Discretization

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 9

Discretization

Guillermo Reyes Carmenaty


October 19th, 2018
A continuous world for a discrete processor

• Why do we need discretization?


– The world we simulate is continuous in nature: one cannot distinguish
different elements within the simulated objects unless we go sub-
atomic.
– The machines we use to simulate understand only discrete values,
depending on how many bits of information the processor can
manage.
– A way for the processor to understand the continuous world must be
found: discretization of the continuity into finite elements.

2
The nature of the Finite Element Method

• What is the Finite Element Method?


– It is a numerical methodology that enables the solution of boundary
condition systems by breaking such systems into smaller, more
manageable, discrete finite elements.
– In our field of application (mechanical analysis), the Finite Element
Method is used to obtain the deformation that occurs in a solid as a
consequence of how some parts of the solid have limited DoFs and
the external forces that are applied to them.
– From that information and some data about material properties,
stress, strain and other interesting values can be obtained.

3
Spatial discretization

• How we discretize geometry


– Given the nature of our method and what we use it for, we must break the studied
geometry into discrete, manageable elements. The FEM ultimately calculates the
displacement of some of the infinitesimal points of the solid, which are interrelated.
How many of this points, which we call nodes, and how we consider their relation
are key to how reliable the results are in relation to what would actually happen in
reality.
– We can consider this relations as unidimensional, bi-dimensional or tri-
dimensional, depending on the complexity and nature of what we are trying to
simulate. The more dimensions we try to incorporate in the calculation, the more
calculation time it will take, and even considering a set number of dimensions, the
amount of relations between nodes can greatly increase or decrease accuracy or
calculation time needed.
– The amount of nodes determines how discrete our system is, and the relations
between them impact calculation time, stability and reliability..
– We must bear in mind the scale of what we are studying: if simulating a building, a
bolt should have few to no elements, but if we are studying how that bolt behaves
in the building, then it should be allocated a generous amount of elements.

4
Meshing particularities

• Adaptive meshing
– We not always need the full detail we anticipated when devising our discretization, and
sometimes finer elements are needed in order to fully appreciate what is happening in a certain
region. Adaptive Meshing Refinement algorithms allow for refinement of the mesh while solving
the simulation without user interaction.
• Multidiscretization
– Not all of the mesh has to be composed of the same type of elements. Some elements are more
suitable than others for certain regions of the geometry, and sometimes not all of the geometry
needs great complexity, so it is sound to use a variety of elements to only use the resources that
are needed.
• H-method and P-method
– When refining mesh there are two possible strategies: one is to use more elements, which
greatly increases the computational cost of running the simulation (H-method), the other is to use
higher polynomial degree elements in order to calculate what happens between nodes more
faithfully through interpolation.

5
Temporal discretization

• How we break up time


– As we break space, so we shall break time. We know how we start
and when we want to end, we want to know how we end and how did
we get there. In order to do that we take small temporal steps from
when we begin until we reach when we end. How small this steps are
is a matter of temporal discretization.
– This time step is function of the geometry of the solid and the stiffness
and density of the material it is made of. If the time step is too large,
the numerical solution may diverge significantly from what would
actually happen, but the smaller it is, the more calculation time we will
need.
• Explicit vs. Implicit
– In explicit calculations, the next step is calculated, in implicit
calculations, the next step is calculated and compared to an
estimation of how it should be. Both ways have drawbacks and
advantages and impact the size of the time step.

6
Conclusions

• We need to break up space


– How we break up the geometry of the piece and how this smaller
parts interact with each other makes the difference between a reliable
but timely calculation and an unreliable or exremely slow one.
• We need to break up time
– How we break up time and how we advance through it greatly
impacts both reliability and actual calculation time needed to
complete the simulation.
• Be wary of interpolation
– We can make an estimate of how the solid would have behaved in a
certain point at a certain time, both of which we hadn’t calculated
initially by interpolating between the two closer points in space/time,
but this result is bound to present greater error.

7
Bibliography

• S. C. Chapra, Applied numerical methods with MATLAB for


engineers and scientists, 3rd ed. New York: McGraw-Hill, 2012.
• K.-J. Bathe, “Finite Element Procedures for Solids and Structures­
Linear Analysis.” MIT Center for Advanced Engineering Study.
• H. Abderrachid, “What is difference between implicit and explicit
FEM in non-linear quasi-static problems? (Answer by
Abderrachid Hamrani),” ResearchGate. Consulted on
17/10/2018.

You might also like