Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and...

14
Final Presentation A. Davey, N. Moniz, T. Spilhaus UMass Dartmouth August 16, 2012 A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 1 / 14

Transcript of Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and...

Page 1: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Final Presentation

A. Davey, N. Moniz, T. Spilhaus

UMass Dartmouth

August 16, 2012

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 1 / 14

Page 2: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Presentation Overview

I Project overview and parameters.

I Motivation for using a particular method to solve the problem at hand.

I Methodology and description of mathematical procedures.

I Numerical results.

I Future goals and possible applications.

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 2 / 14

Page 3: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Problem Overview and Description

I Using finite element method we will attempt to model the steady-state heatprofile across a flat plate with different boundary conditions and right handside functions.

I We will model Poisson’s equation −∇ · (κ(x , y)∇u(x , y)) = f (x , y) on a unitsquare [−1, 1]2.

I Boundary conditions will begin with homogeneous Dirichlet set equal to zeroand then vary as the project progesses.

I κ(x , y) = 16 + ε1x + ε2y

I In the beginning, ε1 and ε2 will be drawn at random from a normal Gaussiandistribution with σ = 1

2 .

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 3 / 14

Page 4: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Advantages and Disadvantages of FEM

I Used in a wide variety of areas tosolve and model military,commercial, and industrialproblems.

I Elements can be individuallycustomized to model objects withcomplex geometries in multipledimensions.

I Computationally intensive if highgrid refinement is used and sourcecode can become quite large.

Figure 1: Finite element simulation of jetturbine engine.

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 4 / 14

Page 5: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Methodology: Outline

I Derive the weak formulation of our partial differential equation.

I Discretize over space (Ω) creating a two-dimensional mesh of arbitraryrefinement.

I Select shape and weight functions.

I Assemble the linear system.

I Apply boundary conditions (either homogeneous Dirichlet or mixed Dirichletand Neumann) and solve system.

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 5 / 14

Page 6: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Methodology: Deriving the Weak Formulation

−∇ · (κ(x , y)∇u(x , y)) = f (x , y)

We then multiply both sides by a test function v .

−v∇ · (κ(x , y)∇u(x , y)) = vf (x , y)

We now integrate over the domain Ω.

−∫

Ωv∇ · (κ(x , y)∇u(x , y)) =

∫Ωvf (x , y)

Then integrating by parts will yield the following.∫Ω∇v · κ∇u(x , y)−

∫Γvn · ∇u =

∫Ωvf (x , y)

Lastly, we apply homogeneous Dirichlet boundary conditions, u(x , y) = 0, along theboundary Γ.∫

Ω∇v · κ∇u(x , y) =

∫Ωvf (x , y)

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 6 / 14

Page 7: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Methodology: Basis Functions and Generating the Mesh

I we find an approximation functionof u to satisfy our equation for allfunctions of v.

uh =∑

j cjφj

I we can now define our test functionas shown below.

v = ciφi , i = 0, · · · ,N

I we also need a mesh onto which wewill map our shape functions.

Figure 2: Example of a generated mesh

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 7 / 14

Page 8: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Methodology: Basis Functions and Generating the Meshcontinued

I then we define the weak form of the discrete problem and substitute in our uh∫Ω∇v · κ∇

∑j cjφj =

∫Ωvf (x , y)

I we then write this in terms of a linear system

Ac = F

I A and F are then replaced with weighted sums over a set of points on eachcell in there domain and we set these as the approximations:

Akij =

∑q∇φi (xkq ) · ∇φj(xkq )wk

q

F ki =

∑q φi (x

kq )f (xkq )wk

q

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 8 / 14

Page 9: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Numerical Results

I several different conditions applied

I average solutions

I average kappa compared to average solution

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 9 / 14

Page 10: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Solutions of each Applied Condition

I

I

I

I

Figure 3: Top-Left: homogeneous Dirichlet conditions, Bottom-Left: Mixed Boundaries,Top-Right: added Gaussian on RHS, Bottom-Right: Sigma changed from 1/2 to 5

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 10 / 14

Page 11: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Average Solutions of each Applied Condition

I

I

I

I

Figure 4: Top-Left: homogeneous Dirichlet conditions, Bottom-Left: Mixed Boundaries,Top-Right: added Gaussian on RHS, Bottom-Right: Sigma changed from 1/2 to 5

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 11 / 14

Page 12: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Comparison overlay of average solutions with sigma 5 andsigma 1/2

Figure 5: Overlay of average of 100 solutions with σ = 12

in green and σ = 5 in red andblue.

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 12 / 14

Page 13: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Convergence of average solution to the average kappasolution

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 13 / 14

Page 14: Final Presentation - compmath.files.wordpress.com · Presentation Overview I Project overview and parameters. I Motivation for using a particular method to solve the problem at hand.

Possible Future Goals

I Modeling more realistic physical systems, such as changing our Kappa to bethe actual heat conductivity of a known material.

I Adding in Time Dependance to the problem being modeled and seeing how itevolves.

I Using FEM to approximate higher dimensional systems

I Modeling Higher Order Methods for higher accuracy results

A. Davey, N. Moniz, T. Spilhaus (UMass Dartmouth) Final Presentation August 16, 2012 14 / 14