Lecture 4

Special Topics of FEA

Somenath Mukherjee Gangan Prathap

Scientist, Director,
Structural Technologies Division, National Institute of Science and
National Aerospace Laboratories (NAL), Information Resources (NISCAIR),
Bangalore, New Delhi,
Karnataka, India India

1
 Lecture 4
Special Topics of FEA
Chapters

1. Shear Locking in Timoshenko beam elements.

2. Error Analysis in Computational
Elastodynamics.
3. Rank Deficiency in elements.

2
 Lecture 4
Special Topics of FEA

Chapter 1

Shear Locking in Timoshenko

beam elements
(A Pathological Problem)

3
 1.1 The Pathological problem of locking
Locking is a pathological problem encountered in formulating a
certain class of elements for structural analysis, although these
elements satisfy completeness and continuity requirements.

• Locking causes slow convergence even for very fine mesh.

• Locking is manifested as Spurious Stiffening and Stress
Oscillations.

Explanations:
(1) Locking is caused by ill conditioning of the stiffness matrix due to the
very large magnitude of the shear stiffness terms as compared to the
those of bending stiffness (Tessler and Hughes).

(2) Locking occurs due to coupling between the shear deformation and
bending deformation, and that it can be eliminated by appropriate de-
coupling (Carpenter et al).

(3) Elements lock because they inadvertently enforce spurious constraints

that arise from inconsistencies in the strains developed from the
assumed displacement functions. (Prathap et al). 4
 1.2 The Shear-flexible beam (Timoshenko)
In the classical Euler beam (meant only for thin
beams), it has been shown that despite the
presence of shear stress in the beam sections, the
shear strain is ignored.
The Euler beam is of infinite shear rigidity (!)

For thick beams (of wider webs), the Euler beam

theory is not valid. Shear deformation of the web
requires shear-flexible formulations. Prof S P Timoshenko

Euler beam Timosheko beam

x
dw/dx
dw/dx
z,w x
=dw/dx NA zx= - dw/dx
5
 Elementary beam theory as constrained media problem
Euler beam model 1
L
dθ
2 L
1
L
dw
2
δΠ = δ EI dx − qwdx + κ θ− =0
2 dx 2 dx
x 0
Equilibrium Equations
0 0

d 2θ dw
EI 2
−κ θ − =0 ....(i )
dx dx
dw/dx
dθ d 2 w
z,w x κ −
dx dx 2
=q i.e.
d
dx
κ θ−
dw
dx
=q ....(ii )

NA Combining (i ) & (ii )

=dw/dx d d 2θ
EI 2 − q = 0 ...(iii )
dx dx
The Euler beam has Boundary conditions at x = 0 & x=L
infinite shear rigidity dθ
Either = 0 or δθ = 0
EI
dx
But the practice of using a dw
Either κ θ− = 0 or δw = 0
large shear rigidity for dx
thin beams creates a As κ → ∞ θ →
dw
problem called Shear dx
Locking in shear-flexible d 4w
beam elements. Equation (iii ) reduces to EI 4
−q →0 6
dx
Equilibrium equations of the Shear flexible (deep) beams
L 2 L 2 L
1 dθ 1 dw
δΠ = δ EI dx + kGA θ − − qwdx = 0
0
2 dx 0
2 dx 0
dw/dx Equilibrium Equations
-dM/dx-V=0
d 2θ dw
EI 2 − kGA θ − =0 ....(i )
dx dx (1.2)
V dw/dx
dθ d 2 w d dw
kGA − =q i.e. kGA θ − =q ....(ii )
dx dx 2 dx dx
Combining (i ) & (ii ) dV/dx=q
V d d 2θ (1.3)
EI 2 − q = 0 ...(iii )
dx
Shear strain dx
Boundary conditions at x=0 & x=L
= - dw/dx=V/(kGA) dθ
Either EI =0 or δθ = 0
(1.1) dx
dw
Either kGA θ − =0 or δw = 0
dx
Shear rigidity: kGA
7
 Shear rigidity of deep beams
dw/dx γ τ=
VQ
γ=
τ
=
VQ
Ib G GIb
L h/2 L h/2
1 τ2 V 2Q 2
U shear = .τ .γ .dV = . .(bdy)dx = . .(bdy)dx
V dw/dx V
2 0 y =− h / 2
2G 0 y = − h / 2 2G ( Ib)
2

L h/2
1 V 1 1 Q2
U shear = V dx = dy
0
2 kGA kGA G −h / 2 I 2b
1
V k= h/2
A Q2
y I2 b
dy
−h / 2
L L
1 V 1 V dw
U shear = V dx = V γ dx γ= =θ −
2 kGA 2 kGA dx
h 0
L
0
L L 2
1 1 1 dw
U shear = V γ dx = kGA(γ ) 2 dx = kGA θ − dx
0
2 0
2 2 0
dx (1.4)

b k is called the shear correction factor

k=5/6 for a rectangular section 8
Example 1. Find the tip deflection of a cantilever subjected to a
concentrated tip load P. (Include shear deformation)
L P

Deflection at the free end :

PL3 PL3 P PL3 3EI
δ= +γL = + L= 1+
3EI 3EI kGA 3EI kGAL2

For thin beams,

kGAL2 EI
→ ∞, 2
→0
EI kGAL

9
1.3 Formulation of the two-noded Timoshenko Beam Element
(Using Linear Lagrangian C0 Shape Functions)
Element displacement and geometry (iso-parametric):
1−ξ 1+ ξ −1 ≤ ξ ≤ 1
wh = N1w1 + N 2 w2 θ h = N1θ1 + N 2θ 2 N1 = N2 =
2 2
e Le 2x
x = N1 x1 + N 2 x2 with x1 = 0, x2 = L x = (ξ + 1), ξ= e
−1
2 L
Le
dx = dξ
2
w1 w1,F1 w2,F2
wh N1 0 N2 0 θ1
h
= = [ N ]{δ e } (1.5) Le
θ 0 N1 0 N2 w2
θ2
1,M1 2,M2

Element Strain vector: ξ = −1 ξ =1

dθ h dx 0 − 1/ L 0 1/ L
(ε ) =
h
= {δ e }
θ h − dwh dx 1 / L (1 − ξ ) / 2 − 1 / L (1 + ξ ) / 2

{ε } = [B]{δ }
h e (1.6) 10
 Element stress resultants :

M EI 0 dθ / dx
= = [ D][ B]{δ e } (1.7)
V 0 kGA θ − dw / dx

Element potential energy:

w1,F1 w2,F2
L 2 L 2 L
1 dθ 1 dw
Π= EI dx + kGA θ − − qwdx − {δ e }T R e }
0
2 dx 0
2 dx 0
Le
1
(
Π = {δ e }T [ K e ]{δ e } − {δ e }T {F e } + {R e }
2
)
1,M1 2,M2
Equilibrium
δΠ = 0 [ K e ]{δ e } = {F e } + {R e } (1.8)

Element Stiffness matrix Element Force vector

1 1
e Le
T e TLe
[ K ] = [ B ] [ D ][ B ] dξ {F } = [ N ] q dξ
−1
2 (1.9) 2 (1.10)
−1 11
 Using a 2 point Gauss integration the stiffness matrix is

1
Le
[ K ] = [ B ] [ D ][ B ] dξ = [ K eb ] + [ K e s ]
e T

−1
2
0 0 0 0 1 Le / 2 −1 Le / 2
EI 0 1 0 −1
kGA Le / 2 ( Le ) 2 / 3 − Le / 2 ( Le ) 2 / 6
[ K e ] = [ K eb ] + [ K e s ] = e + e
L 0 0 0 0 L −1 − Le / 2 1 − Le / 2
0 −1 0 −1 Le / 2 ( Le ) 2 / 6 − Le / 2 ( Le ) 2 / 3

(1.11)

12
FE results of analysis of deep beam cantilever beam under tip load

L P
E=1000
G=375
b=1, h=1
L=4 No of Normalized tip
Locked results elements Displacement
y (Locked)
Observations 1 0.2 (10)-5
•Large errors
h 2 0.8 (10)-5
•A pattern in the
error. 4 0.32 (10)-4

b •Slow convergence 8 0.128 (10)-3

16 0.512 (10)-3
13
 Antidote for shear locking.
Use a 1 point (instead of 2 point) Gauss integration
scheme for the stiffness matrix is

0 0 0 0 1 Le / 2 −1 Le / 2
e e e EI 0 1 0 −1 kGA Le / 2 ( Le ) 2 / 4 − Le / 2 ( Le ) 2 / 4
[ K *] = [ K b ] + [ K s *] = e + e
L 0 0 0 0 L −1 − Le / 2 1 − Le / 2
0 −1 0 −1 Le / 2 ( Le ) 2 / 4 − Le / 2 ( Le ) 2 / 4

(1.12)
1

Normalized
displacement

No. of elements
1 2 3 4 5
Magic! An error in the integration eliminates locking! WHY ? 14
FE results of analysis of deep beam cantilever beam under tip load
L P
E=1000
G=375
b=1, h=1
L=4
Number Normalized tip Normalized tip
y of Displacement Displacement
elements (Locked) (Lock-free,
Reduced Int.)
h 0.2 (10)-5 0.75
1

2 0.8 (10)-5 0.75+0.75/4

b
=0.9375

15
Example problems solved using a single Timoshenko beam element

Observations: Spurious shear oscillations and bending

stiffening for the locked case. 16
 1.4 Explanations for the origin of locking
(The field-consistency paradigm)
L q

Linear displacements: θ h = ax wh = bx
dw h dθ h
Shear strain γ = θ −
h
= ax − b Bending strain =a
dx dx
Rayleigh-Ritz procedure L 0 L3 / 3 − L2 / 2 a 0
EI +κ =
0 0 − L2 / 2 L b qL2 / 2

− 3qL2 qL 1.5qL3
a= b=− −
Element locks when shear 12 EI + κL2 2κ 12 EI + κL2

rigidity is increased
dθ h
indefinitely. As κ → ∞, a → 0, b→0
dx
→0

The parameter a effects both bending and shear strains. It is17 a

spurious constraint that stiffens bending as well as shear strains
 1.5 Explanation of shear locking in the element by Field-
Consistency Theory
The shear strain in the element is θ h - dwh dx = α + βξ
where α = (θ + θ ) 2 − (w − w ) L and β = (θ − θ ) 2
2 1 2 1 2 1

• For thin beams, the shear strain energy term vanishes, leading to two
constraints: α→0 β→0
(First constraint is physically meaningful in terms of the equivalent
Euler beam model, but the second constraint is a spurious one.
The spurious term β effectively enhances the element's bending
stiffness to EI*=EI+kGA(Le)2 /12, where EI and kGA are the bending
and shear rigidities respectively of the actual beam, leading to locking.
wLF wL = I * I = 1 + kGAL2 (12EI ) = 1 + e

e=kGA(Le)2/(12EI)=K/n2 , (l=total beam length, n=total number of

equal elements, Le= element length=l/n).

The parameter e becomes larger for thinner beams, leading to

spuriously high bending stiffness, and spurious shear strain
oscillations in the elements.
18
 1.6 How shear locking is eliminated by reduced integration
The integrand in the element stiffness matrix [Ke] is quadratic, so we need a
2 point Gauss rule for exact integration. This element suffers shear locking.
1
Le
[ K ] = [ B ] [ D ][ B] dξ = [ K e b ] + [ K e s ]
e T

−1
2
0 0 0 0 1 Le / 2 −1 Le / 2
e e e EI 0 1 0 −1 kGA Le / 2 ( Le ) 2 / 3 − Le / 2 ( Le ) 2 / 6
[K ] = [K b ] + [K s ] = e +
L 0 0 0 0 Le −1 − Le / 2 1 − Le / 2
0 −1 0 −1 Le / 2 ( Le ) 2 / 6 − Le / 2 ( Le ) 2 / 3

A reduced integration actually eliminates (ignores) the spurious term of the

shear strain (associated with linear variation in ) so that only constant terms
are needed to be integrated. This elimination of the spurious constraint is done
by a 1 point Gaussian rule for integration.

0 0 0 0 1 Le / 2 −1 Le / 2
EI 0 1 0 −1 kGA Le / 2 ( Le ) 2 / 4 − Le / 2 ( Le ) 2 / 4
[ K e *] = [ K eb ] + [ K e s *] = e + e
L 0 0 0 0 L −1 − Le / 2 1 − Le / 2
0 −1 0 −1 Le / 2 ( Le ) 2 / 4 − Le / 2 ( Le ) 219/ 4
If one uses a Reduced Integration scheme with a one-point rule of
Gauss Quadrature, instead of the two-point rule necessary for
accurate integration in the shear strain energy, it leads to

· Elimination of shear locking by releasing the stiffening constraint β.

· Elimination of spurious shear stress oscillations.

Reduced integration effectively drops the

Second Legendre Polynomial from the shear
strain,
α + βξ → α

20
The Function Space Approach
to Locking Problems

21
 1.7 Definition of the Inner product
The inner product for the Timoshenko beam element is defined
through the symmetric bilinear forms:

h
a(u , u ) = e
{ε }
h T EI
0
0
kGA
{ε }dx
e

{ε }
1
h T EI 0 Le
= {ε } dξ =< ε h , ε >
−1
0 kGA 2

a(u h , u h ) e = {ε }
h T EI
0
0
kGA
{ }
ε h dx
e

{ε } { }
1 e
h T EI 0 h L
= ε dξ =< ε h , ε h >
0 kGA 2 (1.13)
−1

EI=Flexural Rigidity, kGA=Shear Rigidity

22
 1.8 The B Subspace
The B subspace is the space in which the column vectors of the
strain-displacement matrix [B] lie.
0 − 1/ L 0 1/ L (1.14)
[B ] =
1 / L (1 − ξ ) / 2 − 1 / L (1 + ξ ) / 2

The Gram-Schmidt Algorithm for getting the orthogonal basis vectors

spanning the B Space:
{v1} = {b1}
k bk +1, v j (1.15))
{vk +1} = {bk +1} − {v j }
j =1 vj ,vj

After scaling, only TWO NON-ZERO orthogonal basis vectors are

obtained that span the B Space (of 2 dimensions, m=N-R=4-2=2)
0 2/ L (1.16)
{v1} = , {v2 } =
1 ξ
2
B ⊂ Pnr==22 ; P22 = {p} : {p} = {ai }ξi −1 ,−1 ≤ ξ ≤ 1 , { ai } ∈ R 2
i =1

dim( B) = 2 < dim Pnr==22 = 2 × 2 = 4

(1.17) 23
 1.9 Strain projections on the B Subspace;
Shear Locking
Orthogonal Projection of the Analytical Strain onto the B Subspace yields
the FEA computed element strains (best-fits).
< ε ,vj >
{ε } = {ε } =
h m=2

j =1 < v , v >
{v j }, < v1 , v2 >= 0 (1.18)
j j

However, we have problems for thin beams:

1. The bending strain is a lot smaller than the analytical one, showing
that spurious bending stiffness has been introduced through FEA .
2. There is spurious shear strain oscillation in FEA results.
3. Slow Convergence even with many elements.

These are the symptoms of locking

24
 Locked FEA solutions agree with the best-fit strain vector at
the element level. Thus locked solutions are variationally
correct
ε h = ε = Best − fit

< ε ,vj >

{ε } = {ε } = j =1
h m=2
{v j }, < v1 , v2 >= 0
< vj ,vj >

A best fit satisfies the Projection Theorem (Pythagoras)

2 2 2
ε −ε = ε −ε

h 2 2 h 2
Thus ε −ε = ε −ε

i.e. The Energy of the Error= Error of the Energies

25
26
 TABLE 1
Analytical strains and their locked projections as finite element strains
e=kGAL2/(12EI).

Cantilever with tip Cantilever with tip load

moment M0 P
Analytical M 0 / EI PL( 1 + ξ ) /( 2 EI )
{ε } = {ε } =
strain 0 P / kGA
vector
( M 0 / EI ) /( 1 + e )
Locked − −
( PL / 2 EI ) /( 1 + e )
strain {ε } = 6 e M 0ξ {ε } = P 3eξ
(1+ )
vector ( 1 + e ) LkGA kGA 1+ e

{ε }= {ε }= < ε,vj > 0 2/ L

m=2
h
{v j }, v1 v2 {v1} = , {v2 } =
j =1 < v j , v j >
1 ξ

FE Strain vectors exactly agree with these orthogonal projections of analytical strains
27
 TABLE 2
Error norm square for locked strain projections with the linear
two noded Timoshenko beam element. e=kGAL 2/(12EI)
1
2
q =
L
{q}T [D] {q} dξ {q} = {ε }− {ε }
2 −1

Case Locked Solution

Cantilever
2 L 2M 02 e
with tip q = .
2 EI 1 + e
moment, Mo
Cantilever
with tip 2 L (P )2 e 1
q = . +
transverse 2 2EI 1+ e 3
load P

{ε }= {ε }
h 2
q = ε −ε h 2
= ε
2
−ε h 2
28
1.10 The Function Space explanation of locking and its
elimination
The original field-inconsistent [B] matrix is
0 − 1/ L 0 1/ L
[B ] =
1 / L (1 − ξ ) / 2 − 1 / L (1 + ξ ) / 2

Locking occurs because the 2-dimensional B subspace

is field-inconsistent, which cannot be spanned by the
standard basis vectors of its 4-dimensional parent
space P 22 (linear in ξ),
{L1} = [0, 1]T , {L2} = [1, 0]T, {L3} = [0, ξ]T, {L4} = [ξ, 0]T.,
(1.19)
Actually, the field-inconsistent B space is spanned by
non-standard basis vectors, {v } = 0 , {v } = 2 / L
1
1
2
ξ 29
 1.11 Elimination of shear locking
Reduced Integration effectively sets the highest order
Legendre Polynomial ξ in the [B] matrix to zero.
It replaces [B] by a (modified) [B*].
Lock-free strain vector is expressed as,
{ε h *} = 0 − 1 / L 0 1/ L {δ *e } = [B * ] {δ *e } (1.20)
1/ L 1/ 2 1/ L 1/ 2

A new field-consistent space B* emerges from [B*]. This

lockfree, field-consistent space B* is two-dimensional,
and can be spanned by the standard orthogonal basis
vectors, 0 1
{v1*} = , {v2 *} =
1 0 (1.21)
B* ⊂ Pnr==12 ; P12 = {{p} : {p} = { ai } , { ai } ∈ R 2 }
dim( B) = 2 = dim Pnr = 2 × 1 = 2 (1.22)
30
Lockfree stiffness matrix for the Timoshenko beam is obtained
from the field-consistent (lockfree) strain-displacement matrix
[B*] with exact integration

1 e
e T L e e
[ K *] = [ B*] [ D][ B*] dξ = [ K b ] + [ K s *]
−1
2

0 0 0 0 1 Le / 2 −1 Le / 2
e e e EI 0 1 0 − 1 kGA Le / 2 ( Le ) 2 / 4 − Le / 2 ( Le ) 2 / 4
[ K *] = [ K b ] + [ K s *] = e + e
L 0 0 0 0 L −1 − Le / 2 1 − Le / 2
0 −1 0 −1 Le / 2 ( Le ) 2 / 4 − Le / 2 ( Le ) 2 / 4

(1.23)

31
 1.12 Orthogonal Projection on B* space
In general, Reduced Integrated FEA results are NOT
variationally correct. (RI is a variational crime !)
Reduced Integrated FEA strains will agree with the
best-fit solution, provided the following rule holds
good,
{F e E }= − [[B]− [B *]]T [D]{ε }dx = 0 (1.24)
Then: e

< ε ,ν i * >
{ε *} = {ε *} =
h m
{ν i *}, < ν i *,ν j * >= 0 for i≠ j (1.25)
i =1 < ν i *,ν i * >
When this extraneous force {F e E }does not vanish, then
the best-fit solution (on the B* space) will suffer
additional strain from this extraneous force vector,
over the lockfree (reduced integrated) FEA solution.

32
 TABLE 3
Analytical strains and their locked and lockfree projections as finite element strains
e=kGAL2/(12EI).

Cantilever with tip Cantilever with tip

moment M0 load P
Analytical M 0 / EI PL( 1 + ξ ) /( 2 EI )
{ε } = {ε } =
strain 0 P / kGA
vector
Locked ( M 0 / EI ) /( 1 + e ) ( PL / 2 EI ) /( 1 + e )
− −
strain {ε } = 6 e M 0ξ {ε } = P 3eξ
vector (1+ )
( 1 + e ) LkGA kGA 1+ e

Lockfree − M 0 / EI − PL /( 2 EI )
{ ε *} = { ε *} =
strain 0 P / kGA
vector

< ε ,vj >

{ε } = {ε }=
h m=2

j =1 < v , v >
{v j }, < v1 , v2 >= 0 {v1} =
0
1
, {v2 } =
2/ L
ξ
B
j j

{ε *}= {ε *}= < ε,v j >

m=2 0 1
h
{v * j }, v1 v2 {v1*} = , {v2 *} = B*
j =1 < v j , v j > 1 0
33
 TABLE 4
Error norm square for strain projections with the linear two noded Timoshenko beam
element. e=kGAL 2/(12EI)
1
2
q =
L
{q}T [D] {q} dξ {q} = {ε }− {ε }
2 −1

Locked Solution Lockfree

Case Solution
Cantilever
with tip 2 L 2M 02 e 2
q = . q* =0
2 EI 1 + e
moment, Mo
Cantilever
with tip q
2 L (P )2
= .
e
+
1
q* =
2 L
.
( PL )
2

transverse 2 2EI 1+ e 3 2 6EI

load P
{ε }= {ε }
h 2
q = ε −ε h
2
= ε
2
− εh ,
2

{ε }= {ε *}
h q*
2 2
= ε −ε h * = ε
2
− εh*
2
34
A case of variational incorrectness through reduced integration
A cantilever beam with uniformly distributed loading ρ
FI: Field inconsistent, Locked, but variationally correct FE results.
FC: Field consistent, Lock free, Reduced Integrated FE results.
Note that FC (by FEA) deviates from the field-consistent best-fit results.
For this case :
The extraneous force vector
(a non-zero vector) from
Reduced Integration consists
of self-equilibrating moments,
that shift the FC Best-fit from
the FC-FEM results.

{F }= −
e
E [[B ] − [B *]]T [D]{ε }dx
e

0
ρL2 / 12
{FEe } =
0
− ρL2 / 12

{ε *}= {ε *}+ δε *
h
35
 1.13 Lockfree an-isoparametric formulation
(quadratic transverse displacement and linear rotation)

3 _ 2 _
Displacements: w = h
N i wi h
θ = N iθ i (1.26)
i =1 i =1
1− ξ 1+ ξ
Shape functions N1 = N2 = N3 = 1 − ξ 2
2 2

Strain vector: {ε }
h
=
dθ dx
θ − dw / dx
= [B ] δ e { }
{ε }=
h 0 −1/ L 0
− (2ξ − 1) / L (1 − ξ ) / 2 − (2ξ + 1) / L (1 + ξ ) / 2
1/ L 0
4ξ / L
{δ } e
(1.27)

Standard basis vectors spanning 3-dimensional B space:

{v } = [0 , ξ ] , {v } = [1 , 0 ] and {v } = [0 , 1]
1
T
2
T
(1.28) 3
T

36
 Summary
Shear locking in Timoshenko’s Shear Flexible beam element occurs from spurious
constraints that arise from reducing the discretized domain into an Euler beam (of
infinite shear rigidity).

Shear locking is displayed through slow convergence, Spurious bending stiffening

and shear oscillations.

The field consistency paradigm identifies the spurious constraints related to locking,
and suggests methods to eliminate field inconsistency by eliminating the spurious
constraints (thereby enforcing field consistency).

Reduced integration (RI) eliminates shear locking by eliminating the spurious

constraint in the strain.

The function space approach shows that locked strain vector in an element (through
FEA) is actually the orthogonal projection of the analytical strain vector onto a field-
inconsistent subspace B, arising from a field-inconsistent [B] matrix (strain-
displacement matrix). B cannot be spanned by standard orthogonal basis vectors.

FEA through reduced integration (RI) effectively projects the analytical strain vector
onto a field-consistent subspace B*. However, RI is variationally incorrect in general,
and the FE strain vector agrees with the orthogonal projection on B* only when the
spurious extraneous force vector vanishes. 37
 Lecture 4
Special Topics of FEA

Chapter 2

Error Analysis in
Computational Elastodynamics

A comedy of errors…

38
 2.1 Finite Element Elastodynamic Equations using
the Principle of Least Action
2
Action I = L(q,q, t )dt Lagrangian L = T −V
1
Hamilton’s Principle δI = 0 for q(t), q(t1)= q(t2)=0

Lagrange’s Equation for motion

d ∂L ∂L
− = Qi
dt ∂q i ∂qi

Qi = Non − conservative generalised force

39
 In elastodynamics, the equations of motion are generally derived
in a global sense (with element assembly)
N N N N N N
e e e 1 e T
L = T −V = T − U − W = .{δ } [ M e ]{δ e } − e T e e
{δ } [ K ]{δ } − {δ e }T {F e }
e =1 e=1 e =1 e =1 2 e =1 e =1

Element Stiffness Matrix: [ K e ] = [ B ]T [ D][ B]dV

e

Element Consistent Mass Matrix: [ M e ] = [ N ]T [ ρ ][ N ]dV

e
Element Generalized Force Vector
(time dependent) : {F e } = [ N ]T { f (t )}dV
e
With element assembly, we get the global form
1 1
L = T − V = {δ G }T [ M G ]{δ G } − {δ G }T [ K G ]{δ G } − {δ G }T {F G }
2 2

d ∂L ∂L
− =0
Equation of motion dt ∂q i ∂qi

[ M G ]{δ G } + [ K G ]{δ G } = {F G } (2.1)

40
 2.2 Free Vibration Analysis

[ M G ]{δ G } + [ K G ]{δ G } = {0} (2.2)

Let {δ }= {φ}.sin(ω t )
G
n (2.3)

{[K G
}
] − ωn 2 [ M G ] {φ } = 0 (2.4)

{
det [ K G ] − ωn 2 [ M G ] = 0 }
Eigenvalue ωi 2 , Eigenmode {φi },
ωi is natural circular frequency (rad / sec)

41
Orthogonality of the Eigen-modes (Normal modes)

{φi }T [ K G ]{φ j } = 0 i ≠ j, {φi }T [ K G ]{φi } = kii

(2.5)
{φi }T [M G ]{φi } = 0 i ≠ j, {φi }T [M G ]{φi } = mii
Natural Frequencies (rad/sec)
kii
ωi =
mii
kii : generalized modal stiffness for mode i
mii:: generalized modal mass for mode i

qi(t)
kii mii
/\/\/\/\/\ Qi(t)

42
Example 1. Free vibration analysis of a simple cantilever beam using 10 Euler beam elements.
L=1m, b=0.1m, t=0.001m I=2.5×10-7m4, A=3×10-4m2
Density ρ=2722.77 kg/m3, Mass per unit length of the beam is ρA=0.816 kg/m
E=7.1 ×1010 N/m2,

Table 3.1Comparison of the natural frequencies in bending of the uniform cantilever beam obtained by different methods

Natural
Natural Differen Natural
Natural Natural Circular
Circular Natural t Frequency
Different Frequenc Different Circular Frequency
Frequency Frequency methods fn (Hz)
methods y methods Frequency n (rad/sec)
fn (Hz) n fn (Hz)
(rad/sec) n (rad/sec)

Classical
Classical 1447.51 9.095×103
Classical 516.935 3.248×103 solution
82.4915 518.31 solution
solution
9.097×103
FE result 82.4836 518.26 FE result 516.935 3.248×103 FE result 1447.83
43
Example 2. Free vibration analysis of an aircraft wing using Euler beam elements.

7.105 m Leading edge

2.1469m
Trailing edge
Root 0.849 m Tip
Fig (a) A typical subsonic aircraft wing

Fig (b) Length of the wing

divided in 22 elements for
FE formulation

(a) (b)

(c) (d)

Fig 4.3 Normal mode shapes of the wing of the aircraft

(a) First bending mode in y direction (frequency 7.2165 Hz)
(b) Second bending mode in y direction (frequency 21.138 Hz)
(c) Third bending mode in y direction (frequency 50.405 Hz) 44
(d) First torsional mode (frequency 56.8296 Hz)
Example 3. Dynamic Characterization of an aircraft using a Stick Model
Stick
StickModel
ModelofofSARAS
SARASfor
forDynamic
DynamicCharacterization
Characterization
•Beam Model with provision for Bending-Torsion Coupling (Shear Center offset).
•Results for components from in-house code benchmarked with those from detailed FE
model in NASTRAN.

Wing First Symmetric Mode 6.71 Hz (by Stick Model) and 6.72 Hz (by detailed FE model in NASTRAN)

Wing Second Symmetric Mode 18.89 Hz (by Stick Model) and 19.51 Hz (by detailed FE model in NASTRAN)

45
HT Anti-Symmetric Mode 10.4 Hz (by Stick Model) and 9.1 Hz (by detailed FE model in NASTRAN)
 2.3 Definitions of Inner Products in Elastodynamics

[D]= element elastic rigidity matrix

[ ρ ]= element inertia density matrix.
Stiffness-inner product Ne
< a, b >= {a}T [ D ]{b}dx (2.6)
ele =1 ele

Stiffness-norm squared value of the vector {a} is given as

2
a =< a, a > (2.7)

Ne
Inertia-inner product (c, d ) = {c}T [ ρ ]{d }dx (2.8)
ele =1 ele

Inertia-norm squared value of the vector {c} is given as

2 (2.9)
c = ( c, c )
46
 2.4 The Rayleigh Quotient
Free vibration of a system in a given mode can be expressed as
{U ( x, t )} = {u ( x)}eiωt (2.10)
Rayleigh Quotient from exact solutions for displacement and strain
modes u and 2
ε (2.11)
ω =
2
2
u
Let u h and ε h be the approximate modal vector and the strain
vector. Rayleigh Quotient from FEA solution
2
εh (2.12)
(ω h ) 2 = 2
uh

But interestingly (!) for a variationally correct solution,

2 < ε h ,ε >
ω = h
(2.13)
(u , u ) 47
 2.5 The Error Statements of Elastodynamics
Combining equations (5.6) and (5.7), we get one rule
2 h 2 2
ε −ε = ω 2 u − (ω h ) 2 u h
(2.14)
Error of global strain energy = Error of global kinetic energy

Combining equations (2.12) and (2.13), we get another rule,

valid for variationally correct solutions only,

< ε h , ε − ε h >= (u h , ω 2u − (ω h ) 2 u h ) (2.15)

Observation: The Errors in Elastodynamics are decided by

both displacements and strains.

48
 2.6 The Frequency-Error Hyperboloid

It can be shown that for u −u h2 h 2 ε −ε h 2

(ω )
variationally correct + − =1
h2 ω 2 h2
formulations, u ω2 u

Z = ε −ε h (2.16)
H

ωh
X = u −u h 1
E Y =
A
(ω = ω ) ω

Fig 1. Geometric interpretation of eigenvalue analysis of the variationally

correct formulation using Frequency-Error Hyperboloid. Approximate
eigenvalues obtained form a variationally correct formulation lie in the shaded 49
portion of the Hyperboloid.
 2.7 Why is the approximate Rayleigh Quotient higher
than the Exact one ?
(Valid only for variationally correct formulations)
h
u ε h

θu u θε
ε
(u , u h ) < ε ,ε h > cos θ u (u , u h ) ε ε
h
1 ε ε
h
cos(θ u ) = , cos(θε ) = = =
h
uu ε εh cos θε uu h < ε ,ε h > ω 2
u uh

2
h 2
2
cos θ u 1 ε ε 1 2 h 2 (ω h ) 2
= = ω (ω ) =
Hence for variationally cos 2 θε ω 4 u 2 u h 2 ω 4 ω2
correct formulations (2.17)
• Geometrically, the modal displacement vector suffers less deviation
than that of the modal strain vector. Hence (ω h ) 2
2
>1
ω 50
 Example 4: Free Vibration of a Simply Supported Beam
x L

Analytical Approximate
Modal Disp. w = a sin (πx / L ) wh = b
x
1−
x
L L

Modal Strain π
2
d 2 wh 2b
ε = (− d )= a sin (πx / L )
h
2
w / dx 2
ε = − =
L dx 2 L2

Eigenvalue
ω 2 = π 4 EI /( ρAL4 ) (ω h ) 2 = 120 * EI /( ρAL4 )

2 h 2 2 2 h 2 h EI π4
ε −ε = ω u − (ω ) u = a 2 − 4b 2
L3 2

EI
< ε h , ε − ε h >= (u h , ω 2u − (ω h ) 2 u h ) = 4 (πab − b 2 )
L3 51
 2.8 Replacement of Consistent Mass by Lumped Mass;
A variational crime L

Example 5. Free Vibration Analysis

of a Fixed-Fixed Bar
αL (1-α)L
using 2 elements
(First Mode Only shown)
36
33
30
27
24 consistent mass
eigen value

21
18
15
12 A B
9
6
exact solution
3 lumped mass
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
element length ratio

Any variationally incorrect formulation (with Lumped Mass, Reduced Integration

etc.) that does not conform exactly to the Weak form is variationally incorrect.
Variationally incorrect formualtions - Do not satisfy the Hyperboloid Rule
- Cannot guarantee and upper bound of the frequency.
52
 Lecture 4
Special Topics of FEA

Chapter 3

Rank Deficiency in elements

53
 3.1 What is rank deficiency ?
The rank of the stiffness matrix is the dimension of the B subspace
that emerges from the strain-displacement matrix [B], i.e.

Rank [Ke]= dim (B) (3.1)

In the dimension of the B subspace is given by
dim (B)=N-R (3.2)
N= Number of degrees of freedom of the element
R= Number of rigid body motions
To eliminate locking, a reduced order integration effectively converts
the Field-inconsistent [B] matrix into a Field-consistent [B*] matrix, by
simply removing the highest Legendre Polynomial in the field-
inconsistent spurious term of [B].

Using Gram Schmidt algorithm for orthogonal basis vector spanning B*

it can be shown that for some elements

dim (B*) < dim (B) or (N-R*) < (N-R), i.e. R* > R (3.3)

Rank [Ke*] < Rank [Ke] because of introduction of spurious rigid body motions
Reduced integration may introduce rank deficiency 54
 Rank deficiency of the plane stress Quad 4 element
1,2,3 are rigid body modes
4,5,6 are constant strain modes
7,8 are bending modes, but cannot
be sensed by a 1x1 reduced integration

1x1 reduced integration (with sampling

point at element center of zero strain)
effectively considers modes 7 and 8 as
zero energy hour-glass modes (spurious
rigid body motions)
Rank deficiency of this plane stress Quad 4 element is thus 2.

55
56
57
The Mindlin Plate element

58
 Appendix

The basis vectors for spanning

the B subspaces of the elements

59
60
61
62
63
 Some Thoughts
A burning question:
Does Mesh Optimisation Maximize Numerical Entropy?

Total Strain
Analytical Strain Energy with the analytical solution u
Energy remains Invariant (Maximum entropy)

Optimized mesh corresponds to maximum FEA Strain Energy (with

highest possible entropy with the approximations made).
A
FEA Strain Energy with the approximate solution uh depends on
meshing (the position of the middle node). Lower entropy than at A.

2 2
ε ≥ εh

Position of
Middle Node.

64
 Cui bono ? (For whose good ?)
How the best-fit paradigm helps

(a) Gives the exact, but hidden, mechanism of the way the Finite
Element Method works. It shows that computations in FEM
are actually determined in a best-fit manner of the strains (and
stresses), instead of the existing myth that they are based on
displacements.

(b) Helps one to make a priori error estimates for bench mark
problems easily.

( c) Helps one to evaluate the quality of the element that he/she

develops. The origins of the pathological problems of elements
can now be understood, diagnosed and eliminated by appropriate
methods.

65
 When Arts and Science met at the crossroads…

An extract from “ Sanchaita ” by Rabindranath Tagore.

A
A−A
900
v

u P
A
Subspace

66
 Bibliography

Variational Principles of Classical & Computational Solid Mechanics

1 J R Taylor. Classical Mechanics. University Science Books.

2 L Meirovitch. Analytical Mechanics. Dover.
3 C Lanczos. The Variational Principles of Mechanics. Dover.
4 C L Dyms, I H Shames. Solid Mechanics- A Variational Approach. McGraw
Hill.
5 T H Richards. Energy Methods in Stress Analysis. Ellis Horwood.
6 K Washizu. Variational Methods in Elasticity and Plasticity. Pergamon Press.
7 S Rajashekaran, G Sankarasubramanian. Computational Structural
Mechanics. Prentice Hall of India.
8 R Solecki, J R Conant. Advanced Mechanics of Materials. Oxford University
Press.

Linear Algebra

9 L H Edwards, D E Penny. Elementary Linear Algebra. Prentice Hall.

10 G Strang. Linear Algebra. Thomson Brooks/Cole.

67
 Bibliography (Finite Element Method)
Elementary FEM

1. T R Chandrapatla and A D Belegundu. Finite Elements in Engineering. Eastern

Economy Edition.

2. R D Cook, D S Malkus, M E Plesha. Concepts and Applications of Finite Element

Analysis. John Wiley & Sons.

3. J N Reddy. An introduction to the Finite Element Method. McGraw Hill.

4. K H Huebner, D L Dewherst, D E Smith, T G Byrom. The Finite Element Method for

engineers. John Wiley & sons.

5. O C Zienkiewicz. The Finite Element Method. McGraw Hill

Advanced FEM

6. G Strang, G J Fix. An analysis of the Finite element Method. Prentice Hall, NJ. .

7. G Prathap. The Finite Element Method in Structural Mechanics. Kluwer Academic

Press, Dordrecht.

8. T J R Hughes. The Finite Element Method. Dover.

9. O C Zienkiewicz, R L Taylor, J Z Hu. The Finite Element Method; Its basis and
fundamentals, 6-th edition. Butterworth Heinemann. 68
 Recent Publications from NAL

1. S. Mukherjee and G. Prathap 2001 17 (6), pp 385-393. Communications in Numerical

Methods in Engineering. Analysis of Shear Locking in Timoshenko beam elements using a
function space approach.
2. S. Mukherjee and G. Prathap 2002 Sadhana.27(5) 507-526. Analysis of delayed
convergence in the three noded isoparametric Timoshenko beam element using the function
space approach.
3. G. Prathap and S. Mukherjee 2003 Current Science, 85(17), pp 989-994. The engineer
grapples with Theorem 1.1 and Lemma 6.3 of Strang and Fix.
4. H. Mishra and S. Mukherjee 2004 Sadhana 29(6), pp 573-588. Examining the best-fit
paradigm in FEM at element level.
5. S. Mukherjee, P. Jafarali and G. Prathap 2005 Journal of Sound and Vibration. Vol
285(3), pp 615-635. A variational basis for error analysis in finite element elastodynamics.
6. K. Sangeeta , Somenath Mukherjee, and Gangan Prathap 2005 Structural Engineering
and Mechanics Vol 21(5), pp 539-551. A function space approach to study rank deficiency
and spurious modes in finite elements.
7. K. Sangeeta, Somenath Mukherjee and Gangan Prathap 2006 International Journal for
Computational Methods in Engineering Science & Mechanics, Vol 7, pages 1-12.
Conservation of the Best-Fit Paradigm at Element Level.
8. P. Jafarali, M Ammen, S. Mukherjee, G. Prathap, 2007 Journal of Sound and Vibration,
Vol 299(2), pp 196-211. Variational Correctness in Timoshenko beam finite element
elastodynamics.
9. S. Mukherjee, P. Jafarali, (Accepted in October 2008 by Communications in Numerical
Methods in Engineering; available on-line). Prathap’s best-fit paradigm and optimal strain
recovery points in indeterminate tapered bar analysis using linear element. 69
 Thank You
The Blind Men and the Elephant
And so these men of Indostan,
Disputed loud and long,
Each in his own opinion,
Exceeding stiff and strong,
Though each was partly in the right,
And all were in the wrong!
- John Godfrey Saxe (1816-1887)

70