Introduction¶
- FEM discretizes a physical domain into many elements
- then reconnects elements at nodes as if nodes were drops of glue
- results in a set of simultaneous algebraic equations
DOF: Finite!
Basic Concepts¶
graph LR
A[<b>Engineering Problems</b>] --> B[<b>Physical Models</b><br>Governing Eqs + BCs]
B --> C[<b>FEM</b><br>KU = F] Vertical machining center
- Divide the domain into a number of small, simple elements
- Obtain the algebraic equations for each element
-
Put all the element equations together
\[ \mathbf{KU} = \mathbf{F} \]- \(\mathbf{K}\): property matrix
- \(\mathbf{U}\): nodal displacement vector
- \(\mathbf{F}\): action
General Steps: Structural Problem¶
- Identify the geometry of the problem
-
Select a displacement function
Axial Tension
-
choose \(u(x) = a_0 + a_1 x\) for the displacement function
you can't choose \(u(x) = a_0 + a_2 x^2\), because it doesn't satisfy the convergence condition
-
determine \(a_0, a_1\) by BC: \(u(0) = u_i, u(L) = u_j\)
\[ \begin{aligned} a_0 &= u_i \\ a_1 &= \frac{u_j - u_i}{L} \end{aligned} \] -
rewrite \(u(x)\) in terms of nodal displacements \(u_i, u_j\):
\[ \begin{aligned} u(x) &= u_i + \frac{x}{L} (u_j - u_i) \\ &= \Big(1 - \frac{x}{L}\Big) u_i + \frac{x}{L} u_j \\ :&= N_i u_i + N_j u_j \end{aligned} \]- \(N_i, N_j\): shape functions
-
-
Define the strain/displacement and stress/strain relationships
\[ \boldsymbol{\varepsilon} = \nabla \mathbf{u}, \quad \boldsymbol{\sigma} = \mathbf{D} \boldsymbol{\varepsilon} \] -
Derive the element stiffness matrix and equations
\[ \mathbf{k}_e \mathbf{u}_e = \mathbf{f}_e \]- 直接平衡法
- 能量法
- 最小势能原理(弹性材料)
- Castigliano 定理(弹性材料)
- 虚功原理(任何材料)
- 最小残差法(weighted residual method)
- Galerkin method
- Assemble the element equations to obtain the global equations, introduce BCs
\[ \mathbf{k}_e \mathbf{u}_e = \mathbf{f}_e \xrightarrow{\text{Direct stiffness method}} \mathbf{KU} = \mathbf{F} \]
Commercial FEM Software Packages
- ANSYS (General-purpose, widely used in industry)
- Abaqus (General-purpose, widely used in academia)
- COMSOL (Multiphysics)
- NASTRAN (General-purpose)
- HyperMesh (Pre-processing, no solver)
- COSMOS
Errors of FEM¶
FEM only obtains approximate solutions with inherent errors. The errors can be reduced by:
- Discretization error
- Modeling error
- Numerical error
History of FEM
- 1941, Hrenikoff (Lattice framework to model membrane and plates)
- 1943, Courant (Shape functions for triangular subregions)
- 1954, Argyris and Kelsey (Matrix structural analysis method)
- 1956, Turner (Direct stiffness method)
- 1960, Clough (Coined the term "finite element method", plane problems)
- 1965, Zienkiewicz (First book on FEM)
- 1965, 冯康 (Dicretization numerical method based on variational principles)