Heat equation (FE)

In this tutorial, the heat equation (first steady and then unsteady) is solved using finite-elements.

Theory

This example shows how to solve the heat equation with eventually variable physical properties in steady and unsteady formulations:

\[ \rho C_p \partial_t u - \nabla . ( \lambda u) = f\]

We shall assume that $f, \, \rho, \, C_p, \, \lambda \, \in L^2(\Omega)$. The weak form of the problem is given by: find $ u \in \tilde{H}^1_0(\Omega)$ (there will be at least one Dirichlet boundary condition) such that:

\[ \forall v \in \tilde{H}^1_0(\Omega), \, \, \, \underbrace{\int_\Omega \partial_t u . v dx}_{m(\partial_t u,v)} + \underbrace{\int_\Omega \nabla u . \nabla v dx}_{a(u,v)} = \underbrace{\int_\Omega f v dx}_{l(v)}\]

To numerically solve this problem we seek an approximate solution using Lagrange $P^1$ or $P^2$ elements. Here we assume that the domain can be split into two domains having different material properties.

Steady case

As usual, start by importing the necessary packages.

using Bcube
using LinearAlgebra
using WriteVTK

First we define some physical and numerical constants

const htc = 100.0 # Heat transfer coefficient (bnd cdt)
const Tr = 268.0 # Recovery temperature (bnd cdt)
const phi = 100.0
const q = 1500.0
const λ = 100.0
const η = λ
const ρCp = 100.0 * 200.0
const degree = 2
const outputpath = joinpath(@__DIR__, "..", "..", "myout", "heat_equation/")

Read 2D mesh

mesh_path = joinpath(@__DIR__, "..", "..", "input", "mesh", "domainSquare_tri.msh")
mesh = read_msh(mesh_path)

Build function space and associated Trial and Test FE spaces. We impose a Dirichlet condition with a temperature of 260K on boundary "West"

fs = FunctionSpace(:Lagrange, degree)
U = TrialFESpace(fs, mesh, Dict("West" => 260.0))
V = TestFESpace(U)

Define measures for cell integration

dΩ = Measure(CellDomain(mesh), 2 * degree + 1)

Define bilinear and linear forms

a(u, v) = ∫(η * ∇(u) ⋅ ∇(v))dΩ
l(v) = ∫(q * v)dΩ

Create an affine FE system and solve it using the AffineFESystem structure. The package LinearSolve is used behind the scenes, so different solver may be used to invert the system (ex: solve(...; alg = IterativeSolversJL_GMRES())) The result is a FEFunction (ϕ). We can interpolate it on mesh centers : the result is named Tcn.

sys = AffineFESystem(a, l, U, V)
ϕ = Bcube.solve(sys)
Tcn = var_on_centers(ϕ, mesh)

Compute analytical solution for comparison. Apply the analytical solution on mesh centers

T_analytical = x -> 260.0 + (q / λ) * x[1] * (1.0 - 0.5 * x[1])
Tca = map(T_analytical, get_cell_centers(mesh))

Write both the obtained FE solution and the analytical solution to a vtk file.

mkpath(outputpath)
dict_vars =
    Dict("Temperature" => (Tcn, VTKCellData()), "Temperature_a" => (Tca, VTKCellData()))
write_vtk(outputpath * "result_steady_heat_equation", 0, 0.0, mesh, dict_vars)

Compute and display the error

@show norm(Tcn .- Tca, Inf) / norm(Tca, Inf)

Unsteady case

The code for the unsteady case if of course very similar to the steady case, at least for the beginning. Start by defining two additional parameters:

totalTime = 100.0
Δt = 0.1

Read a slightly different mesh

mesh_path = joinpath(@__DIR__, "..", "..", "input", "mesh", "domainSquare_tri_2.msh")
mesh = read_msh(mesh_path)

The rest is similar to the steady case

fs = FunctionSpace(:Lagrange, degree)
U = TrialFESpace(fs, mesh, Dict("West" => 260.0))
V = TestFESpace(U)
dΩ = Measure(CellDomain(mesh), 2 * degree + 1)

Compute matrices associated to bilinear and linear forms, and assemble

a(u, v) = ∫(η * ∇(u) ⋅ ∇(v))dΩ
m(u, v) = ∫(ρCp * u ⋅ v)dΩ
l(v) = ∫(q * v)dΩ

A = assemble_bilinear(a, U, V)
M = assemble_bilinear(m, U, V)
L = assemble_linear(l, V)

Compute a vector of dofs whose values are zeros everywhere except on dofs lying on a Dirichlet boundary, where they take the Dirichlet value

Ud = assemble_dirichlet_vector(U, V, mesh)

Apply lift

L = L - A * Ud

Apply homogeneous dirichlet condition

apply_homogeneous_dirichlet_to_vector!(L, U, V, mesh)
apply_dirichlet_to_matrix!((A, M), U, V, mesh)

Form time iteration matrix (note that this is bad for performance since up to now, M and A are sparse matrices)

Miter = factorize(M + Δt * A)

Init the solution with a constant temperature of 260K

ϕ = FEFunction(U, 260.0)

Write initial solution to a file

mkpath(outputpath)
dict_vars = Dict("Temperature" => (var_on_centers(ϕ, mesh), VTKCellData()))
write_vtk(outputpath * "result_unsteady_heat_equation", 0, 0.0, mesh, dict_vars)

Time loop

itime = 0
t = 0.0
while t <= totalTime
    global t, itime
    t += Δt
    itime = itime + 1
    @show t, itime

    # Compute rhs
    rhs = Δt * L + M * (get_dof_values(ϕ) .- Ud)

    # Invert system and apply inverse shift
    set_dof_values!(ϕ, Miter \ rhs .+ Ud)

    # Write solution (every 10 iterations)
    if itime % 10 == 0
        dict_vars = Dict("Temperature" => (var_on_centers(ϕ, mesh), VTKCellData()))
        write_vtk(
            outputpath * "result_unsteady_heat_equation",
            itime,
            t,
            mesh,
            dict_vars;
            append = true,
        )
    end
end

This page was generated using Literate.jl.