Professional Documents
Culture Documents
Lecture 5 PDF
Lecture 5 PDF
Explicit
Implicit Analysis u
u Implicit
q The dependent variables are defined by coupled sets
of equations, and either a matrix or iterative
technique is needed to obtain the solution
q Unknown appears on both sides of the equation
• x(t+h) = x(t) + h f(x(t+h))
u Internal and external forces are summed at each node point u A global stiffness matrix is computed, inverted and applied to
and a nodal acceleration is computed by dividing by nodal the nodal out of balance force to obtain a displacement
mass increment
..
[M]{x}n = [Fexternal ]n − [Finternal ]n .. ..
[M ]{ x} n+1 + [ K ]{ x} n+1 = [Fexternal] n+1 − [ Finternal]n − [ M ]{ x} n
u Solution is advanced by numerical integration of the above u Large numerical effort required to form, store and factorize
computed acceleration in time the stiffness matrix
u Courant condition limits largest stable time step u Typically involve a relatively small number of expensive time
u Typically requires many relatively inexpensive time steps steps
u Well suited for dynamic simulations such as impact and crash u Well suited for static and quasi-static simulations
(short duration) q Quasi -static analysis: “time” represents a monotonically
increasing parameter which characterizes the evolution of the
loading
1
Activating Implicit Analysis Implicit Keywords
u Three types of analysis can be performed u *CONTROL_IMPLICIT_GENERAL (required)
q Fully explicit (default) q Activates implicit mode
q Fully implicit q Defines implicit time step size
u Implicit governing equations contain two problems to solve u A linear system of equations of the form K.Du=R must
.. ..
be solved within each equilibrium iteration
[ M ]{x }n+ 1 + [ K ]{x}n +1 = [ Fexternal] n +1 − [ Finternal] n − [ M ]{x}n
u Stiffness matrix K is inverted and applied to the
residual load R, yielding a displacement increment Du
u Nonlinear Problem u Storing and solving this linear system represent a
q Find displacements x which satisfy equilibrium Fext =Fint large portion of the memory and CPU costs of an
q Both K, Fext and Fint can be nonlinear functions of x implicit analysis
q Iterative search employed using Newton-based methods
u Several different linear equations solvers are
u Linear Problem available, including direct (Gaussian elimination) and
q Solve system of linear algebraic equations, [k]{x}=[f] iterative (conjugate gradient, Lanczos) methods
q Must solve during every nonlinear iteration u Very important to allow enough memory for the
q Great CPU and memory cost makes this problem important stiffness matrix factorization
2
Non-linear Equation Solver Element Formulations
u Several different non-linear equations solvers are u Selected using *SECTION keywords
available. All of them are iterative
u Default elements often use single point integration
u When the norms of displacement and energy are
with hourglass control
reduced below user prescribed tolerances (default 1.0e-
3 and 1.0e-2 respectively), equilibrium is reached within q Good for explicit, where element costs can dominate
sufficient accuracy, the iteration process is said to have solution
converged, and the solution proceeds to the next time q Bad for implicit, since hourglass modes cause
step convergence trouble
u For implicit analysis it is generally more effective to
use more expensive element formulations which
are less susceptible to hourglass instability
Good
Good
Bad Bad
3
Hourglass Control Element Formulations for Implicit Analysis
u Shell Elements
q S/R Hughes Liu shell element #6
• 2x2 selective -reduced integration
• 6 DOF per node (dx, dy, dz, rx, ry, rz)
• Most expensive
q Fully integrated shell element #16
• 2x2 integration with enhanced strain formulation
• 6 DOF per node (dx, dy, dz, rx, ry, rz)
• Least expensive of 2x2 elements
Bad
Bad
u Solid Elements
q Hughes Liu brick element #2
• 3 DOF per node (dx, dy, dz)
Good Good • 2x2x2 selective reduced integration
• Most expensive
4
Loads Automatic Time Step Control
Model Setup
Honeycomb Block
Implicit Analysis u
u Implicit Analysis
q Crosshead speed –
10 mm/sec
q Output – Rwforc
and nodout
5
Dyna Input Dyna Output
u X-axis,
q R.Vol = Current vol/Initial vol
q R.Vol = L1L2L3’/L1L2L3
q R.Vol = (L3-d)/L3
q R.Vol = (150-d)/150
u Y-axis,
q Stress = Force/Area
q Stress = F/(122.52)
6
Comparison