Simulating Multi-Rigid-Body Dynamics with Contact and Friction

1. Simulating Multi-Rigid-Body Dynamics with Contact and Friction Mihai Anitescu SIAM OPTIMIZATION MEETING May 10, 1999

2. Rigid Multi-Body Dynamics Simulation Applications • Industrial simulation and design for aircraft, ships, heavy vehicles (product design -- production, 5 years). • Flexible Manufacturing Systems design and redesign (commissioning+tuning, 3 years). • Appropriate simulation could reduce lead times by up to 50%. • Simulating virtual environments (+real time = interactive).

3. Insert Picture Virtual Environment Application Example • Setting up the real installation for training is very expensive. • The virtual environment can be easily modified for different scenarios, but simulating it in real-time is essential. • Simplified model: rigid bodies with joints and contacts.

4. Initial model requirements • Acceleration based Newton laws. • Rigid bodies. • No inter penetration, nonnegative contact impulses and passive contact • Joint constraints • Coulomb friction or reasonable approximations. • Impact resolution (Poisson hypothesis).

5. Classical Constrained Dynamics

6. Inconsistency example • State space representation • Constraint , initially

7. Frictional Inconsistency • What went wrong? Newton’s laws rather than Coulomb’s model. • Configuration inconsistent, not only singular. • P. Painleve: Governing equations of motion do not have a solution in general( Comptes Rendus Acad. Sci. Paris, 1895). • Approaches for the continuous case. • Solutions in a distributional sense. • Differential inclusions instead of differential equations (Moreau, 1986; Monteiro-Marques, 1993).

8. Avoiding Frictional Inconsistencyin time-stepping schemes • Time-stepping: If a continuous model exists, how should the associated numerical integration scheme look like? • Friction is estimated, and the frictionless problem is solved (Lotstedt,1982). Problems: starting configuration and the impact case. • The inconsistent configuration is treated as a collision(Baraff,1993). Problems: arbitrary collisions and the impact case. • Newton's law with impulses and velocities (Stewart & Trinkle 1995; Anitescu & Potra, 1996).

9. Revised model requirements • Velocity based Newton laws, by combining time stepping with impulse resolution. • Rigid bodies. • No inter penetration, nonnegative contact impulses and passive contact • Joint constraints • Coulomb friction or reasonable approximations. • Impact resolution (Poisson hypothesis).

10. Insert Model Slides Framework and notations used • The rigid multi-body system is described by its generalized coordinates, q, and generalized velocities, v. • The contacts (joints) are described by inequalities (equalities). • The ``tangent plane'' for a contact constraint is defined by the cone generators • M: mass matrix, symmetric positive definite; h: the time step k: external force.

11. Contact configuration described by the (generalized) distance function , which allows for some interpenetration. Feasible set . Contact impulses are compressive The normal velocity for non inter penetration is velocity-based (as opposed to Stewart & Trinkle, where it is position-based): Complementarity guarantees energetic passivity of contact: Contact Model

12. is the normal impulse and is the tangential impulse is the total impulse in Newton Euler world coordinates . Classical Coulomb model requires the 3D contact impulse to lie in a circular cone. If relative motion exists, the frictional impulse has to be opposite to the relative velocity (Maximum Dissipation) Contact Description

13. Polygonal approximation of Coulomb cone

14. The Integration Step

15. Matrix Form of The Integration Step

16. Linear Complementarity Problems (LCPs) • Linear and Quadratic Programming are LCPs, via primal-dual formulations. • If , a solution exists, (by a Brouwer fixed point argument), and can be computed by Lemke’s algorithm (pivotal, similar to simplex). • M has to be copositive

17. Interior Point Algorithms (IP) • It takes corrected Newton steps to compute a solution if M is positive semi definite. The central path is well defined. • In the convex case IP algorithms have polynomial complexity, but other extensions are unclear.

18. Theorem • Consider the mixed LCP above. If M is a symmetric positive definite matrix, N a copositive matrix and b a nonnegative vector, then the mixed LCP has a solution. Lemke’s algorithm (with anti cycling strategy) will find a solution of the LCP obtained by eliminating x,y.

19. Properties of the formulation • The model is solvable for any configuration. • Discrete Maximum Dissipation Principle (MDP): The frictional impulse maximizes the dissipation over all feasible contact impulses, given • If the mass matrix is constant, the kinetic energy at the new step cannot exceed the the kinetic energy of the system with no constraint enforced (contacts are passive).

20. Energy/ stability properties • Assumption: M constant. • If the external force is of the form • c depends only on M and d. h has to be sufficiently small but depending only on c, T.

21. Constraint stability • There are constants C,H depending on the problem data, but not on the time-step h, such that • for any h < H (q(t) is obtained by linear interpolation).

22. Difficult Friction Configuration • For any friction coefficient we obtain multiple solutions. • Does this result in undesirable properties of the LCP matrix?

23. Column sufficiency?

24. Row sufficiency?

25. Properties of the LCP matrix. • The matrix Q is neither row, nor column sufficient. • As a result, it cannot be in the class, which makes it difficult to use polynomial-time interior-point algorithms. • Lemke’s algorithm appears to be the only one which is guaranteed to provide a solution to this model. • However, further analysis may reveal good matrix properties for pointed friction cones. • Since model needs to solve a succession of these LCPs anyway, other approaches may provide (computationally) better matrices.

26. Differential Inclusions • v has bounded variation • K(t) is closed and convex set valued

27. Theorem • For a pointed friction cone, there exists • The quantities are obtained by linear interpolation. (D. Stewart)

28. Theorem(II) • For one contact, maximal dissipation holds

29. Accomplishments • The time-stepping model to has a solution regardless of the configuration and dimension of the problem. Impulsive solutions are accommodated. • Static and dynamic friction are treated in an unitary manner. • The constraints are guaranteed to be satisfied within any given tolerance. • The numerical scheme is dissipative and, as a result the velocities are uniformly bounded. • A differential inclusion model with energy dissipation satisfied in the limit.

30. 2-dimensional examples • Long bar with multiple contacts: • low friction (mu=0.05, e=0.6) • high friction.(mu=0.2, e=0.6) • 2 blocks on table • low friction (mu=0.01, e=0.2) • high friction (mu=0.1, e=0.2) • Multiple collisions with high restitution, high friction (mu=0.3, e=0.9). • Three blocks on table, low friction (mu=0.05, e=0.6). • Several blocks, (mu=0.2, e=1.0)

31. Open Problems • Can a consistent differential inclusion approach be formulated for general configurations? • Under which conditions does the maximum dissipation principle apply in the multiple contact case? • Can other properties of the time-stepping scheme be used in the continuous time context (such as the bound on the square of the variation of the velocities)? • Is the solution set of the friction LCP convex? (it is not true for arbitrary copositive matrices), • Can the LCP be recast as a convex optimization problem?

32. Future Research • Treat the Coulomb cone directly rather than using a polygonal approximation (partly solved APS98 ). • Design stable numerical algorithms to handle stiffness (implicit or linearly implicit, work in progress). • Nonlinear formulations (NCP). • Investigate Interior-Point Algorithms for solving the friction LCP or NCP. • Are higher-order methods relevant? • Formulate an efficient piecewise DAE strategy.