Advanced Structural Analysis

Lecture 07 (example with MATLAB programming)

Date: 02 Nov (Thu)
Lecturer: Myung-Jin Choi ([email protected])

Example. Transient response of two degrees-of-freedom system (Bathe, 2006, Examples 9.1,
9.3, 9.6, 9.7). Solve the differential equation

MÜ(t) + Fint (t) = Fext , (1)

for a linear two degrees-of-freedom (DOFs) system, i.e., U = [U1 , U2 ]T , and

Fint (t) = KU(t), (2)

with      
2 0 6 −2 0
M= , K= , Fext = , (3)
0 1 −2 4 10
and the initial conditions

U(t) = 0, and U̇(t) = 0 at t = 0, (4)

using the following three methods.

(a) exact solution,
(b) central difference method,
(c) Newmark method (constant average acceleration method).

1
Solution
(a) We divide the solution procedure into the following three steps.
(Step 1: Change of the basis)
We can decouple the given differential equations for U(t), by mode superposition. We
introduce a new basis {ϕ1 , ϕ2 }, such that
 
  U1 (t)
U(t) = e1 e2
U2 (t)
 
  X1 (t)
= ϕ1 ϕ2 . (5)
X2 (t)

The selected new base vectors are eigenvectors, obtained from the generalized eigenvalue
problem
Kϕi = ωi2 Mϕi , (6)
with 
1 if i = j,
ϕT
i Mϕj = (7)
0 if i ̸= j,
for i, j ∈ {1, 2}. We obtain the eigenvalues ω12 = 2, ω22 = 5, and the corresponding eigenvec-
tors
 √  ( 1p )
1/√3 2/3
ϕ1 = , and ϕ2 = 2p . (8)
1/ 3 − 2/3
Let Φ := [ϕ1 , ϕ2 ], and X := [X1 , X2 ]T . Then Eq. (5) can be rewritten as

U(t) = ΦX(t). (9)

Note that Φ does not depend on time. Further, Eq. (7) can be rewritten as

ΦT MΦ = 12×2 , (10)

where 12×2 denotes the 2 × 2 identity matrix.

(Step 2: Decoupling of the given equation)

Substituting Eq. (9) into Eq. (1) gives

MΦẌ(t) + KΦX(t) = Fext . (11)

Multiplying ΦT to both sides of Eq. (11), and using Eq. (10), we have

Ẍ(t) + K̃X(t) = F̃ext , (12)

with
 
T 2 0
K̃ := Φ KΦ = , (13)
0 5
 √ 
10/p 3
F̃ext := ΦT Fext = . (14)
−10 2/3

2
Then, Eq. (12) can be rewritten as
10
Ẍ1 + 2X1 − √ = 0, (15a)
3
√
10 2
Ẍ2 + 5X2 − √ = 0, (15b)
3
with the initial conditions

X1 (0) = Ẋ1 (0) = 0, (16a)

X2 (0) = Ẋ2 (0) = 0. (16b)

(Step 3: Solution of the decoupled differential equation)

Let
Xi (t) = A cos (ωi t) + B sin (ωi t) + C, i ∈ {1, 2} . (17)
Then, using the conditions of Eqs. (15) and (16), we obtain
5  √ 
X1 (t) = √ 1 − cos 2t , (18a)
3
√ 
r 
2
X2 (t) = −2 1 − cos 5t . (18b)
3
Substituting Eq. (18) into Eq. (9), we obtain the final solution
5 √ 2 √
U1 (t) = − cos 2t + cos 5t + 1, (19a)
3 3
5 √ 4 √
U2 (t) = − cos 2t − cos 5t + 3. (19b)
3 3

3
(b) We obtain the solution at the discrete set of time steps, {t0 , t1 , · · · , tn , tn+1 , · · · , tN }, and
the solution at the current (n + 1)th step is obtained, based on the equilibrium condition of
Eq. (1) at the previous nth step, which is so called an explicit method. That is, we calculate
n+1
U such that
M n Ü + K n U = Fext , (20)
and
0
U = 0 U̇ = 0. (21)
Here, we consider a central difference method, where we employ

n 1 n+1
U − n−1 U


U̇ = (22a)
2∆t
n 1 n+1
U + n−1 U − 2 n U .


Ü = (22b)
∆t2
Substituting Eq. (22b) into Eq. (20) gives
1 1 2
2
M n+1 U = −n R − 2 M n−1 U + 2 M n U. (23)
∆t ∆t ∆t
The solution procedure is divided into the following two steps.
(Step 1: Initial calculation)
We first calculate the initial acceleration. Imposing the condition of Eq. (1) at time t = 0,
and applying the initial conditions of Eq. (4), we have
 
0
Ü = −M−1 K 0 U̇ − Fext
= M−1 Fext
  
1/2 0 0
=
0 1 10
 
0
= . (24)
10
−1
Second, we calculate U, as

−1 1
U = ∆t2 0 Ü + 
0 0
U̇ − ∆t U̇
2  
0
= ∆t2 (25)
5

(Step 2: Solution for each time step)

n+1
At every step n + 1 = 1, 2, 3, ..., we solve Eq. (23) to obtain U. If required, we calculate
n
U̇ and n Ü, using Eqs. (22a) and (22b), respectively.

4
(c) We obtain the solution at the discrete set of time steps, {t0 , t1 , · · · , tn , tn+1 , · · · , tN }, and
the solution at the current (n + 1)th step is obtained, based on the equilibrium condition
of Eq. (1) at the current n + 1th step, which is so called an implicit method. That is, we
calculate n+1 U such that
M n+1 Ü + K n+1 U = Fext , (26)
and
0
U = 0 U̇ = 0. (27)
Here, we consider a Newmark method, with the constant average acceleration method (trape-
zoidal rule), so that

n+1 1 n 
Ü + n+1 Ü ∆t,
U̇ = n U̇ + (28a)
2  
n+1 n n 1 1n 1 n+1
U = U + U̇∆t + Ü + Ü ∆t2
2 2 2
n 1 n n+1

= U+ U̇ + U̇ ∆t, (28b)
2
and combining Eqs. (28a) and (28b) gives

n+1 4 n+1 4 n
U − nU − U̇ − n Ü.


Ü = (28c)
∆t2 ∆t
Substituting Eqs. (28b) and (28c) into Eq. (26) gives
 
4 4 4
2
M + K n+1 U = Fext + 2 M n U + M n U̇ + M n Ü. (29)
∆t ∆t ∆t

The solution procedure is divided into the following two steps.

(Step 1: Initial calculation)
We first calculate the initial acceleration
 
0
Ü = −M−1 K 0 U̇ − Fext
= M−1 Fext
  
1/2 0 0
=
0 1 10
 
0
= . (30)
10

(Step 2: Solution for each time step)

n+1 n+1
At every step n + 1 = 1, 2, 3, ..., we solve Eq. (29) to obtain U. Then, we calculate Ü
and n+1 U̇, using Eqs. (28a) and (28b), respectively.

For the computation results, see the attached MATLAB code.

5
References
Bathe, K.-J. (2006). Finite element procedures. Klaus-Jurgen Bathe.