavo Gioia Shanfu Zheng

3. Singapore 2003 cuiti ñ o@rutgers Context Cascade of Length and Time Scales Particle Dynamics Discrete Element Monte Carlo Mesoscopic Models Continuum Length Scale (particle size) Time Scale (particle size/wave speed) 10 -3 - 10 0 One or less Phase Fields Fast Fourier FEM Atomistics Molecular Dynamics 10 0 - 10 1 1 to 10 3 particles 10 1 - 10 2 10 3 to 10 6 particles 10 2 - 10 3 10 6 to 10 9 particles > 10 3 Too many Bonding Energies Activation Barrier Phase Transformations Particle Packing Particle Rearrangement Particle Mixing Granulation Early Consolidation Particle Contact Stress Concentration Internal Localization Fracture Highly Confined Flow Consolidation Particle deformation Sintering Effective Behavior Design Models Sub-particle Resolution Range Supra-particle Resolution Range 10 -1 -10 2 10 5 -10 6 > 10 6 10 4 -10 5 10 2 -10 3

4. Singapore 2003 cuiti ñ o@rutgers Supra-Particle Resolution * If a “good”constitutive relation is provided Continuum INPUT From: Parameter fitting with macroscopic experiments OUTPUT Large regions / Long times Global Constrains (shapes, friction) Good for component DESIGN* Particle Dynamics Discrete Element Monte Carlo OUTPUT Limited to small regions and/or Short Times GAP INPUT From sup-particle resolution simulations (Multiscale)sup-particle From DIRECT Experimental Particle-Level Studies (Higashitani, Granick, …)

5. Singapore 2003 cuiti ñ o@rutgers Aside: MD/DEM Domain Computation Time/Physical Time PROPORTIONAL  pressure*number of particles/particle size Ct/Pt PROPORTIONAL p*n/d P o (mixing range) P = 10 3 P o (consolidation range) 100 3 10 3 Volume Sample = V o Volume Sample = V = V o /1,000 p n Constant Ct/Pt 100 10 -9 100 10 -6 n d Volume Sample = V o e.g. = 1mm 3 Volume Sample = V = V o /1,000,000,000 e.g. V = 1  m 3 Constant Ct/Pt

6. Singapore 2003 cuiti ñ o@rutgers Aside: PD/DEM Domain Computation Time/Physical Time PROPORTIONAL  pressure*number of particles/particle size Ct/Pt PROPORTIONAL p*n/d PRESSURE Case 1a: Constant Ct/Pt (for example, 1 CPU hour/ 1 second of physical time) Constant particle size If pressure is increased 1,000 the sample size is decreased 1,000 times (for example, from 1mm 3 to 100  m 3 ) Case 1b: Constant sample size (for example, 1mm 3 ) Constant physical time (for example 1 second) Constant particle size If pressure is increased 1,000 the Computational time is increased by 1,000 times (for example, 1 CPU hour to 1.5 CPU month) PARTICLE SIZE Case 2a: Constant Ct/Pt (for example, 1 CPU Hour/ 1 second of physical time) Constant pressure If particle size is reduced by 1,000 the sample size is decreased 10 9 times (for example, from 1mm 3 to 1  m 3 ) Case 2b: Constant sample size (for example, 1mm 3 ) Constant physical time (for example 1 second) Constant pressure If particle size is reduced by 1,000 the Computational time is increased 10 9 times (for example, 1 CPU Hour to 116,000 CPU years) High pressures and small particles conspired against PD/DEM simulations

7. Singapore 2003 cuiti ñ o@rutgers Supra-Particle Resolution Particle Dynamics Discrete Element Monte Carlo GRANULAR QUASI CONTINUUM Continuum INPUT From sup-particle resolution simulations (Multiscale)sup-particle From DIRECT Experimental Particle-Level Studies (Higashitani, Granick, …) OUTPUT Large regions / Long times Global Constrains (shapes, friction) Good for component DESIGN* * If a “good”constitutive relation is provided INPUT From: Parameter fitting with macroscopic experiments OUTPUT Limited to small regions and/or Short Times

8. Singapore 2003 cuiti ñ o@rutgers Granular Quasi-Continuum Allows for explicit account of the particle level response on the effective behavior of the powder Provides estimates of global fields such as stress, strain density Is numerically efficient, can also be improved by using stochastic integration Provides variable spatial resolution Is not well posed to handle large non- affine motion of particles Particle deformation is only considered in an approximate manner (as in PD/DEM) GOOD BAD

9. Singapore 2003 cuiti ñ o@rutgers Identifying Processes and Regimes Mixing Transport Granulation Characteristics: Large relative motion of particles Differential acceleration between particles Large number of distinct neighbors Low forces among particles Long times, relatively slow process Quasi steady state Discharge Die Filling Vibration Characteristics: Large relative motion of particles Differential acceleration between particles Large number of distinct neighbors Low forces among particles Short times Transient Early Consolidation Pre- compression Characteristics: Limited relative motion of particles Low particle acceleration Same neighbors Quasi-static Low forces among particles Small particle deformation (elastic) Consolidation Characteristics: No relative motion of particles Low acceleration Same neighbors Quasi-static Sizable forces among particles Small particle deformation (elastic + plastic) Compact Formation Characteristics: No relative motion of particles Low acceleration Same neighbors Quasi-static Large forces among particles Large particle deformation

10. Singapore 2003 cuiti ñ o@rutgers Identifying Numerical Tools (which can use direct input from finer scale) Mixing Transport Granulation PD/DEM/MC Discharge Die Filling Vibration PD/DEM/MC Ballistic Deposition Early Consolidation Pre- compression PD/DEM/MC Consolidation GCC Compact Formation GQC OUR SCOPE Numerical tools appropriate for process

11. Singapore 2003 cuiti ñ o@rutgers Numerical Strategies for assembly of particles Continuum Plasticity models for frictional materials such as Drucker-Prager Discrete Molecular Dynamics Discrete Element Method Continuum with Microstructure

12. Singapore 2003 cuiti ñ o@rutgers Continuum with Microstructure Cyclic

13. Singapore 2003 cuiti ñ o@rutgers Numerical Strategies for assembly of particles Continuum Plasticity models for frictional materials such as Drucker-Prager Continuum with Microstructure Discrete Molecular Dynamics Discrete Element Method Granular Quasi Continuum PARTICLES POWDERS (discrete) (continuum)

14. Singapore 2003 cuiti ñ o@rutgers Constrain kinematics of the particles by overimposing a displacement field described by a set of the displacements in a set of points (nodes) and a corresponding set of interpolation functions (a FEM mesh) A quasi-continuum approach FEM MeshSet of Particles Combined System Granular Quasi-Continuum

15. Singapore 2003 cuiti ñ o@rutgers Nomenclature umum rmrm rnrn r mn RmRm RnRn R mn Particle m Particle n

16. Singapore 2003 cuiti ñ o@rutgers Governing Equations PVW Euler Equation Local Equilibrium

17. Singapore 2003 cuiti ñ o@rutgers Constrained Kinematics Displacement Field is constrained by the selection of the mesh Displacement of Node a Displacement of particle m Relative Displacement of particle m with respect to particle n

18. Singapore 2003 cuiti ñ o@rutgers Discrete Version

19. Singapore 2003 cuiti ñ o@rutgers Local/Non-local Formulation  u u*(  m ) u(  n ) u(  m ) u*(  n ) nn mm d

20. Singapore 2003 cuiti ñ o@rutgers Additional Freedom/ Mesh Adaption Indicator Additional relaxation by adding internal nodes Still within the framework of FEM

21. Singapore 2003 cuiti ñ o@rutgers Load Paths

22. Singapore 2003 cuiti ñ o@rutgers Mueth, Jaeger, Nagel 2000 Comparison with Experiment Experiment Simulation

23. Singapore 2003 cuiti ñ o@rutgers Simulations Pre-rearrangement

24. Singapore 2003 cuiti ñ o@rutgers Simulations Post-rearrangement

25. Singapore 2003 cuiti ñ o@rutgers Macroscopic Behavior

26. Singapore 2003 cuiti ñ o@rutgers Macroscopic Behavior RED – Group I BLUE – Group II

27. Singapore 2003 cuiti ñ o@rutgers Macroscopic Behavior RED – Group I BLUE – Group II

28. Singapore 2003 cuiti ñ o@rutgers Comparison with Experiment Experiment Simulation

29. Singapore 2003 cuiti ñ o@rutgers Adjusting Force Fields Elastic Properties Plastic Properties Fracture Properties BULK PROPERTIES SURFACE PROPERTIES Lactose 1.5% PEG 0.5% Glycerin (Courtesy of P&G P. Mort and M. Roddy) Experimental Numerical

30. Singapore 2003 cuiti ñ o@rutgers Numerical Studies Unit Cell Approach

31. Singapore 2003 cuiti ñ o@rutgers Numerical Studies Yield Stress Effects

32. Singapore 2003 cuiti ñ o@rutgers Numerical Studies Young Modulus Effects

33. Singapore 2003 cuiti ñ o@rutgers Numerical Studies Hardening Effects

34. Singapore 2003 cuiti ñ o@rutgers Materials Solid Materials Granulated Materials HPDE (high density polyethylene) PEG 8000 (Polyethylene glycol) Lactose 1.5% PEG (binder) 0.5% Glycerin (lubricant)

35. Singapore 2003 cuiti ñ o@rutgers Experimental Studies PEG 8000 PEG 3350 PEG 1450 Glycerin 0% 0.5% 1.0% 1.5%

36. Singapore 2003 cuiti ñ o@rutgers Experimental Studies Glycerin (Yield Stress)

37. Singapore 2003 cuiti ñ o@rutgers Experimental Studies PEG ( Elastic Modulus + Hardening )

38. Singapore 2003 cuiti ñ o@rutgers Future Emerging trend in the material system design: using computer simulations to probe the space of best candidates for manufacturing. We have developed a basic framework capable of simulating the consolidation process, including: –die filling (with/without cohesion) –particle rearrangement in both 2D and 3D –powder densification in both 2D and 3D, with materials with a wide range of elastic and plastic properties powders with a wide range of size distributions particle different shapes (by agglomeration of spherical particles) wall friction complex shapes dynamic effects FROM TO Specific applications More validation Pre die filling Post ejection behavior, in particular bonding using the GQC This approach will provide estimates of the local and global strength of the material by following the temperature and pressure dependent evolution of the bonding process