You are on page 1of 20

CEG8315

Advanced structural modelling

3. Implementation – linear statics

3e. Isoparametric formulation

School of Engineering

School of Engineering
ISOPARAMETRIC FORMULATION

The isoparametric formulation makes it possible to generate elements that are non-rectangular and have
curved sides, whilst remaining compatible with neighbouring elements.

These shapes have obvious uses in grading a mesh from coarse to fine, in modelling arbitrary shapes, and
in modelling curved boundaries.

The isoparametric family includes elements for plane, solid, plate, and shell problems. There are also
special elements for fracture mechanics and elements for non-structural problems (e.g., thermal, electrical,
etc.).

School of Engineering
ISOPARAMETRIC FORMULATION

School of Engineering
ISOPARAMETRIC BAR ELEMENT

The basic principles of the isoparametric formulation can be demonstrated by the formulation of a
straight 3-node bar element.

Natural co-ordinate is 𝜉 , with 𝜉 = ±1 defining the ends of the bar, regardless of the physical length, L.

Note that 𝜉 = 0 is not necessarily at the geometric centre of the bar, and that the origin of the cartesian
(or global) co-ordinate x may not (will normally not) coincide with 𝜉 = 0 .

Natural co-ordinate 𝜉 is attached to the bar and remains an axial co-ordinate regardless of how the bar
is orientated in the global co-ordinates 𝑥, 𝑦, 𝑧. For convenience in this example, and not necessity, 𝜉 and
𝑥 are colinear.

School of Engineering
ISOPARAMETRIC BAR ELEMENT

We begin with an assumed field, quadratic in this case, written in terms of natural co-ordinate 𝜉 for both
co-ordinate 𝑥 and axial displacement 𝑢:

𝑥 = 𝐍 𝑥! 𝑥" 𝑥# T and 𝑢 = 𝐍 𝑢! 𝑢" 𝑢# T,

! !
where, 𝐍 = "
(−𝜉 + 𝜉 ") "
(𝜉 + 𝜉 ") 1 − 𝜉 " .

School of Engineering
ISOPARAMETRIC BAR ELEMENT

Construction of the element stiffness matrix requires the strain-displacement relation. In the case of the
bar element, there is only one strain, which is the axial strain 𝜖! ,

𝑢!
%& % % %' %
𝜖$ = = 𝐍 𝑢" , where, = %$ %'
%$ %$ %$
𝑢#
%
The chain rule for must be used because 𝐍 is expressed in terms of 𝜉 and not 𝑥.
%$

%'
Unfortunately, is not immediately available.
%$

%$
We must first calculate its inverse, from 𝑥 = 𝐍 𝑥" 𝑥# 𝑥$ T, with
%'
! ! !"
𝐍 = (−𝜉 + 𝜉 ") (𝜉 + 𝜉 ") 1 − 𝜉 " . Let J = , then,
" " !#

𝑥! 𝑥!
% ! !
J=
%'
𝐍 𝑥" = "
(−1 + 2𝜉) "
(1 + 2𝜉) −2𝜉 𝑥" .
𝑥# 𝑥#
J is called a Jacobian. It can be regarded as a scaling factor that describes the physical length 𝑑𝑥
associated with a reference length 𝑑𝜉, such that 𝑑𝑥 = J 𝑑𝜉.

School of Engineering
ISOPARAMETRIC BAR ELEMENT

Calculating the Jacobian – an example.

Assume 𝑥" = 0, 𝑥# = 𝐿, 𝑥$ = 𝐿/2.

𝑥! 𝑥!
% ! !
J = %' 𝐍 𝑥" = "
(−1 + 2𝜉) "
(1 + 2𝜉) −2𝜉 𝑥"
𝑥# 𝑥#

0
! !
J= (−1 + 2𝜉) (1 + 2𝜉) −2𝜉 𝐿 = 𝐿/2.
" "
𝐿/2

Note
% ' # '
𝑑𝑥 = J 𝑑𝜉 ⟶ ∫$ 𝑑𝑥 = ∫&' J 𝑑𝜉 = 𝐿 (( = 𝐿 ⟶ ok.
&'

School of Engineering
ISOPARAMETRIC BAR ELEMENT

Using the Jacobian to calculate the element stiffness matrix.

The bar element stiffness matrix is defined as:

) !

𝐤 = 2 𝐁 * 𝐴𝐸 𝐁 𝑑𝑥 = 2 𝐁 * 𝐴𝐸 𝐁 J 𝑑𝜉
( +!

where,

1 𝑑 1 1 1
𝐁 = 𝐍 = (−1 + 2𝜉) (1 + 2𝜉) −2𝜉
J 𝑑𝜉 J 2 2

Only if node 3 is at the centre of the bar does J reduce to the constant value J = L/2. The specific form
of J depends on the numerical values assigned to 𝑥" , 𝑥# , 𝑥$ .

In general, J is a function of 𝜉, meaning that 𝐁 will contain 𝜉 in both numerator and denominator of
every term.

Therefore, the expression for the stiffness matrix 𝐤 cannot be conveniently integrated in closed form.
In practice, numerical integration (e.g., Gauss quadrature) is used instead.

School of Engineering
ISOPARAMETRIC FORMULATION
The preceding 3-node bar element illustrates some of the concepts and manipulations associated with
isoparametric elements.

Using the same principles, a four-node plane element of arbitrary geometry (left image below) in cartesian
co-ordinates, is mapped to a square in isoparametric space 𝜉 − 𝜂 (right image below).

In the form of a square, the element formulation will be consistent with adjoining isoparametric
quadrilaterals. Integration is again performed between limits ±1 in 𝑑𝜉𝑑𝜂, with the Jacobian, J, scaling the
surface area of the physical element (left) with respect to the isoparametric mapping (with area of 4).

For 3D elements, the Jacobian scales volumes.

School of Engineering
GAUSS QUADRATURE

c b a
k = ò ò ò BT E B dx dy dz
24´24 - c - b - a 24´6 6´6 6´24

School of Engineering
GAUSS QUADRATURE

+1
I= ∫ φ dξ ≈ W φ 1 1
+W2φ2 +W3φ3 +!+Wnφn
−1

+1 +1 % +1 (
I= ∫ ( )
∫ φ ξ ,η d ξ dη ≈ ∫ '&∑Wiφ (ξi ,η)*) dη = ∑∑ WiW jφ (ξi ,η j )
−1 −1 −1 i i j

+ +1 +1
I= ∫ ∫ ∫ φ (ξ ,η,ζ ) d ξ dη dζ ≈ ∑∑∑ W W W φ (ξ ,η ,ζ i j k i j k
)
−1 −1 −1 i j k

Order n Degree of Sampling (Gauss) point locations Weight factors


precision ξi Wi

1 1 0. 2.

2 3 ±1 ⁄ √3 1.

3 5 ± √0.6 5/9
0. 8/9

School of Engineering
GAUSS QUADRATURE

+1
I= ∫ φ dξ ≈ W φ1 1
+W2φ2 +W3φ3 +!+Wnφn
−1

Order n Degree of Sampling (Gauss) point [G.P.] locations


Weight factors
precision ξi Wi

1 1 0. 2.

2 3 ±1 ⁄ √3 1.

3 5 ± √0.6 5/9
0. 8/9
'
Example: 𝐼 = ∫&' 𝑥 ( 𝑑𝑥

1 G.P., 𝐼 = 0 # ×2 = 0

# #
2 G.P., 𝐼 = 1⁄ 3 ×1 + 1⁄ 3 ×1 = 2⁄3

# #
3 G.P., 𝐼 = 0.6 × 5⁄9 + 0.6 × 5⁄9 + 0 # × 8⁄9 = 0.6× 5⁄9 ×2 = 6⁄10 × 5⁄9 ×2 = 6⁄9 = 2⁄3

Analytical solution = 𝑥 $ ⁄3|"%" = 1⁄3 − − 1⁄3 = 2⁄3

School of Engineering
ISOPARAMETRIC FORMULATION – VALIDITY
The critical test of element validity is the patch test. Necessary conditions for passing the patch test are
interelement continuity and completeness of the polynomial field used to define the element shape
functions. Isoparametric elements containing these features should have properties that would be
expected to enable them to pass the patch test.

Passing the patch test (or fulfilling continuity and completeness requirements) says nothing about the
accuracy of isoparametric elements in a coarse mesh, convergence rate with mesh refinement, or how
accuracy declines as element geometry is distorted from a compact and regular shape.

As sides become curved and as side nodes become unevenly spaced, an element tends to lose its ability
to represent quadratic and higher polynomials in 𝑥, 𝑦, 𝑧 co-ordinates.

School of Engineering
REQUIRED ORDER OF QUADRATURE
For numerically integrated elements, “full integration” is defined as a quadrature rule sufficient to provide
the exact integrals of all terms 𝑘&' in the element stiffness matrix if the element is undistorted (e.g., if a
quadratic element has straight sides and mid-side nodes). The same “full integration” rule will not exactly
integrate all 𝑘&' if sides are curved or if side nodes are offset from the midpoints, since J is not constant
throughout the element. The use of full integration is the only way of avoiding potential pitfalls such as
mesh instabilities.

A lower-order quadrature rule – “reduced integration” – may be useful to reduce computational cost (by
using fewer sampling points in the Gauss quadrature scheme) or to soften an element (countering the
overly stiff behaviour associated with an assumed displacement field). Softening comes about because
certain higher-order polynomial terms vanish at Gauss points of a lower-order rule, so that these terms
make no contribution to the strain energy, and consequently reducing the stiffness. Reduced integration
may increase the accuracy of a finite element analysis. It should not be used solely to lower
computational cost.

The number of Gauss points has a lower limit – in the limit of mesh refinement, the element volume must
be integrated exactly. For an isoparametric element based on an assumed displacement field, the best
quadrature rule is usually the lowest-order rule that computes volume correctly and does not produce
instability.

School of Engineering
ELEMENT AND MESH INSTABILITIES
An “instability” may also be referred to as a spurious singular mode. In structural mechanics, an instability
may be known as a mechanism, a kinetmatic mode, an hourglass mode, or a zero-energy mode.

The term “zero-energy mode” refers to a nodal displacement vector D that is not a rigid-body motion but
nevertheless produces zero strain energy D ( K D /2. Instabilities arise because of shortcomings in the
element formulation process, such as the use of a low-order Gauss quadrature rule.

In the present context, an instability has nothing to do with buckling problems of structures.

A structure that appears adequately constrained may yet have an instability that makes K singular, or
unstable elements may combine to form a structure that is stable but unduly susceptible to certain load
patterns, so that computed displacements are excessive.

School of Engineering
ELEMENT AND MESH INSTABILITIES

Explaining the term “zero-energy mode”


The strain energy in an element, 𝑈) , is written as,

1 (
1 ( 𝑻
1 𝑻
𝑈) = d k 𝐝 = d I B E B 𝑑𝑉 𝐝 = I 𝝐 E 𝝐 𝑑𝑉
2 2 *! 2 *!

When k is computed using numerical integration (e.g., Gauss quadrature), it contains only the
information that can be sensed at the sampling points (e.g., Gauss points) of the quadrature rule.

If it happens that strains 𝝐 = B d are zero at all sampling points for a certain deformation mode d ,
then 𝑈) will vanish (e.g., 𝑈) = 0) for that d , and, according to the equation above, k will be a zero-
stiffness matrix. If 𝑈) = 0 when d is not a rigid-body motion, then an instability is present.

An element that displays a mechanism is said to be “rank deficient” – the rank (minimum number of
linearly independent rows or columns) of k is less than the number of element d.o.f. minus the number
of rigid-body modes.

An instability in an existing k can be detected by means of an eigenvalue test.

(a) Mesh of four bilinear elements,


showing single Gauss points in each
element. (b, c, d) Possible mechanisms
(“hourglass” modes).

School of Engineering
REMARKS ON STRESS COMPUTATION
Element stresses, 𝝈

𝝈 = E 𝐁 𝐝 − 𝝐𝟎 + 𝝈𝟎

In isoparametric elements, 𝐁 is a function of the natural co-ordinates, and 𝝈 contains stresses referred
to in the (local) cartesian co-ordinate system 𝑥, 𝑦, 𝑧 .

Q. Where in the element should stresses be calculated?


A. For isoparametric elements, stresses are often most accurately calculated at Gauss points of a
quadrature rule one order less than that required for full integration of the element stiffness matrix.

Stresses at Gauss points can be interpolated or extrapolated to other points in the element. Extrapolation
to element nodes is useful in assessing mesh refinement requirements.

Example of natural co-ordinate systems


used in extrapolation of stresses from
Gauss points to element nodes – e.g., @
point 3, r = s = 1, and 𝜉 = 𝜂 = 1⁄ 3.

School of Engineering
EFFECT OF ELEMENT GEOMETRY
Accuracy in predicting displacements and stresses is affected by element distortion, changes in Gauss
quadrature rule, and changes in element aspect ratio.

The table below illustrates the behaviour of the bilinear element when its k is formed by four-point Gauss
quadrature. Results are expressed as the ratio of computed value to the value given by beam theory,
illustrating the poor performance of the bilinear element in bending (particularly for deep beams), and the
decline in accuracy with increasing element distortion.

Stresses and deflections in cantilever beams of constant thickness under transverse tip load P. Length = 10, depth = 2,
𝜐 = 0.25. Values from beam theory = 1.000, of which 3% of 𝑣! is due to transverse shear deformation.

School of Engineering
GEOMETRIC DISTORTION
The example elements illustrated below have very poor geometries, and should not be used . If used, and
if surrounded by elements of acceptable geometry, stresses will be poor in and very near the distorted
element, but reasonable at some distance away because of St. Venant’s principle.

In the first element (a), there is a singularity at node 3, where the determinant of the Jacobian, J , is
zero. In elements (b) – (d), part of each element falls outside the intended boundaries and J < 0 at one
of the Gauss points. These distortions do not prevent the elements from passing constant-strain patch
tests, but drastically reduce the ability of the elements to represent more complicated states of
deformation.

Badly shaped elements, showing


Gauss points bounded by lines of
constant 𝜉 and 𝜂. (a) Interior angle at
node 3 is 180o. (b) Node 1 moved to
the centre of the original rectangle. (c)
Node 7 moved from top mid-side to
corner 3. (d) Two side nodes moved
to centre of the original rectangle.

School of Engineering
CEG8315
Advanced structural modelling

3. Implementation – linear statics

3e. Isoparametric formulation

School of Engineering

School of Engineering

You might also like