0% found this document useful (0 votes)
12 views

Lecture Note04

Uploaded by

liu377997
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
12 views

Lecture Note04

Uploaded by

liu377997
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 47

结构动力学及其工程应用

Structural Dynamics and Its Applications

Multi-Degree-of-Freedom (MDOF)
Systems

阳佳桦,Ph.D.,副教授
广西大学土木建筑工程学院,工程力学研究中心
Email: [email protected]
Dr. Jia-Hua Yang, Ph.D., Associate Professor
College of Civil Engineering and Architecture
Scientific Research Center of Engineering Mechanics
Guangxi University
Contents
 Harmonic Excitation

 Periodic Loading

 Unit Impulse

 General Excitation

2
Equation of Motion
 We use a 3-DOF system for illustrating the concepts. In practice,
finite element method is used for constructing the equation of
motion. Damping is not considered yet for simplicity, but we’ll return
to it later after introducing natural frequencies and mode shapes.

3
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Force equilibrium for each mass:
𝑓1 𝑡 − 𝑘1 𝑥1 + 𝑘2 𝑥2 − 𝑥1 = 𝑚1 𝑥ሷ 1
𝑓2 𝑡 − 𝑘2 𝑥2 − 𝑥1 + 𝑘3 𝑥3 − 𝑥2 = 𝑚2 𝑥ሷ 2
𝑓3 𝑡 − 𝑘3 𝑥3 − 𝑥2 = 𝑚3 𝑥ሷ 3
Re-arrange:
𝑚1 𝑥ሷ 1 + 𝑘1 + 𝑘2 𝑥1 − 𝑘2 𝑥2 = 𝑓1 𝑡
𝑚2 𝑥ሷ 2 − 𝑘2 𝑥1 + 𝑘2 + 𝑘3 𝑥2 − 𝑘3 𝑥3 = 𝑓2 𝑡
𝑚3 𝑥ሷ 3 − 𝑘3 𝑥2 + 𝑘3 𝑥3 = 𝑓3 𝑡
In matrix form
𝑚1 0 0 𝑥ሷ 1 𝑘1 + 𝑘2 −𝑘2 0 𝑥1 𝑓1 𝑡
0 𝑚2 0 𝑥ሷ 2 + −𝑘2 𝑘2 + 𝑘3 −𝑘3 𝑥2 = 𝑓2 𝑡
0 0 𝑚3 𝑥ሷ 3 0 −𝑘3 𝑘3 𝑥3 𝑓3 𝑡

4
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
𝐌
ณ 𝐗ሷ
ณ + 𝐊
ณ 𝐗
ณ = ณ
𝐅
mass acceleration stiffness displacement loading
matrix vector matrix vector
𝑥ሷ 1 𝑥1 𝑓1 𝑡
𝐗ሷ = 𝑥ሷ 2 and 𝐗 = 𝑥2 and 𝐅 = 𝑓2 𝑡
𝑥ሷ 3 𝑥3 𝑓3 𝑡
The complete equation of motion of a MDOF system is

 In applications of vibration and dynamics, 𝐂 is seldom constructed


from first principles, e.g., finite element method. Damping is
accounted empirically through ‘modal damping ratios’.

5
Au SK, 2018. Operational Modal Analysis. Modeling, Bayesian Inference, Uncertainty Laws. Springer.
Positive Definiteness and Energy
 𝐌, 𝐂, 𝐊 are real symmetric and positive definite:

Their entries are all real values;


symmetric: 𝐌 = 𝐌 𝑇 , 𝐂 = 𝐂 𝑇 , 𝐊 = 𝐊 𝑇
positive definite: …
The positive definite property:
1. 𝐌 reflects positive kinetic energy for non-zero velocity;
2. 𝐂 reflects energy dissipation rate for non-zero velocity;
3. 𝐊 reflects that the structure is stable, and has positive strain energy

6
Au SK, 2018. Operational Modal Analysis. Modeling, Bayesian Inference, Uncertainty Laws. Springer.
Natural Frequencies and Mode Shapes
 Undamped free vibration is studied first to introduce the
characteristics of MDOF systems. They have multiple natural
frequencies, each of which has a spatial vibration pattern called mode
shape.

We will follow an intuitive derivation (at least I think) to obtain its


solution. Suppose the solution contains two parts, a spatial and a time-
dependent part, which can be separated. We have

where is a constant vector and 𝜂(𝑡) is a scalar function of 𝑡.


This assumption may not be true, but we will see how to make
adjustments.
7
Au SK, 2018. Operational Modal Analysis. Modeling, Bayesian Inference, Uncertainty Laws. Springer.
Substituting gives

Facts:
1. This equation should hold for all 𝑡.
𝐊𝛙 and 𝐌𝛙 do not depend on 𝑡, so must be a constant.
2. Pre-multiplying 𝛙𝑇 gives

positive definite positive definite

So we let be a negative constant. 𝜔 is taken to be a real


constant, because the negative 𝜔 gives the same solution and we ignore
it.
8
Au SK, 2018. Operational Modal Analysis. Modeling, Bayesian Inference, Uncertainty Laws. Springer.
Now we have two equations:

The first one describes the undamped free vibration of a SDOF system,
and the second is the generalized eigenvalue problem.
The eigenvalue problem looks for the eigenvalue 𝜔2 (square of the
natural frequency) and the eigenvector (mode shape). It can be shown
that there are n different possible solutions , and
correspondingly n ‘modal responses’ , which are n sets of
independent modal coordinates. The response can be represented by the
set of modal responses and mode shapes, modal superposition

9
Au SK, 2018. Operational Modal Analysis. Modeling, Bayesian Inference, Uncertainty Laws. Springer.
Eigenvalue Problem
The eigenvalue problem can be written as
𝐊 − 𝜔2 𝐌 𝛟 = 𝟎
It is a set of 𝑁 homogeneous linear equations. To have non-trivial
solution requires the determinant of the coefficient matrix to be equal to
zero
𝐊 − 𝜔2 𝐌 = 0
Expanding the determinant gives a 𝑁th order polynomial of 𝜔2 , so there
are 𝑁 roots and the eigenvalue problem becomes
𝐊 − 𝜔𝑖2 𝐌 𝛟𝑖 = 𝟎 for 𝑖 = 1, … , 𝑁
Therefore, there are 𝑁 modes with one natural frequency 𝜔𝑖 and one
mode vector of amplitude (to be determined) 𝛟𝑖 (mode shape) for each
10
mode.
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Notes for mode shapes.
the exact vibration amplitude at each DOF cannot be determined but only
the “relative” amplitudes. That is if 𝛟𝑖 satisfies the set of 𝑁
homogeneous linear equations, 𝑎𝛟𝑖 also satisfies, where 𝑎 is a constant
(it can be positive or negative). Thus, 𝛟𝑖 only gives the “shape” of
vibration amplitudes of the multi-DOF system, and it is called the mode
shapes of the 𝑖 th mode of vibration.
The solution of all natural frequencies and mode shapes can be expressed
𝜔12
𝜔22
𝛀=

2
𝜔𝑁
𝜙11 𝜙12 ⋯ 𝜙1𝑁
𝜙 𝜙22 ⋯ 𝜙2𝑁
𝚽 = 𝛟1 , 𝛟2 , ⋯ , 𝛟𝑁 = 21
⋮ ⋮ ⋱ ⋮
𝜙𝑁1 𝜙𝑁2 ⋯ 𝜙𝑁𝑁
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Notes on the modal parameters.

The mode corresponding to the lowest natural frequency is called the


fundamental mode, which is of particular importance.

It is believed that the fundamental mode (or the first mode) of a


structural system is the most easily excited one. It is generally the
easiest one to be measured in a field test (or vibration test under
laboratory conditions).

Other modes are usually called the higher modes (or higher
harmonics). In general, the mode shapes for lower modes are
relatively simple (without fluctuations in space). However, the mode
shapes for higher modes become more and more complicated (with
more and more fluctuations in space)

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Example. Determine the natural frequencies and mode shapes of a two-
story shear building model.
𝑘1 + 𝑘2 −𝑘2 9.953 −6.578
𝐊= = ×106 N/m
−𝑘2 𝑘2 −6.578 6.578
𝑚1 0 2700 0
𝐌= = kg
0 𝑚2 0 5400
𝑘 + 𝑘 − 𝜔 2𝑚 −𝑘2 𝜙1
2 1 2 𝑛 1 0
𝐊 − 𝜔𝑛 𝐌 𝛟 = 2 =
−𝑘2 𝑘2 − 𝜔𝑛 𝑚2 𝜙 2 0
𝑘1 + 𝑘2 − 𝜔𝑛2 𝑚1 𝜙1 − 𝑘2 𝜙2 = 0
⟹൝
−𝑘2 𝜙1 + 𝑘2 − 𝜔𝑛2 𝑚2 𝜙2 = 0
𝑘1 + 𝑘2 − 𝜔𝑛2 𝑚1 −𝑘2
2 =0
−𝑘2 𝑘2 − 𝜔𝑛 𝑚2

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
2 2
𝑚1 𝑚2 𝜔𝑛 − 𝑘1 + 𝑘2 𝑚2 + 𝑚1 𝑘2 𝜔𝑛2 + 𝑘1 𝑘2 = 0
6 2 2
1.4580 × 10 𝜔𝑛 − 7.1507 × 1010 𝜔𝑛2 + 2.2201 × 1013 = 0
𝜔12 = 333.0930 and 𝜔22 = 4571.3514
The natural circular frequencies (in rad/s) of the two-story shear building are:
𝜔1 = 18.2508 rad/s and 𝜔2 = 67.6118 rad/s
 Calculate mode shapes. First mode:
2
2 𝑘1 + 𝑘 2 − 𝜔1 𝑚1 −𝑘2 𝜙11 0
𝐊 − 𝜔1 𝐌 𝛟1 = 2 =
−𝑘2 𝑘2 − 𝜔1 𝑚2 𝜙21 0
9.0536𝜙11 − 6.5780𝜙21 = 0
⟹ቊ
−6.5780𝜙11 + 4.7793𝜙21 = 0
Since this set of linear equations is homogeneous, there is only 𝑁 − 1
independent equations. Both of the above equations can be transformed to the
same one:

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
𝜙11 − 0.7266𝜙21 = 0
It is clear that the exact values of 𝜙11 and 𝜙21 cannot be obtained, but
only the relative values. If it is assumed that 𝜙21 = 1, then 𝜙11 =
− 0.7266. Thus, the first mode shape
𝜙11 0.7266
𝛟1 = =
𝜙21 1
The frequency of the 2nd mode is substituted into the eigenvalue problem
to obtain the mode shape of the 2nd mode:
2
2 𝑘1 + 𝑘 2 − 𝜔2 𝑚1 −𝑘2 𝜙12 0
𝐊 − 𝜔2 𝐌 𝛟2 = =
−𝑘2 𝑘2 − 𝜔22 𝑚2 𝜙22 0
0.2390𝜙12 + 0.6578𝜙22 = 0
⟹ቊ
0.6578𝜙12 + 1.8107𝜙22 = 0
These two equations comes down to the same one:
𝜙12 + 2.7527𝜙22 = 0
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
If it is assumed that 𝜙22 = 1, then 𝜙12 = −2.7527
𝜙12 −2.7527
𝛟2 = =
𝜙22 1
Finally, the mode shape matrix
0.7266 −2.7527
𝚽 = 𝛟1 , 𝛟2 =
1 1
It is clear that the vibration of the two DOFs are in-phase in mode 1. That
is the vibrations are in the same direction with zero-degree phase angle.
However, the two DOFs vibrates out-of-phase (i.e., with 180° phase
angle) in mode 2.

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Example. Determine the natural frequencies and mode shapes of a
three-story shear building with 𝑚1 = 𝑚2 = 𝑚3 = 1 kg and 𝑘1 = 𝑘2 =
𝑘3 = 10 N/m.
𝑚1 0 0 1 0 0
𝐌 = 0 𝑚2 0 = 0 1 0 kg
0 0 𝑚3 0 0 1
𝑘1 + 𝑘2 −𝑘2 0 20 −10 0
𝐊 = −𝑘2 𝑘2 + 𝑘3 −𝑘3 = −10 20 −10 N/m
0 −𝑘3 𝑘3 0 −10 10
𝐊 − 𝜔𝑛2 𝐌 𝛟 =
𝑘1 + 𝑘2 − 𝜔𝑛2 𝑚1 −𝑘2 𝜙1 0
−𝑘2 𝑘2 + 𝑘3 − 𝜔𝑛2 𝑚2 −𝑘3 𝜙2 = 0
−𝑘3 𝑘3 − 𝜔𝑛2 𝑚3 𝜙3 0

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 The determinant of the coefficient matrix should be zero to ensure non-
trivial solutions

𝑘1 + 𝑘2 − 𝜔𝑛2 𝑚1 −𝑘2
−𝑘2 𝑘2 + 𝑘3 − 𝜔𝑛2 𝑚2 −𝑘3 =0
−𝑘3 𝑘3 − 𝜔𝑛2 𝑚3
Expanding the determinant and re-arranging
𝑘1 + 𝑘2 − 𝜔𝑛2 𝑚1 𝑘2 + 𝑘3 − 𝜔𝑛2 𝑚2 𝑘3 − 𝜔𝑛2 𝑚3 − 𝑘32
+ 𝑘2 −𝑘2 𝑘3 − 𝜔𝑛2 𝑚3 = 0
2 3 2 2
𝑎 𝜔𝑛 +𝑏 𝜔𝑛 + 𝑐𝜔𝑛2 + 𝑑 = 0
𝑎 = −𝑚1 𝑚2 𝑚3
𝑏 = 𝑚2 𝑚3 𝑘1 + 𝑘2 + 𝑚1 𝑚2 𝑘2 + 𝑘3 + 𝑚1 𝑚3 𝑘3
𝑐 = 𝑘2 𝑘2 𝑚2 − 𝑘1 𝑚1 − 𝑘1 + 𝑘2 𝑘2 𝑚3 + 𝑘3 𝑚3 + 𝑘3 𝑚2
𝑑 = 𝑘1 𝑘2 𝑘3
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
The Cardano's Method (Brilliant.org., 2021) is employed. The first root
of the cubic equation 𝑎𝑥 3 + 𝑏𝑥 2 + 𝑐𝑥 + 𝑑 = 0 is
3 3 𝑏
𝑥1 = 𝛼+ 𝛼2 + 𝛽3 + 𝛼− 𝛼2 + 𝛽3 −
3𝑎
9𝑎𝑏𝑐 − 27𝑎2 𝑑 − 2𝑏 3
𝛼=
54𝑎3
3𝑎𝑐 − 𝑏 2
𝛽=
9𝑎2
After solving for the first root, 𝑥1 , the cubic equation is written as
𝑥 − 𝑥1 𝐴𝑥 2 + 𝐵𝑥 + 𝐶 = 0
Now if we can find 𝐴, 𝐵 and 𝐶, we can solve the quadratic equation
𝐴𝑥 2 + 𝐵𝑥 + 𝐶 = 0 to get the other two roots. To do this, expanding the
above equation gives

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Brilliant.org. (2021) Cardano's Method. Retrieved 19:50, April 5, 2021, from https://brilliant.org/wiki/cardano-method/.
𝐴𝑥 3 + 𝐵 − 𝑥1 𝐴 𝑥 2 + 𝐶 − 𝑥1 𝐵 𝑥 − 𝑥1 𝐶 = 0
Compare with the original equation
𝑎𝑥 3 + 𝑏𝑥 2 + 𝑐𝑥 + 𝑑 = 0
With the coefficient relationship of the 3 arrows, we obtain the
coefficients of the quadratic equation
𝐴𝑥 2 + 𝐵𝑥 + 𝐶 = 0
𝐴 = 𝑎, 𝐵 = 𝑏 + 𝑥1 𝐴 and 𝐶 = 𝑐 + 𝑥1 𝐵 = − 𝑑Τ𝑥1
Now we can solve the quadratic equation to get another two roots.
Summary
1. Calculate 𝛼 = 129.6296 and 𝛽 = −77.7778
2. Caltulate the first root 𝑥1 = 32.4698
3. Calculate 𝐴 = −1.0000, 𝐵 = 17.5302 and 𝐶 = −30.7979
4. Solve to get the other two roots 𝑥2 = 1.9806, 𝑥3 = 15.5496

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Brilliant.org. (2021) Cardano's Method. Retrieved 19:50, April 5, 2021, from https://brilliant.org/wiki/cardano-method/.
The three eigenvalues (roots sorted in ascending order) are:
𝜔12 = 1.9806, 𝜔22 = 15.5496, 𝜔32 = 32.4698
𝑓1 = 0.2240 Hz, 𝑓2 = 0.6276 Hz, 𝑓3 = 0.9069 Hz
Next, the eigenvalues are substituted back into the eigenvalue problem,
respectively, for calculating the corresponding mode shape:
18.0194 −10.0000 0 𝜙11 0
𝐊 − 𝜔12 𝐌 𝛟1 = −10.0000 18.0194 −10.0000 𝜙21 = 0
0 −10.0000 8.0194 𝜙31 0
Gauss elimination gives
1.0000 0 −0.4450 𝜙11 0
0 1.0000 −0.8019 𝜙21 = 0
0 0 0 𝜙31 0

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Brilliant.org. (2021) Cardano's Method. Retrieved 19:50, April 5, 2021, from https://brilliant.org/wiki/cardano-method/.
By assuming 𝜙31 = 1, 𝜙11 = 0.4450 by the first equation (row 1), and
𝜙21 = 0.8019 by the second equation (row 2). The mode shape of the
1st mode is
𝜙11 0.4450
𝛟1 = 𝜙21 = 0.8019
𝜙31 1
The mode shape of the 2nd mode.
4.4504 −10.0000 0 𝜙12 0
𝐊 − 𝜔22 𝐌 𝛟2 = −10.0000 4.4504 −10.0000 𝜙22 = 0
0 −10.0000 −5.5496 𝜙32 0
1.0000 0 1.2470 𝜙12 0
0 1.0000 0.5550 𝜙22 = 0
0 0 0 𝜙32 0
𝜙12 −1.2470
𝛟2 = 𝜙22 = −0.5550 By assuming 𝜙32 = 1
𝜙32 in System Identification
Lam HF, Yang JH, 2023. Vibration Testing and Applications 1 of Civil Engineering Structures. CRC.
The mode shape of the 3rd mode.
−12.4698 −10.0000 0 𝜙13 0
𝐊 − 𝜔32 𝐌 𝛟3 = −10.0000 −12.4698 −10.0000 𝜙23 = 0
0 −10.0000 −22.4698 𝜙33 0
𝜙13 1.8019
𝛟3 = 𝜙23 = −2.2470 𝚽 = 𝛟1 , 𝛟2 , 𝛟3 =
𝜙33 1
0.4450 −1.2470 1.8019
0.8019 −0.5550 −2.2470
1 1 1

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Eigenvalue Problem: Practical
Considerations
 It is relatively easy for a two-DOF system (solving a quadratic
equation), and it is not easy for a three-DOF system (solving a cubic
equation). For system with four or more DOFs, there is no analytical
formula available for calculating the roots of a polynomial with
order higher than three. We introduce a practical way for solving
eigenvalue problem, i.e., use MATLAB function eig().
The syntax of eig() is:
>> [V,D]=eig(A,B)
where A and B are the two input matrices dimensions 𝑁 × 𝑁. The
corresponding eigenvalue problem is
𝐀𝐕 = 𝐁𝐕𝐃

24
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
D is the diagonal matrix of eigenvalues, and V is the matrix of
eigenvectors:
𝑑1
𝑑2
𝐃=

𝑑𝑁
𝑣11 𝑣12 ⋯ 𝑣1𝑁
𝑣21 𝑣22 ⋯ 𝑣2𝑁
𝐕 = 𝐯1 , 𝐯2 , ⋯ , 𝐯𝑁 = ⋮ ⋮ ⋱ ⋮
𝑣𝑁1 𝑣𝑁2 ⋯ 𝑣𝑁𝑁
It is clear that the two input matrices are 𝐀 = 𝐊 and 𝐁 = 𝐌, by
comparing to our eigenvalue problem
𝐊𝚽 = 𝐌𝚽𝛀
𝐀𝐕 = 𝐁𝐕𝐃

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Example. Use MATLAB to
calculate the natural frequencies
and mode shapes of an 8-story
shear building model with constant
inter-story stiffness of 1000 N/m
and constant floor mass of 1 kg
through the entire building. Plot
the mode shapes in MATLAB.

Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Orthogonal Property of Mode Shapes
 Orthogonal property of mode shapes is a very important property in
dynamics. It is the foundation for decomposing physical responses
into independent coordinates in modal space, i.e., modal
superposition.
Considering the 𝑎th mode of vibration in the eigenvalue problem
𝐊𝛟𝑎 = 𝜔𝑎2 𝐌𝛟𝑎
Pre-multiply 𝛟𝑇𝑏 of the 𝑏th vibration mode
𝛟𝑇𝑏 𝐊𝛟𝑎 = 𝜔𝑎2 𝛟𝑇𝑏 𝐌𝛟𝑎
Similarly, one can consider the 𝑏th mode and pre-multiply it with 𝛟𝑇𝑎
𝛟𝑇𝑎 𝐊𝛟𝑏 = 𝜔𝑏2 𝛟𝑇𝑎 𝐌𝛟𝑏

27
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Taking transpose for both sides of the mode 𝑎 equation
𝑇 𝑇 2 𝑇 𝑇
𝛟𝑏 𝐊𝛟𝑎 = 𝜔𝑎 𝛟𝑏 𝐌𝛟𝑎 ⟹
𝑇 𝑇 𝑇 𝑇
𝐊𝛟𝑎 𝛟𝑏 = 𝜔𝑎2 𝐌𝛟𝑎 𝑇 𝜔𝑎2
𝑇
𝛟𝑏
𝛟𝑇 𝐊 𝑇 𝛟 = 𝜔2 𝛟𝑇 𝐌 𝑇 𝛟
Due to symmetry, 𝑎 𝑏 𝑎 𝑎 𝑏
𝐊 𝑇 = 𝐊, 𝐌 𝑇 = 𝐌 𝛟𝑇𝑎 𝐊𝛟𝑏 = 𝜔𝑎2 𝛟𝑇𝑎 𝐌𝛟𝑏
Subtract the equation for mode 𝑏, 𝛟𝑇𝑎 𝐊𝛟𝑏 = 𝜔𝑏2 𝛟𝑇𝑎 𝐌𝛟𝑏
0 = 𝜔𝑎2 − 𝜔𝑏2 𝛟𝑇𝑎 𝐌𝛟𝑏
Substitute into here and we
When 𝑎 is not equal to 𝑏, 𝜔𝑎2 − 𝜔𝑏2 ≠ 0, so get orthogonal property of
stiffness matrix
𝛟𝑇𝑎 𝐌𝛟𝑏 = 0, for 𝑎 ≠ 𝑏
Substitution directly gives the orthogonal property of stiffness matrix
𝛟𝑇𝑎 𝐊𝛟𝑏 = 0, for 𝑎 ≠ 𝑏

28
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 When 𝑎 = 𝑏, we can define the modal mass of mode 𝑎 based on the
orthogonal property of the mass matrix
𝑚𝑎 = 𝛟𝑇𝑎 𝐌𝛟𝑎
 When 𝑎 = 𝑏, we can also define the modal stiffness of mode 𝑎 based
on the orthogonal property of the stiffness matrix
𝑘𝑎 = 𝛟𝑇𝑎 𝐊𝛟𝑎

29
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Mode Shape Scaling and Mode shape
Normalization
 Eigenvectors are only determined up to a scaling factor. In
computations, a scaling constraint is applied to uniquely determine
the eigenvector. One common way is to scale it to have unit
‘Euclidean norm’, i.e., , which is equivalent to having the
sum of the squared values of the vector to be equal to 1.

Another way is to scale it to be 1 at a particular DOF. We have done


this for the shear building examples. Scaling affects the modal
response but not the physical response; nature does not care about
what scaling we use. It is a good practice to indicate the scaling applied
when presenting results.

30
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Another popular normalization method is using the orthogonal
property. We normalize the mode shape by dividing with a scaling
factor, i.e., the modal mass
1 1
𝛙𝑖 = 𝛟𝑖 = 𝛟𝑖 , for 𝑖 = 1, … , 𝑁.
𝑇 𝑚𝑖
𝛟𝑖 𝐌𝛟𝑖

Similar to the mode shape matrix, it is convenient to define the mass


normalized mode shape matrix as:
𝜓11 𝜓12 ⋯ 𝜓1𝑁
𝜓21 𝜓22 ⋯ 𝜓2𝑁
𝚿 = 𝛙1 , 𝛙2 , ⋯ , 𝛙𝑁 =
⋮ ⋮ ⋱ ⋮
𝜓𝑁1 𝜓𝑁2 ⋯ 𝜓𝑁𝑁

31
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
The orthogonal property of mass normalized mode shapes to the
system mass matrix becomes:
0, for 𝑖 ≠ 𝑗
𝛙𝑇𝑖 𝐌𝛙𝑗 =ቊ
1, for 𝑖 = 𝑗

Now use the mass normalized mode shape for the orthogonal property
of stiffness matrix:
𝛙𝑇𝑖 𝐊𝛙𝑗 = 𝜔𝑗2 𝛙𝑇𝑖 𝐌𝛙𝑗

Substituting the above property for the stiffness gives


0, for 𝑖 ≠ 𝑗
𝛙𝑇𝑖 𝐊𝛙𝑗 =ቊ 2
𝜔𝑖 , for 𝑖 = 𝑗

32
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
The orthogonal property of mass normalized mode shapes in matrix
form:
𝚿 𝑇 𝐌𝚿 = 𝐈 and 𝚿 𝑇 𝐊𝚿 = 𝛀
1 𝜔12
1 𝜔22
𝐈= and 𝛀 =
⋱ ⋱
1 2
𝜔𝑁
Similar to the mode shape matrix, it is convenient to define the mass
normalized mode shape matrix as:
𝜓11 𝜓12 ⋯ 𝜓1𝑁
𝜓21 𝜓22 ⋯ 𝜓2𝑁
𝚿 = 𝛙1 , 𝛙2 , ⋯ , 𝛙𝑁 =
⋮ ⋮ ⋱ ⋮
𝜓𝑁1 𝜓𝑁2 ⋯ 𝜓𝑁𝑁

33
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Example. Consider the following two-story shear building model,
determine the mass normalized mode shapes, and verify the
orthogonal property.
The mode shapes have been calculated before. Calculate the modal
masses.
𝑇 0.7266 1 2700 0 0.7266 −2.7527
𝚽 𝐌𝚽 =
−2.7527 1 0 5400 1 1
6.8253 0
= × 103 kg
0 25.8589
1 0.0088
𝛙1 = 𝛟1 =
6.8253 × 103 0.0121
1 −0.0171
𝛙2 = 𝛟2 =
25.8589 × 103 0.0062

34
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
To verify the orthogonal property of mode shapes
0.0088 0.0121 2700 0 0.0088 −0.0171
=
−0.0171 0.0062 0 5400 0.0121 0.0062
𝚿 𝑇 𝐌𝚿
1 0
= =𝐈
0 1
0.0088 0.0121 9.953 −6.578 0.0088 −0.0171
= ×106
−0.0171 0.0062 −6.578 6.578 0.0121 0.0062
𝚿 𝑇 𝐊𝚿 𝜔12 0
333.0930 0
= = 2 =𝛀
0 4571.3514 0 𝜔2

35
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Response Computation of MDOF
Systems: Modal Superposition
The mode shapes constitute an orthogonal basis, so the displacement of
a system can be written as the superposition of different modal
components:
𝑁𝑚

𝐱 𝑡 = ෍ 𝛗𝑚 𝑞𝑚 𝑡
𝑚=1
where each modal component is the product of a vibration amplitude
(temporal variation) 𝑞𝑚 (𝑡) and a vibration shape (spatial variation)
𝝋𝑚 ; 𝑚 is the mode number; 𝑁𝑚 is the total number of modes
considered. The above modal superposition also applies for velocity
and acceleration:
𝑁𝑚 𝑁𝑚

𝐱ሷ 𝑡 = ෍ 𝛗𝑚 𝑞ሷ 𝑚 𝑡 , 𝐱ሶ 𝑡 = ෍ 𝛗𝑚 𝑞ሶ 𝑚 𝑡
36 𝑚=1 𝑚=1
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Consider an under-damped dynamic system
𝐌𝐱ሷ 𝑡 + 𝐂𝐱ሶ 𝑡 + 𝐊𝐱 𝑡 = 𝐟(𝑡)
Substituting the modal decomposition
𝑁𝑚 𝑁𝑚 𝑁𝑚

෍ 𝐌𝛗𝑚 𝑞ሷ 𝑚 𝑡 + ෍ 𝐂𝛗𝑚 𝑞ሶ 𝑚 𝑡 + ෍ 𝐊𝛗𝑚 𝑞𝑚 𝑡 = 𝐟(𝑡)


𝑚=1 𝑚=1 𝑚=1
How about damping?
Pre-multiplying 𝛗𝑇𝑛 gives
𝑁𝑚 𝑁𝑚 𝑁𝑚

෍ 𝛗𝑇𝑛 𝐌𝛗𝑚 𝑞ሷ 𝑚 𝑡 + ෍ 𝛗𝑇𝑛 𝐂𝛗𝑚 𝑞ሶ 𝑚 𝑡 + ෍ 𝛗𝑇𝑛 𝐊𝛗𝑚 𝑞𝑚 𝑡 = 𝛗𝑇𝑛 𝐟(𝑡)


𝑚=1 𝑚=1 𝑚=1

Using the orthogonality conditions for mass and stiffness matrices,


only the terms when 𝑚 = 𝑛 remains, i.e., the modal mass and modal
stiffness

37
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Assuming that orthogonality property also holds for damping matrix
𝛗𝑇𝑛 𝐂𝛗𝑚 = 0, when 𝑚 ≠ 𝑛
𝛗𝑇𝑚 𝐂𝛗𝑚 = 𝐶𝑚 , when 𝑚 = 𝑛
where 𝐶𝑚 is defined as modal damping of the 𝑚-th mode.
Finally, the MDOF equation of motion can be decoupled to the SDOF
one for the 𝑚-th mode:
𝑀𝑚 𝑞ሷ 𝑚 𝑡 + 𝐶𝑚 𝑞ሶ 𝑚 + 𝐾𝑚 𝑞𝑚 𝑡 = 𝛗𝑇𝑚 𝐟(𝑡)
Dividing 𝑀𝑚 gives
𝛗𝑇 𝐟(𝑡)
2𝑞 𝑚
𝑞ሷ 𝑚 𝑡 + 2𝜔𝑚 𝜉𝑚 𝑞ሶ 𝑚 + 𝜔𝑚 𝑚 𝑡 =
𝑀𝑚
where 𝜔𝑚 is the natural frequency of the 𝑚-th mode; 𝜉𝑚 is the
damping ratio of the 𝑚-th mode
2
𝐾𝑚 𝐶𝑚
𝜔𝑚 = , 2𝜔𝑚 𝜉𝑚 =
𝑀𝑚 𝑀𝑚

38
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
 Summary of response calculation using modal superposition.
1. Calculate natural frequencies {𝜔𝑚 : 𝑚 = 1, ⋯ , 𝑁𝑚 } and mode
shapes {𝛗𝑚 : 𝑚 = 1, ⋯ , 𝑁𝑚 } by solving the eigen value problem
𝐊𝛗 = 𝜔2 𝐌𝛗.
2. Determine the modal initial displacement 𝐪 0 =
𝑇
𝑞1 0 , 𝑞2 0 , ⋯ , 𝑞𝑁𝑚 (0) and modal initial velocity 𝐪ሶ 0 =
𝑇
𝑞ሶ 1 0 , 𝑞ሶ 2 0 , ⋯ , 𝑞ሶ 𝑁𝑚 (0) given the physical displacement and
velocity 𝐱 0 and 𝐱(0). ሶ
𝑁𝑚

𝐱 𝑡 = ෍ 𝛗𝑚 𝑞𝑚 𝑡 = 𝚽𝐪 𝑡 ⟹ 𝐪 0 = 𝚽 −1 𝐱 0 , 𝐪ሶ 0 = 𝚽 −1 𝐱(0)

𝑚=1
3. For each mode, solve the modal SDOF equation of motion with
modal initial condition to get modal response 𝑞𝑚 𝑡
𝛗 𝑇 𝐟(𝑡)
2 𝑚
𝑞ሷ 𝑚 𝑡 + 2𝜔𝑚 𝜉𝑚 𝑞ሶ 𝑚 + 𝜔𝑚 𝑞𝑚 𝑡 =
𝑀𝑚
39
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
4. Transform the modal responses back to the physical responses:
𝑁𝑚

𝐱 𝑡 = ෍ 𝛗𝑚 𝑞𝑚 𝑡
𝑚=1

40
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
More on Damping: Rayleigh
We have considered damping modeling by modal damping ratios. Now
Let’s see how to construct damping matrix. Rayleigh damping matrix
is constructed as a linear combination of the system stiffness and mass
matrices to ensure its orthogonal property
𝐂 = 𝑎𝐌 𝐌 + 𝑎𝐊 𝐊
Pre- and post-multiplying by 𝚿𝑇 and 𝚿, respectively, one obtains
the following based on the orthogonal property:
𝚿𝑇 𝐂𝚿 = 𝑎𝐌𝚿𝑇 𝐌𝚿 + 𝑎𝐊 𝚿𝑇 𝐊𝚿
𝑐𝑚 = 𝑎𝐌 𝑀𝑚 + 𝑎𝐊 𝐾𝑚 , for 𝑖 = 1, … , 𝑁
𝑐𝑚 /𝑀𝑚 = 𝑎𝐌 + 𝑎𝐊 𝐾𝑚 /𝑀𝑚
2𝜁𝑖 𝜔𝑖 = 𝑎𝐌 + 𝑎𝐊 𝜔𝑖2 , for 𝑖 = 1, … , 𝑁.

41
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
We have only two unknown factors, and only two damping ratios are
used for solving the following two equations:

1 𝜔𝑖
𝑎𝐌 + 𝑎𝐊 = 𝜁𝑖
2𝜔𝑖 2
1 𝜔𝑗
𝑎𝐌 + 𝑎𝐊 = 𝜁𝑗
2𝜔𝑗 2
The thing is that we need to obtain the damping ratios of two modes,
any two modes are OK.

42
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
More on Damping: Caughey Damping
Rayleigh damping allows one to fit the damping ratios of two modes
only. If one can find one additional matrix with orthogonal property,
the damping for one additional mode can be fitted. Consider pre-
multiplying 𝛟𝑇𝑏 𝐊𝐌−1 to the eigenvalue problem 𝐊𝛟𝑎 = 𝜔𝑎2 𝐌𝛟𝑎
𝛟𝑇𝑏 𝐊𝐌 −1 𝐊𝛟𝑎 = 𝜔𝑎2 𝛟𝑇𝑏 𝐊𝐌 −1 𝐌𝛟𝑎 = 𝜔𝑎2 𝛟𝑇𝑏 𝐊𝛟𝑎
As 𝐊 has the orthogonal property, we can prove that 𝐊𝐌 −1 𝐊 on the
right-hand side also has the orthogonal property.
2
Next, pre-multiplying 𝛟𝑇𝑏 𝐊𝐌−1 (note the power introduced) to the
eigenvalue problem This is just proved to be orthogonal!
𝑇 2
𝛟𝑏 𝐊𝐌 −1 𝐊𝛟𝑎 = 𝜔𝑎2 𝛟𝑇𝑏 𝐊𝐌−1 𝐊𝐌−1 𝐌𝛟𝑎 = 𝜔𝑎2 𝛟𝑇𝑏 𝐊𝐌−1𝐊𝛟𝑎
2
Similarly, 𝐊𝐌−1 𝐊 on the left-hand side also has the orthogonal
property.
43
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
𝑙
By induction, 𝐊𝐌−1 𝐊 , for any positive integer 𝑙 , have the
orthogonal property. There is an infinite number of matrices from this
series. Thus, one can fit the damping ratios of as many modes as one
wants. Now we have found the good candidate that is orthogonal for
constructing orthogonal damping matrix.
To include the original two (i.e., 𝐌 and 𝐊) in the series, the series of
matrices can be expressed as 𝐌 𝐌 −1 𝐊 𝑙 , for any positive integer 𝑙
(including zero).
 When 𝑙 = 0, 𝐌 𝐌 −1 𝐊 0 = 𝐌

 When 𝑙 = 1, 𝐌 𝐌 −1 𝐊 1 = 𝐊 We just saw these two before.

 When 𝑙 = 2, 𝐌 𝐌 −1 𝐊 𝐌 −1 𝐊 = 𝐊𝐌 −1 𝐊

 When 𝑙 = 3, 𝐌 𝐌 −1 𝐊 𝐌 −1 𝐊 𝐌 −1 𝐊 = 𝐊𝐌 −1 2 𝐊.

44
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
Caughey damping matrix can be constructed as, by using the
orthogonal matrix series we just found:
𝑁𝜁 −1

𝐂 = 𝐌 ෍ 𝑎𝑙 𝐌 −1 𝐊 𝑙

𝑙=0

These coefficients are to be calculated using damping ratios of


different modes
To calculate 𝑎𝑙 , pre- and post-multiply the mass normalized mode
shapes 𝚿𝑏𝑇 and 𝚿𝑎 ,
𝚿𝑏𝑇 𝐂𝚿𝑎
𝑇 −1 −1 2
= 𝚿𝑏 𝐌 ቆ𝑎0𝚿𝑎 + 𝑎1 𝐌 𝐊 𝚿𝑎 + 𝑎2 𝐌 𝐊 𝚿𝑎 + ⋯
−1 𝑙 −1 𝑁𝜁−1
+ 𝑎𝑙 𝐌 𝐊 𝚿𝑎 + ⋯+ 𝑎𝑁𝜁−1 𝐌 𝐊 𝚿𝑎 ቇ

45
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
The first term on the RHS is non-zero only when 𝑎 = 𝑏
𝑎0 𝚿𝑎𝑇 𝐌𝚿𝑎 = 𝑎0
The second term on the RHS is non-zero only when 𝑎 = 𝑏, due to the
orthogonal property of the matrix series we found 𝐊𝚿𝑎 = 𝜔𝑎2 𝐌𝚿𝑎
𝑎1 𝚿𝑎𝑇 𝐌𝐌−1 𝐊𝚿𝑎 = 𝑎1 𝚿𝑎𝑇 𝐊𝚿𝑎
= 𝑎1 𝚿𝑎𝑇 𝜔𝑎2 𝐌𝚿𝑎 = 𝑎1 𝜔𝑎2
Similarly, the third term on the RHS is non-zero only when 𝑎 = 𝑏
𝑎2 𝚿𝑎𝑇 𝐌𝐌 −1 𝐊𝐌 −1 𝐊𝚿𝑎 = 𝑎2 𝚿𝑎𝑇 𝐊𝐌 −1 𝐊𝚿𝑎 = 𝑎2 𝚿𝑎𝑇 𝐊𝐌−1 𝜔𝑎2 𝐌𝚿𝑎
= 𝑎2 𝜔𝑎2 𝚿𝑎𝑇 𝐊𝚿𝑎 = 𝑎2 𝜔𝑎2×2
By mathematical induction, the fourth term can be expressed as
2 𝑙−1
𝑎3𝜔𝑎2×3, …, the 𝑙-th term in this series is 𝑎𝑙−1 𝜔𝑎
For one mode 𝑖, we have the equation for the Cauhey damping series
1 2 𝑁𝜁 −1 −1
𝜁𝑖 = 𝑎0 𝜔𝑖−1 + 𝑎1 𝜔𝑖 + 𝑎2 𝜔𝑖3 + ⋯ + 𝑎𝑙 𝜔𝑖2𝑙−1 + ⋯ + 𝑎𝑁𝜁−1 𝜔𝑖
46 2
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.
For one mode 𝑖, we have the equation for the Cauhey damping series
𝚿𝑖𝑇 𝐂𝚿𝒊
= 𝚿𝑖𝑇 𝐌 ቀ𝑎0 𝚿𝑖 + 𝑎1 𝐌 −1 𝐊 𝚿𝑖 + 𝑎2 𝐌 −1 𝐊 2 𝚿𝑖 + ⋯ + 𝑎𝑙 𝐌 −1 𝐊 𝑙 𝚿𝑖 + ⋯
𝑁𝜁 −1
+ 𝑎𝑁𝜁−1 𝐌 −1 𝐊 𝚿𝑖 ቁ
2 𝑁𝜁 −1
2𝜁𝑖 𝜔𝑖 = 𝑎0 + 𝑎1 𝜔𝑖2 + 𝑎2 𝜔𝑖4 + ⋯+ 𝑎𝑙 𝜔𝑖2𝑙 + ⋯+ 𝑎𝑁𝜁 −1 𝜔𝑖
1 −1 3 2𝑙−1 2 𝑁𝜁 −1 −1
𝜁𝑖 = 𝑎0 𝜔𝑖 + 𝑎1 𝜔𝑖 + 𝑎2 𝜔𝑖 + ⋯ + 𝑎𝑙 𝜔𝑖 + ⋯ + 𝑎𝑁𝜁−1 𝜔𝑖
2
Assuming that the damping ratios of the first 𝑁𝜁 modes are given,
2𝑁𝜁−3 𝑎0
1ൗ𝜔1 𝜔1 𝜔1 ⋯ 𝜔1 3 2𝑙−1 ⋯ 𝜔1 𝑎1
𝑎 𝜁1
2𝑁 −3 2
1 1ൗ𝜔2 𝜔2 𝜔23 ⋯ 𝜔22𝑙−1 ⋯ 𝜔2 𝜁 𝜁
⋮ = ⋮2
2 ⋮ ⋮ ⋮ ⋮ ⋮ 𝑎𝑙
3 2𝑙−1 2𝑁𝜁−3 ⋮ 𝜁𝑁 𝜁
1ൗ𝜔𝑁𝜁 𝜔𝑁𝜁 𝜔𝑁 ⋯ 𝜔𝑁 ⋯ 𝜔𝑁 𝑎
𝜁 𝜁 𝜁 𝑁𝜁−1

47
Lam HF, Yang JH, 2023. Vibration Testing and Applications in System Identification of Civil Engineering Structures. CRC.

You might also like