AERODYNAMICS AND AEROELASTICITY

Master in Mechanical Engineering

3. Galerkin Method

Pedro Leal Ribeiro

2022-2023
APPLIED MECHANICS SECTION
FACULTY OF ENGINEERING OF THE UNIVERSITY OF PORTO

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 1

 3.1 Method of Weighted Residuals
• In the method of Weighted Residuals, as occurred in Rayleigh-Ritz method,
a form is assumed to the solution.
• This assumed solution must have unknown parameters, which are
determined so that some error measured is minimised.

• The method of Weighted Residuals works directly with the partial differential
equation of motion and boundary conditions.
• By the way, Rayleigh-Ritz does not employ the partial differential equations of
motion, nor does it lead to ordinary differential equations of motion, it employs
Rayleigh’s quotient, hence the name. Rayleigh-Ritz method leads to an eigenvalue
problem. Be careful with some literature.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 2

• The equation of motion of unidimensional continuous systems,
experiencing forced vibrations, can often be written as
v 2
( )
f ( x, t ) − L k ( x ) L ( v ( x, t ) ) = m ( x ) 2 ( x, t )
t
x3  z
b l x2  y
where
L (...) - differential operator x1  x

k ( x ) - related to stiffness h

m ( x ) - related to mass
f ( x, t ) - force or moment per unit length
v ( x, t ) - displacement or angular displacement

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 3

 • For example, in flexural vibration of beams:
 2 (...)
L (...) = k ( x) = E ( x) I ( x) m ( x) =  ( x) A( x)
x 2
x3  z
b l x2  y
So
v 2
(
f ( x, t ) − L k ( x ) L ( v ( x, t ) ) ) = m ( x ) 2 ( x, t )
t
x1  x
h
becomes
  2
f ( x, t ) − 2  E ( x ) I ( x )
 2
( v ( x, t ) )   2v
 = m ( x ) 2 ( x, t )
x  x 2 
 t

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 4

• In the method of weighted residuals, one assumes a trial function F(x, t)
that, in general does not satisfy the equation of motion.

• A measure of the error is given by function R(F,x,t)

 2F
(
R ( F, x, t ) = f ( x, t ) − L k ( x ) L ( F ( x, t ) ) ) − m ( x ) 2 ( x, t )
t
known as the residual.

• R(F,x,t) varies along the domain, to obtain a value for the error over the entire
solution domain, one could consider integrating R(F,x,t) over x. However, in a
straight integration, the negative and positive errors at different points may
cancel each other. Furthermore, such an integral would provide one equation
only.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 5

• In an attempt to avoid error cancellation, R(F,x,t) is first multiplied by
a weighting function and then integrated over the solution domain.
• Multiplying the residual by one weighting function i(x), one obtains a
weighted residual
  2F 
( )
R ( F, x, t )  i ( x ) =  f ( x, t ) − L k ( x ) L ( F ( x, t ) ) − m ( x ) 2 ( x, t )   i ( x )
t
 
• At this stage, known functions of x have been used to express F(x,t), so
derivatives with respect to x can be made.
• Integrating over the domain we receive one ordinary differential equation
where the sole variable is time (t):
   2F 
 ( )
 f ( x, t ) − L k ( x ) L ( F ( x, t ) ) − m ( x )  2 ( x, t )    i ( x ) dx = 0
  t 

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 6

• The approximation or trial function can be written as
n
F ( x, t ) =  q j ( t )  j ( x )
j =1

• Since there are n unknown parameters, n weighting functions i(x) should

also be chosen, so that n weighted residuals exist.

• Integrating each weighted residual in the domain:

   n   n d2q j (t ) 
  f ( x, t ) − L  k ( x ) L   q j ( t )  j ( x )   − m ( x )    j ( x )    i ( x ) dx = 0
   dt 2 
  j =1   j =1  
where qj(t) are unknown coefficients.
• Since j=1..n, a system of n ordinary differential equations on qj(t) is obtained.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 7

➢ Approximation functions j(x) are chosen by the user and should (references in the
last slide; most of these rules appeared in Rayleigh-Ritz method)
n
• be linearly independent, that is    ( x) = 0  
j =1
j j j = 0, j = 1, 2, , n;
• have continuous derivatives up to, at least, the order of the problem - order of the
highest derivative with respect to x in the equation of motion -minus the number of
integrations by parts performed;
• respect all boundary conditions, but if integration by parts is applied it is enough that
the geometric - also known as essential or kinematic - boundary conditions are
respected;

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 8

➢ [continuation] Approximation functions j(x) are chosen by the user and should
(references in the last slide; most of these rules appeared in Rayleigh-Ritz method)

• form a complete series; here, a series of functions is understood to be

complete if the mean square error tends to zero as one increases the number
of functions; with v(x,t) representing the exact solution, we will express this
condition as: 2
 n 
lim   v ( x, t ) −   j ( x ) a j ( t )  dx = 0
n→
 j =1 
at all instants t; most often, in practice, time and space are dealt with separately.
The set of approximation functions must allow for the representation of all terms in
an exact solution. The solution will not converge if a term that is present in the
exact solution is not included in the assumed solution (Bhatti, 2005). Suppose for
example that monomial x2 is present in the exact solution, but the assumed one
only contains odd powers of x, in this case, convergence is not achieved.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 9

 3.2 Galerkin’s method
• Galerkin’s method is a method of Weighted Residuals where the weighting
functions coincide with the trial or approximation functions

   n   n d2q j (t ) 
  f ( x, t ) − L  k ( x ) L   q j ( t )  j ( x )   − m ( x )    j ( x )   i ( x ) dx = 0
   dt 2 
  j =1   i =1  
j=1..n

  n   n d2q j (t ) 
 L  k ( x ) L   q j ( t )  j ( x )   i ( x ) dx +  m ( x )    j ( x ) i ( x ) dx =  f ( x, t )i ( x ) dx
   2
  j =1   i =1 dt 

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 10

• Introducing
(
kij =  i ( x ) L k ( x ) L ( j ( x ) ) dx) mij =  m ( x )i ( x ) j ( x )dx

Qi ( t ) =  i ( x ) f ( x, t ) dx

• The following set of ordinary differential equations in the time domain is

obtained
n n

 m q (t ) +  k q (t ) = Q (t )
i =1
ij j
i =1
ij j i j=1..n

In matrix form
 M q ( t ) +  K q ( t ) = Q ( t )

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 11

 3.3 Eigenvalue problem
3.3.1 Weighted residuals
• In free vibrations, the equation of motion becomes
v 2
( )
− L k ( x ) L ( v ( x, t ) ) = m ( x ) 2 ( x, t )
t
• Inserting v(x,t)=F (x)• cos(wt-), with in the equation of motion one obtains
the characteristic equation in differential form
( )
L k ( x ) L ( F ( x )) = w 2m ( x ) F ( x )

• The approximation or trial function is now written as

n
F ( x ) =  q j j ( x )
j =1

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 12

• Defining, as before, a weighted residual and integrating in the domain:
   n   n 2 
  − L  k ( x ) L   q j j ( x )   + m ( x )   w q j j ( x )    i ( x ) dx = 0
     
  j =1  j =1 

   n   n 2 
  i ( x )  L  k ( x ) L   q j j ( x )   dx =   i ( x ) m ( x )   w q j j ( x ) dx


 j =1 
  j =1 

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 13

3.3.2 Galerkin’s method
• As said, in Galerkin’s method the approximation functions are also the
weighting functions. Hence, one has:
   n   n 2 
 i ( x )  L  k ( x ) L   q j j ( x )   dx =  i ( x ) m ( x )   w q j j ( x ) dx


 j =1 
  j =1 
leading to n n

 ij j
k
i =1
q = w 2
 mij q j
i =1
j=1..n

  
K q = w 2
 M q
with the coefficients given in slide 11.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 14

3.4 Application to beam bending
• The bending of Bernoulli-Euler beams is governed by

 2 
f ( x, t ) − 2   E ( x ) I ( x )
 2
( v ( x, t ) )    2v
  =  ( x ) A ( x ) 2 ( x, t )
x   x 2 
 t


so that the operators given in slide 4 apply:

 2 (...)
k ( x) = E ( x) I ( x) m ( x) =  ( x) A( x) L (...) =
x 2

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 15

• Consider now, for the sake of simplicity that the material is homogeneous
and the cross section of the beam is constant, then
E ( x) = E I ( x) = I A( x) = A  ( x) = 

• The partial differential equation of motion becomes

 4 v ( x, t ) v 2
f ( x, t ) − EI =  A 2 ( x, t )
x 4
t
• Make the following approximation n
v ( x, t )    j ( x ) q j ( t )
j =1

4  n  2  n 
f ( x, t ) − EI 4   q j (t ) j ( x )  =  A 2   q j (t ) j ( x ) 
x  j =1  t  j =1 

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 16

  n d 4 j ( x )   n d 2 q j (t ) 
f ( x, t ) − EI   q j (t )  =  A    j ( x ) 
 j =1 dx 4   j =1 dt 2
  
• The coefficients of the mass and stiffness matrix in the equations of
motion
 M q ( t ) +  K q ( t ) = Q ( t )
are
d 4 j ( x )
kij =  EIi ( x ) 4
dx in general, kij kji
0 dx

mij =   Ai ( x )  j ( x ) dx , mij= mji

0

• The coefficients of generalised external forces are

Qi ( t ) =  i ( x ) f ( x, t ) dx

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 17

• One may decide to carry out integrations by parts in kij :

d 4 j ( x )  d 3 j ( x )  di ( x ) d 3 j ( x )
kij =  EIi ( x ) dx  kij = EI  i ( x )  − 0 EI dx
dx 4  3
dx 0 dx dx 3

0

 d 3 j ( x )   di ( x ) d 2 j ( x )  d 2i ( x ) d 2 j ( x )
 kij = EI  i ( x )  − EI   + 0 EI dx
 3
dx 0 2 2 2
  dx dx 0 dx dx

• In many cases, the boundary conditions allow one to eliminate the

first terms in kij, so that we arrive at
d 2i ( x ) d 2 j ( x )
 kij =  EI 2 2
dx with kij= kji
0 dx dx

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 18

• The coefficients kij, of Galerkin’s method after integration by parts,
coincide with the ones of Rayleigh-Ritz method.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 19

About Galerkin and the Rayleigh-Ritz methods
▪ Both methods are powerful tools for finding approximate solutions for problems
in dynamics. Rayleigh-Ritz is in fact a method derived to solve eigenvalue
problems, leading, in our case, to natural frequencies and natural mode shapes of
vibration. A method in some way similar to Rayleigh-Ritz can be applied to forced
vibration problems, but one should use another designation for this method.
▪ Rayleigh-Ritz method requires that the kinetic and potential energies are known;
its extension to forced vibrations involves the virtual works of “non-conservative”
forces.
▪ Provided the differential equations of motion are available, Galerkin’s method can
be applied both to free or forced vibrations and does not require that the
energies are known; hence it is more general than Rayleigh-Ritz.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 20

 3.5 Example – Ordinary differential equations of
motion, forced vibration of a cantilever beam
l
For the cantilever beam here represented, obtain
ordinary differential equations of motion in the time z w(x,t)
domain applying Galerkin’s method. Employ the two x
following functions, which respect the geometric
(essential or kinematic) boundary conditions.
f(x,t)
2
 x
2
x x
1 ( x) = 1 −  2 ( x) = 1 −  Cantilever beam
   
1 ( x) 2 ( x)

0 1 0 x 1
x

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 21

 2 2
 x x x
w ( x, t )  F ( x, t ) = q1 ( t ) 1 −  + q2 ( t ) 1 − 
   
The stiffness coefficients that result from integration by parts are

 d 3 j ( x )   di ( x ) d 2 j ( x )  d 2i ( x ) d 2 j ( x )
kij = EI  i ( x )  − EI   +  EI dx
 dx 0
3  dx dx 0 0
2
dx 2
dx 2
 
At x=0, the boundary conditions are:
 3 w ( 0, t )  2 w ( 0, t )
Q ( 0, t ) = EI =0 M ( 0, t ) = EI =0
x 3
x 2

At x=l:
w
w( , t ) = 0 ( ,t) = 0
x

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 22

Imposing the boundary conditions, the stiffness coefficients become
d 2i ( x ) d 2 j ( x )
kij =  EI 2 2
dx
dx0 dx
Hence, the determination of kij proceeds as in the Rayleigh-Ritz example
d 21 ( x ) 2 d 22 ( x ) 2 ( 2 − 3x )
2
= 2 2
=− 3
dx dx
d 21 ( x ) d 21 ( x ) d 21 ( x ) d 22 ( x )
k11 = EI  dx = ? k12 = k21 = EI  2 2
dx = ?
0 dx 2
dx 2 0 dx dx

d 22 ( x ) d 22 ( x )
k22 = EI  2 2
dx = ?
0 dx dx

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 23

 Imposing the boundary conditions, the stiffness coefficients become
d 2i ( x ) d 2 j ( x )
kij =  EI 2 2
dx
dx0 dx
Hence, the determination of kij proceeds as in the Rayleigh-Ritz example
d 21 ( x ) 2 d 22 ( x ) 2 ( 2 − 3x )
2
= 2 2
=− 3
dx dx
d 21 ( x ) d 21 ( x ) d 21 ( x ) d 22 ( x ) 2
k11 = EI  dx = EI
4 k12 = k21 = EI  2 2
dx = − EI 3
0 dx 2
dx 2 3 0 dx dx
 4 2
 3 − 3
d 22 ( x ) d 22 ( x )
k22 = EI  dx = EI
4
 K  = EI  2 
0 dx 2 dx 2 3
− 4 
 3 3 


2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 24

Had we used the stiffness coefficients prior to integration by parts
d 4 j ( x )
kij =  EIi ( x ) 4
dx
0 dx

we would have to find other approximation functions, because

d 41 ( x ) d 31 ( x ) d 42 ( x )
4
= 3
=0 4
=0
dx dx dx

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 25

• The coefficients of the mass matrix are (again, the same equation was used in
Rayleigh-Ritz)
mij =  A i ( x ) j ( x ) dx
0 1 1 
A A A 5 30 
m11 = m12 = m21 = m22 = M  =  A  1 1 
5 30 105  
 30 105 

• The vector of generalised forces is

2
 x
2
x x
Q1 ( t ) =  f ( x, t )  1 −  dx Q2 ( t ) =  f ( x, t )  1 −  dx
   
For an uniform force f ( x, t ) = f 0  g ( t )

f0  f0 
Q1 ( t ) =  g (t ) Q2 ( t ) =  g (t )
3 12

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 26

• Equations of motion if the force is uniform in space

1 1   4 2 1
− 3
5 30   q1 ( t )   3  q1 ( t )   3 
A    + EI    =    f0 g (t )
1 1  q2 ( t )  − 2 4   q2 ( t )   1 
 30 105   3 3 
 12 

• If g(t) is given, this equation can be solved by methods addressed in another

course. After computing q1(t) and q2(t) an approximation for the transverse
displacement w(x,t) is obtained
2 2
 x x x
w ( x, t )  F ( x, t ) = q1 ( t ) 1 −  + q2 ( t ) 1 − 
   

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 27

 Homework
• Verify if the following functions respect the
boundary conditions l

 − x   −x w(x,t)
1 ( x) = 1 − cos   2 ( x) = 1 − cos   
z
x
 2   

f(x,t)
• obtain the corresponding ordinary differential
equations of motion in the time domain applying Cantilever beam
Galerkin’s method.

2022-2023 AERODYNAMICS AND AEROELASTICITY – M.EM - PEDRO LEAL RIBEIRO 28

References
• L. Meirovitch; Computational methods in structural dynamics, Alphen aan den
Rijn, The Netherlands, Sijthoff & Noordhoff, 1980.
• M. Asghar Bhatti, Fundamental Finite Element Analysis and Applications, John
Wiley & Sons, 2005.
• Galerkin, B. G. “Sterzhni i plastinki: Riady v nekotorykh voprosakh uprugogo
ravnovesiia sterzhnei i plastinok.” Vestnik inzhenerov, 1915, vol. 1, no. 19, pp. 897-
908. [not verified, but it is quoted elsewhere as the original reference for
Galerkin’s method]

2022-2023 Aerodynamics and Aeroelasticity – M.EM - Pedro Leal Ribeiro 29