Nonlinear finite elements/Nonlinear axial bar weak form

< Nonlinear finite elements

Symmetric weak form for the axially loaded bar

Let us start with the weak form and derive a symmetric stiffness matrix. We have,


    {
    A~\cfrac{d}{dx}\left[E(u)\cfrac{du}{dx}\right] + ax = 0 ~.
    }

Once again, we multiply with a weighting function and integrate over an element to get


    \int_{\Omega} A~\cfrac{d}{dx}\left[E(u)\cfrac{du}{dx}\right]w~dx + 
    \int_{\Omega} axw = 0 ~.

Integrating the first term by parts, we get


    \left.AE(u)\cfrac{du}{dx}w\right|_{\Gamma_t} - 
    \int_{\Omega} AE(u)\cfrac{du}{dx}\cfrac{dw}{dx}~dx + 
    \int_{\Omega} axw = 0 ~.

Taking the internal force terms to the left and the external force terms to the right, we get


    \int_{\Omega} AE(u)\cfrac{du}{dx}\cfrac{dw}{dx}~dx  =  
    \int_{\Omega} axw  +
    \left.AE(u)\cfrac{du}{dx}w\right|_{\Gamma_t} ~.

Note that, at x = L,


     f = A\sigma = AE(u)\cfrac{du}{dx} = R ~.

Also, at x = 0, we have w = 0. The external surface force is zero at all other points. Therefore,


    \int_{\Omega} AE(u)\cfrac{du}{dx}\cfrac{dw}{dx}~dx  =  
    \int_{\Omega} axw  + Rw|_{x=L} ~.

The finite element approximation

Let, the trial and weighting functions be


    u = \sum_j u_j N_j~;~\qquad w = N_i ~.

Then, we get


    \int_{\Omega} AE(u) \left(\sum_j u_j \cfrac{dN_j}{dx}\right)
                 \cfrac{dN_i}{dx}~dx  =  
    \int_{\Omega} axN_i  + RN_i|_{x=L} ~.

Take the constants outside the integral sign as usual, and we can write


    A\sum_j\left(\int_{\Omega} E(u) \cfrac{dN_i}{dx}\cfrac{dN_j}{dx}~dx\right)
            u_j = \int_{\Omega} axN_i  + RN_i|_{x=L} ~.

The above can be written in the usual form as


    K_{ij} u_j = f_i \,

where


    {
      \begin{align}
        K_{ij} & := A\int_{\Omega} E(u) \cfrac{dN_i}{dx}\cfrac{dN_j}{dx}~dx \\
        f_i  & := \int_{\Omega} axN_i  + RN_i|_{x=L} ~.
      \end{align}
    }

This stiffness matrix is symmetric. However, we are still left with the problem of dealing with the E(u) part.

Dealing with E(u)

The expression of E(u) is


    E(u) := E_0\left(1-\cfrac{du}{dx}\right)~.

In terms of the trial function, we have


    E(u) = E_0\left(1-\sum_k u_k\cfrac{dN_k}{dx}\right)~.

I have chosen to call the index k in order to distinguish it from the index names that we have already used, that is i and j.

Plugging the expansion for E(u) into the expression for K_{ij} we get


    K_{ij} = AE_0\int_{\Omega}\left(1-\sum_k u_k\cfrac{dN_k}{dx}\right) 
                              \cfrac{dN_i}{dx}\cfrac{dN_j}{dx}~dx ~.

Notice that the stiffness matrix continues to remain symmetric when we do this because the sum over k includes all the relevant nodes.

The stiffness matrix is now a function of the unknown displacements u_i and the system of equations can be written as


     \mathbf{K}(\mathbf{u}) \mathbf{u} = \mathbf{f}~.

This is a nonlinear system of equations and cannot be solved directly (as was done for linear systems).

Making things a bit more concrete

Let us get a bit more explicit about what all that leads to.

Assume that the bar is divided into two linear elements. Therefore, there are three global nodes: 1, 2, and 3. Each element has two nodes which have local numbers 1 and 2.

The shape functions within each element are


    N_1 = \cfrac{x_2 - x}{h}, \qquad N_2 = \cfrac{x - x_1}{h}

where x_1 is the coordinate of node 1 of the element, and x_2 is the coordinate of node 2 of the element. The length of the element is h.

The derivatives of the shape functions are


    \cfrac{dN_1}{dx} = -\cfrac{1}{h}, \qquad \cfrac{dN_2}{dx} = \cfrac{1}{h}~.

The element stiffness matrix terms can be written as


    K_{ij} = AE_0\int_{x_1}^{x_2}
      \left(1-u_1\cfrac{dN_1}{dx}-u_2\cfrac{dN_2}{dx}\right) 
                              \cfrac{dN_i}{dx}\cfrac{dN_j}{dx}~dx

where u_1 is the displacement of local node 1, and u_2 is the displacement of local node 2.

Plugging in the values of the derivatives of the shape functions, we get

\begin{align}
    K_{11} & = \cfrac{AE_0}{h^2}
             \int_{x_1}^{x_2}\left(1+\cfrac{u_1}{h}-\cfrac{u_2}{h}\right)~dx \\
    K_{12} & = -\cfrac{AE_0}{h^2}
             \int_{x_1}^{x_2}\left(1+\cfrac{u_1}{h}-\cfrac{u_2}{h}\right)~dx \\
    K_{21} & = -\cfrac{AE_0}{h^2}
             \int_{x_1}^{x_2}\left(1+\cfrac{u_1}{h}-\cfrac{u_2}{h}\right)~dx \\
    K_{22} & = \cfrac{AE_0}{h^2}
             \int_{x_1}^{x_2}\left(1+\cfrac{u_1}{h}-\cfrac{u_2}{h}\right)~dx 
  \end{align}

The terms inside the integrals are the same for all the coefficients of the stiffness matrix (this is not true in general), and we have


    \int_{x_1}^{x_2}\left(1+\cfrac{u_1}{h}-\cfrac{u_2}{h}\right)~dx = 
       h + u_1 - u_2 ~.

Therefore,

\begin{align}
    K_{11} & = \cfrac{AE_0}{h^2}(h + u_1 - u_2) \\
    K_{12} & = -\cfrac{AE_0}{h^2}(h + u_1 - u_2) \\
    K_{21} & = -\cfrac{AE_0}{h^2}(h + u_1 - u_2) \\
    K_{22} & = \cfrac{AE_0}{h^2}(h + u_1 - u_2) 
  \end{align}

In matrix form, the element stiffness matrix is


    \mathbf{K} = \cfrac{AE_0}{h^2}(h + u_1 - u_2)\begin{bmatrix}1 & -1 \\-1 & 1
           \end{bmatrix} ~.

For simplicity, let us set the body force term a to zero. Let us also assume that both elements have equal lengths. Then the assembled global system of equations (for the two element mesh) is


    \cfrac{AE_0}{h^2}
    \begin{bmatrix}
      h+u_1-u_2 & -(h+u_1-u_2) & 0 \\
      -(h+u_1-u_2) & (h+u_1-u_2)+(h+u_2-u_3) & -(h+u_2-u_3) \\
      0 & -(h+u_2-u_3) & (h+u_2-u_3)
    \end{bmatrix}
    \begin{bmatrix} u_1 \\ u_2 \\ u_3 \end{bmatrix} = 
    \begin{bmatrix} f_1 \\ 0 \\ R \end{bmatrix} ~.

After simplification, we have


    \cfrac{AE_0}{h^2}
    \begin{bmatrix}
      h+u_1-u_2 & -(h+u_1-u_2) & 0 \\
      -(h+u_1-u_2) & (2h+u_1-u_3) & -(h+u_2-u_3) \\
      0 & -(h+u_2-u_3) & (h+u_2-u_3)
    \end{bmatrix}
    \begin{bmatrix} u_1 \\ u_2 \\ u_3 \end{bmatrix} = 
    \begin{bmatrix} f_1 \\ 0 \\ R \end{bmatrix} ~.

The stiffness matrix is a function of u_1, u_2, and u_3.

The boundary condition at x = 0 is u_1 = 0. Therefore, the reduced system of equations can be written as


    \cfrac{AE_0}{h^2}
    \begin{bmatrix}
      (2h-u_3) & -(h+u_2-u_3) \\
      -(h+u_2-u_3) & (h+u_2-u_3)
    \end{bmatrix}
    \begin{bmatrix} u_2 \\ u_3 \end{bmatrix} = 
    \begin{bmatrix} 0 \\ R \end{bmatrix} ~.

The reduced stiffness matrix is a function of u_2 and u_3.

How can we find u_2 and u_3 under these circumstances? We will use the Newton-Raphson numerical technique.

This article is issued from Wikiversity - version of the Wednesday, March 10, 2010. The text is available under the Creative Commons Attribution/Share Alike but additional terms may apply for the media files.