Vous êtes sur la page 1sur 7

UNIVERSITY OF BRISTOL

FACULTY OF ENGINEERING

Third Year Examination for the Degrees of


Bachelor and Master of Engineering

JANUARY 2017 1.5 Hours

FINITE ELEMENT ANALYSIS 3M


MENG33111

This paper contains three questions.


Answer all questions.

All questions carry 12 marks each.


The maximum for this paper is 36 marks.

TURN OVER ONLY WHEN TOLD TO START WRITING

1
Q1 Figure Q1 shows a frame structure comprised of 4 bars connected at frictionless pin-joints. All bars
have the same Young’s modulus, 𝐸. Bars 1 to 3 have cross-sectional area 𝐴, while bar 4 has cross-
sectional area (2√2)𝐴. The local node numbering for each bar is shown on the right. The element
stiffness matrix, 𝐊 (𝐞) , for each bar relates the nodal displacements, 𝐮 = {𝑢1𝑥 𝑢1𝑦 𝑢2𝑥 𝑢2𝑦 }𝑇 , to
the nodal forces, 𝐟 = {𝑓1𝑥 𝑓1𝑦 𝑓2𝑥 𝑓2𝑦 }𝑇 . In the subscripts, 1 or 2 denotes the local node number
and 𝑥 or 𝑦 denote direction.

Figure Q1

The element stiffness matrices, 𝐊 (1) to 𝐊 (4) , for the 4 bars, and the form of the global stiffness matrix
𝐊 are shown below:
1 0 −1 0 0 0 0 0
𝐴𝐸 0 0 0 0 𝐴𝐸 0 1 0 −1
𝐊 (1) = { }, 𝐊 (2) = { },
𝐿 −1 0 1 0 𝐿 0 0 0 0
0 0 0 0 0 −1 0 1
1 0 −1 0 1 1 −1 −1
(3) 𝐴𝐸 0 0 0 0 (4) 𝐴𝐸 1 1 −1 −1
𝐊 = { }, 𝐊 = { },
𝐿 −1 0 1 0 𝐿 −1 −1 1 1
0 0 0 0 −1 −1 1 1
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
∙ ∙ ? ? ? ? ∙ ∙
𝐴𝐸 ∙ ∙ ? ? ? ? ∙ ∙
𝐊= 𝐿 ∙ ∙ ? ? ? ? ∙ ∙ .
∙ ∙ ? ? ? ? ∙ ∙
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
{ ∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙}
(a) Calculate the terms indicated by ? in the 4 × 4 sub-matrix of 𝐊.
(4 marks)
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
∙ ∙ 2 1 0 0 ∙ ∙
𝐴𝐸 ∙ ∙ 1 2 0 −1 ∙ ∙
𝐊=
𝐿 ∙ ∙ 0 0 1 0 ∙ ∙
∙ ∙ 0 −1 0 1 ∙ ∙
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
{∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙}
1 mark per correct quadrant.

2
(b) Carefully explain (i) why only the 4 × 4 sub-matrix of 𝐊 calculated in part (a) is needed to
determine the displacement of the structure under load and (ii) how the reaction forces at nodes
1 and 4 could then be calculated.
(4 marks)
(i) The constraints at nodes 1 and 4 mean than 𝑢1𝑥 = 𝑢1𝑦 = 𝑢4𝑥 = 𝑢4𝑦 = 0. Consequently in the global
stiffness equation, 𝐟 = 𝐊𝐮, the columns of 𝐊 (column numbers 1, 2, 7 and 8) corresponding to these DOF can
be removed as they will be multiplied by zero displacements (1 mark). The applied loads can only be applied
at nodes 2 and 3 in either the 𝑥 or 𝑦 directions. These correspond to rows 3, 4, 5 and 6 in 𝐊. Hence the central
4 × 4 submatrix of 𝐊 describes 4 equations in 4 unknown displacements, which can then be solved (1 mark).
(ii) Once the displacements at all nodes have been calculated, the complete stiffness matrix can be multiplied
by the displacement vector to yield the externally applied forces at all nodes (1 mark for recognising that this
is the point where more of 𝐊 than just the middle 4 × 4 bit is now needed). More efficiently, a new 4 × 4 sub-
matrix (specifically: rows 1, 2, 7, 8; columns 3, 4, 5, 6) of 𝐊 can be multiplied by the 4 non-zero displacements
(1 mark for recognising this).
(c) Bar 4 is removed from the structure. Calculate the terms indicated by ? in 𝐊 for the resulting
3-bar structure, and use these to explain why a static analysis of the resulting 3-bar structure
will fail.
(4 marks)
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
∙ ∙ 1 0 0 0 ∙ ∙
𝐴𝐸 ∙ ∙ 0 1 0 −1 ∙ ∙
𝐊=
𝐿 ∙ ∙ 0 0 1 0 ∙ ∙
∙ ∙ 0 −1 0 1 ∙ ∙
∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙
{∙ ∙ ∙ ∙ ∙ ∙ ∙ ∙}
2 marks for getting 𝐊 right.
Physically, the structure is now a mechanism (1 mark) and a static analysis is meaningless. In terms of an FE
calculation, the relevant part of the stiffness matrix now contains two rows (rows 4 and 6 in the global matrix;
rows 2 and 4 in the sub-matrix) that are (by inspection) linearly dependent. It will therefore have a determinant
of zero and cannot be inverted.

3
Q2 A cantilever structure must be designed to support a coolant pipe in a nuclear power station. One
possible design is shown in Figure Q2. The cantilever is manufactured from welded steel plates and
under normal operation must support a distributed load of 10 kN at the position shown due to the
weight of the pipe and its contents.

Figure Q2

(a) Explain (in words) the key differences in formulation between linear and quadratic finite
elements, and discuss their relative merits.
(3 marks)
Linear/quadratic refers to the order of the shape functions used to approximate displacement within the
element for given nodal displacements (1 mark). In isoparametric elements the shape functions also relate
natural coordinates to global physical coordinates, thus a quadratic isoparametric element can have parabolic
sides to provide a better approximation to curved geometries (1 mark). The advantage of quadratic elements
is that they generally provide a better approximation to the true displacement field than linear ones, hence a
smaller number of nodes can be used to achieve a given level of accuracy – the global matrices will also be
smaller. The penalty for this is that the element formulation is more complex and computationally costly (1
mark).
(b) Why is symmetry often exploited in finite element models? Identify any planes of symmetry
in the design shown in the figure and carefully state the constraints that should be applied to
nodes on the symmetry plane(s).
(3 marks)
Symmetry is exploited to reduce the size of the model. It may also increase model accuracy as in a smaller
model there is less accumulation of numerical error (1 mark). The plane of symmetry in the design is along
its vertical centreline in the end elevation plane (i.e. in the 𝑥 − 𝑦 plane). On this line, the displacement in the
𝑧 direction should be constrained to zero (1 mark), as should rotations about the 𝑥 and 𝑦 axes (1 mark).
(c) A finite element model of the design shown in the figure is tested by applying a distributed
load of 10 kN over the area shown. Under this load, the displacement of cantilever tip is found
to be 2 mm. Discuss how you would validate this result.
(2 marks)

4
To validate the result it would be useful first to estimate the deflection of a simpler cantilever of similar
dimensions using analytical textbook solutions. This would serve to confirm that the FEA result was of the
right order (1 mark). The mesh should then be progressively refined and the behaviour of key parameters of
interest (e.g. tip deflection, peak stress) observed (1 mark).
(d) The safety case for the power station requires various failure scenarios to be modelled. In one
scenario, a component elsewhere in the power station fails and the weight of a large pump is
then also supported by the cantilever. The peak load that the cantilever could experience in this
scenario is estimated to be 500 kN. Finite element analysis is to be used to test whether the
cantilever will still prevent the pipe from falling under this load. What additional effects and
input data would have to be considered in the finite element model in order to model this?
Discuss possible outcomes.
(4 marks)
Based on the result from part (c), if the cantilever was linear, such a load would cause a displacement of
500/10 × 2=100 mm. This is clearly far greater than required to stay in the linear regime, therefore nonlinear
analysis is required (1 mark). Yielding of the material is inevitable, so the model will also need to include
appropriate material model of yielding behaviour (1 mark). One outcome is that the material of cantilever fails
catastrophically by fracture. A likely alternative is that the bracket yields and plastically deforms to some final
position under the applied load. The question would be whether the final position was still horizontal enough
to be able to support the pipe – or whether the pipe would e.g. slide off (up to 2 marks depending on how
intelligent discussion is).

5
Q3 It is proposed to use the 2-noded, 1-dimensional bar element shown in Figure Q3 to analyse the
dynamics of wave propagation along long slender structures. The positions of nodes 1 and 2 are 𝑥1
and 𝑥2 respectively. The natural coordinate, 𝑠, is defined as having value 𝑠1 = 0 at node 1 and value
𝑠2 = 1 as node 2. The displacement within the element in terms of the nodal displacement vector, 𝐮 =
{𝑢1 𝑢2 }𝑇 , is given by:
𝑑(𝑠) = 𝐍(𝑠)𝐮
where the shape functions, 𝑁1 (𝑠) and 𝑁2 (𝑠), in the matrix 𝐍(𝑠) = {𝑁1 (𝑠) 𝑁2 (𝑠)} have the following
properties:
0, 𝑖≠𝑗
𝑁𝑖 (𝑠𝑗 ) = {
1, 𝑖=𝑗

Figure Q3

(a) Assuming a linear displacement distribution within the element, write down suitable shape
functions, 𝑁1 (𝑠) and 𝑁2 (𝑠).
(2 marks)
By inspection:
𝑁1 = 1 − 𝑠
𝑁2 = 𝑠
Alternatively, write shape function as 𝑁𝑖 = 𝑎𝑖 𝑠 + 𝑏𝑖 , form matrix equations for each, e.g.
0 1 𝑎1 1
{ } {𝑏 } = { }
1 1 1 0
and then solve, e.g.
𝑎1 0 1 −1 1 1 −1 1 −1
{𝑏 } = { } { } = −{ }{ } = { }
1 1 1 0 −1 0 0 1
so 𝑁1 = −𝑠 + 1 = 𝑠 − 1 as above and repeat for 𝑁2 .
𝑑
(b) The strain in the element is 𝜖 = 𝑑𝑥 𝑑. The matrix, 𝐁(𝑠), relates strain to nodal displacements
according to:
𝜖(𝑠) = 𝐁(𝑠)𝐮
Calculate the terms in 𝐁 and state its size.
(4 marks)
Strain is 𝑑𝑑⁄𝑑𝑥 but we have displacements as a function of 𝑠. Chain rule:
𝑑𝑑 𝑑𝑑 𝑑𝑠
𝜖= =
𝑑𝑥 𝑑𝑠 𝑑𝑥
First term
𝑑𝑑 𝑑 𝑑
= 𝐍(𝑠)𝐮 = {1 − 𝑠 𝑠}𝐮 = {−1 1}𝐮
𝑑𝑠 𝑑𝑠 𝑑𝑠
Shape functions give relationship between 𝑠 and 𝑥
𝑥(𝑠) = 𝑁1 (𝑠)𝑥1 + 𝑁2 (𝑠)𝑥2 = (1 − 𝑠)𝑥1 + 𝑠𝑥2 = 𝑥1 + 𝑠(𝑥2 − 𝑥1 )

6
Rearrange
𝑥 − 𝑥1
𝑠=
𝑥2 − 𝑥1
So
𝑑𝑠 1
=
𝑑𝑥 𝑥2 − 𝑥1
Therefore final answer is
1
𝜖= {−1 1}𝐮
𝑥2 − 𝑥1
So
1
𝐁= {−1 1}
𝑥2 − 𝑥1
It’s a 1x2 matrix.
(c) The general expression for an element stiffness matrix, 𝐊 (𝑒) , is based on integrating the strain
energy density over the volume of the element, 𝑉, according to:

𝐊 (𝑒) = ∭ 𝐁 𝐓 𝐃𝐁 𝑑𝑉
𝑉

where 𝐃 is the material stiffness matrix. Assuming the element has constant cross-sectional
area 𝐴 and Young’s modulus 𝐸 adapt the above expression for the bar element so that the
integral is with respect to the natural coordinate 𝑠. State the size and terms in matrix D in your
expression. Evaluate the expression to obtain the stiffness matrix 𝐊 (𝑒) for the element.
(6 marks)
Integration is over volume but element has constant cross section and no variation of strain across it. Therefore
volume integral can be reduced to single integral over element length multiplied by area:
𝑥2

𝐊 (𝑒) = ∭ 𝐁 𝐓 𝐃𝐁 𝑑𝑉 = 𝐴 ∫ 𝐁 𝐓 𝐃𝐁 𝑑𝑥
𝑉 𝑥1
Now need to change integration variable to 𝑠. From above
𝑑𝑥 = (𝑥2 − 𝑥1 )𝑑𝑠
So answer is
1

𝐊 (𝑒) = 𝐴(𝑥2 − 𝑥1 ) ∫ 𝐁 𝐓 𝐃𝐁 𝑑𝑠
0
Matrix is 𝐃 = {𝐸} and is of size 1x1. Integrand is therefore
1 −1 1 1 −1
𝐁 𝐓 𝐃𝐁 = 𝐸 { } {−1 1} = 𝐸 { }
(𝑥2 − 𝑥1 ) 1
2 (𝑥2 − 𝑥1 ) −1 1
2

And final answer is


1
1 1 −1 𝐴𝐸 1 −1
𝐊 (𝑒) = 𝐴(𝑥2 − 𝑥1 ) ∫ 𝐸 { } 𝑑𝑠 = { }
(𝑥2 − 𝑥1 ) −1 1
2 (𝑥2 − 𝑥1 ) −1 1
0