ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 01: 17 January 2006 Introduction

Dynamics according to Galilei Galileo (Two New Sciences, 1636): A subject of never-ending interest. Isaac Newton (The Principia, 1687): We offer this work as the mathematical principles of philosophy; for all the difculty of philosophy seems to consist in this from phenomena of motions to investigate the forces of nature, and then from these forces to demonstrate the other phenomena; and to this end the general propositions in the rst and second book are directed.

What we will learn/acquire in this course: 1. Kinematics which are essential for the description of motions of masses. 2. Ability to derive the equations of motion for complex dynamical systems. 3. Understanding of vibration phenomena for structural elements. 4. Ability to model complex vibration problems by reduced number of equations. 5. Applications of dynamical principles for engineering design applications. Above All, Become Excellent Dynamists!

Transition from Statics to Dynamics Statics For ce equilibrium : f or a f r ee body i Moment equilibrium : ar ound point P
j fi j = 0

Dynamics fi j m i ai = 0 j M P j r P j (m j a j ) = 0

j MP j = 0

Observation: The crucial aspect of dynamics is the need to compute the acceleration vector for every mass in the system.

Landaus Uniqueness Theorem If all the position vectors {x1 , x2 , ..., xn } and the velocity vec tors { 1 , x2 , ..., xn } for n particles are given at some instant, x the accelerations { 1 , x2 , ..., xn } at that instant are uniquely x dened. The relations between the accelerations, velocities and position vectors are called the equations of motion. L. D. Landau and E. M. Lifshits, Mechanics (3rd ed.), Pergamon Press, 1959.

Reference Frame and Position Vector Consider two frames, K and K , where V is constant in both magnitude and direction, and in which the properties of space and time are the same.
Example of Two Inertial Frames

Frame K'

Frame K

r r' Vt

O r = Vt + r'

O'

Reference Frame and Position Vector-Contd The kinematic relations between frames K and K are: r = Vt + r r=V+r v=V+v r=r a=a Conclusion: The acceleration vectors are the same in all inertial frames! Hence, Galileos relativity principle holds. In all inertial frames, the laws of mechanics are the same, which is referred to Galileos relativity principle, one of the most important principles of mechanics.

Newtons Eight Denitions

DEFINITION I: The quantity of matter[mass] is the measure of the same, arising from its density and bulk conjointly. DEFINITION II: The quantity of motion[linear momentum] is the measure of the same, arising from the velocity and quantity of matter conjointly. DEFINITION III: The vis insita, or the innate force of matter[inertia force], is a power of resisting, by which every body, as much as in it lies, continues in its present state, whether it be rest, or of moving uniformly forwards in a right[straight] line. DEFINITION IV: An impressed force is an action exerted upon a body, in order to change its state, either of rest, or of uniform motion in a right line.

Newtons Eight Denitions - contd

DEFINITION V: A centripetal force is that by which bodies are drawn or impelled, or any way tend, towards a point as to a centre. DEFINITION VI: An absolute centripetal force is the measure of the same, proportional to the efcacy of the cause that propagates it from the centre, through the spaces round about. DEFINITION VII: The accelerative quantity of a centripetal force is the measure of the same, proportional to the velocity which it generates in a given time. DEFINITION VIII: The motive[motion-causing] quantity of a centripetal force is the measure of the same, proportional to the motion which it generates in a given time.

Newtons Three Laws Law I: Every body continues in its state of rest, or of uniform motion in a right line, unless it is compelled to change that state by forces impressed upon it. Law II: The change of motion is proportional to the motive force impressed; and is made in the direction of the right line in which the force is impressed. Law III: To every action there is always opposed an equal reaction; or, the mutual actions of two bodies upon each other are always equal, and directed to contrary parts.

Test of Galileos Relativity Principle

B
X X' x' d x mg k(x-d) = kx' ma

m k
o
Free-free state

o'
Frame K' (mass center)

Frame K

Static displacement

Assumed dynamic state

Test of Galileos Relativity Principle-contd Position vector in frame K : X = h d + x Position vector in frame K : X = x Relation between the two frames : x = x d Equation of motion in frame K : mg k(x d) m = 0 x x Equation of motion in frame K : mg kx m = 0

Test of Galileos Relativity Principle-contd EOM in frame K : m + kx = (mg + kd)(static eq.!) = 0 x x EOM in frame K : m + kx = mg Remark 1: The mass-point based coordinate system (frame K) must include the dead weight mg in the equations of motion. Remark 2: The ground based coordinate system (frame K ) does not require to account for the static equilibrium. Remark 3: When the ground itself moves, i.e., buildings subjected earthquakes and automobile riding on wavy roads, one has to modify h to X = xg + h d + x where xg is the motion of the ground. A similar modication must be made for X .

Test of Galileos Relativity Principle-contd What happens to the mass point-based coordinate system when the ground moves?

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures

Lecture 02: 19 January 2006 Vibration of Two-DOF Systems

1. Equations of Motion Consider a two-DOF model as shown in Fig. 1. The free-body diagrams for masses M1 and m 2 are also shown in Fig. 1. Note that the inertia forces for both masses are associated with minus sign, for the inertia forces can be considered as resisting forces (cf., f ma = 0). Summing the forces acting on each mass, the equations of motion for the coupled two-mass-spring-damper system can be written as f = f 1 (t) M1 x1 + {k2 (x2 x1 ) + c2 (x2 x1 ) K 1 x1 } = 0 f = f 2 (t) m 2 x2 {k2 (x2 x1 ) + c2 (x2 x1 )} = 0 M1 x1 = f 1 (t) K 1 x1 + k2 (x2 x1 ) + c2 (x2 x1 ) m 2 x2 = k2 (x2 x1 ) c2 (x2 x1 ) (1)

For M1 : For m 2 :

Two DOF Spring-Mass-Damper Model

x k
K
M


C

Free Body Diagram for Two DOF Spring-Mass-Damper Model

(-M
K1 x 1
M

x1

f1(J )

)
x k2 ( 2 x c2 (2
-

x k2 ( 2 x1 ) x1 ) 2 c2 (x

x1 ) x1 )
m

(- m
B

Fig. 1 Two DOF Simplied Vertical Motion Model

x2

As the preceding equation involves two displacements, x1 and x2 , its general solution involves complex matrix differential algebra. For design considerations, however, important insight can be gained by considering the special case of forcing function given by f 1 (t) = F1 e j t , f 2 (t) = F2 e j t (2) so that the solution assumes the form of x1 (t) = X 1 e j t x2 (t) = X 2 e Substituting (2) and (3) into (1), we obtain 2 M1 X 1 = F1 K 1 X 1 + k2 (X 2 X 1 ) + jc2 (X 2 X 1 ) m 2 X 2 = F2 k2 (X 2 X 1 ) jc2 (X 2 X 1 )
2 j t

(3)

(4)

In order to solve for X 1 and X 2 , lets rearrange the preceding equation to read (2 M1 + jc2 + K 1 + k2 ) X 1 = F1 + (k2 + jc2 ) X 2 ( m 2 + jc2 + k2 ) X 2 = F2 + (k2 + jc2 )X 1
2

(5)

Solving for X 1 and X 2 we obtain H2 ()F1 + H12 ()F2 X1 = 2 H1 () H2 () H12 () X2 = H1 ()F2 + H12 ()F1 2 H1 ()H2 () H12 ()

(6)

H1 () = (2 M1 + jc2 + K 1 + k2 ) H2 () = (2 m 2 + jc2 + k2 ) H12 () = (k2 + jc2 ) Although (6) appears to be very complex, several simplications are possible to aid vibration designers. This is studied below.

2. What Are Vibration Modes and Mode Shape? It turns out that the motions of X 1 and X 2 given by (6) are not randomly independent as the solution may suggest. They are interlinked by the property called mode shapes. Understanding the physical properties of mode shapes is important if one is tasked to design structures subject to vibrations. As a motivation, let us consider the following special 2-DOF case: m 1 = m 2 = 1, k1 = k2 = 2.618 2 , c2 = 0, f 1 (t) = 0 (7)

For this model, we apply two different excitations: f 2 (t) = 0.2 sin (0.98 t) sin (0.98 2.618t) (8)

Figure 2 illustrates the two responses subject to the two excitations specied in (8). Observe that, for the case of f 2 (t) = 0.2 sin (0.98 t), both mass m 1 and mass m 2 are moving in phase, that is, they move in the same direction in time. However, when the system is subjected to f 2 (t) = sin (0.98 2.618t), mass m 1 and mass m 2 are moving out phase, that is, they move in the opposite directions in time. In other words, depending on the excitation frequency, the motions of the two masses are drastically different. To understand this strange phenomena, one has to understand the roles of modes and mode shapes. To this end, let us recast (1) in a matrix form with c = 0: m1 0 0 m2 x1 x2 + k1 + k2 k2 k2 k2 x1 x2 = f 1 (t) f 2 (t) (9)

The characteristic equation of the above coupled 2-dof differential equation(9) can be obtained as follows. First, we assume the solution of their homogeneous equations in the form x1 x2 = x1 x2 e jt (10)

2-dof system response due to harmonic input at mass 2 with omega= 0.98*omega 1
2.5

2.5

2-dof system response due to harmonic input at mass 2 with omega= 0.98*omega2

m2
2

m2

Position of Mass 2

Position of
Mass Positions for Mode 2

Mass Positions for Mode 1

Mass 2

1.5

1.5

k2

m1

m1

Position of Mass 1

Position of Mass 1
0.5 0.5

k1

0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Time

Time

Figure 2 Motions of two masses under two different excitations

Second, substituting this into (9) with f 1 = f 2 = 0 yields: [2 m1 0 0 k + k2 + 1 m2 k2 k2 x ] 1 k2 x2 =0 (11)

Hence, the characteristic equation is obtained by requiring that the above equation has a nontrivial solution: k1 + k2 k2 2 k1 k2 + ) + =0 m1 m2 m1m2 (12) Note that, with m 1 = m 2 = 1, k1 = k2 = 2.618 2 , the two roots of the above characteristic equation are given by det k1 + k2 2 m 1 k2 k2 k2 2 m 2 =0 4 ( n1 = , n2 = 2.618 (13)

These two values are called characteristic values whose square roots are called the natural frequencies or vibration modes of the system.

The corresponding eigenvectors can be computed from the second row of (11): x2 k2 x1 + k2 x2 = 0
2

x1 k2 2 2 = =1 x2 k2 k2

(14)

Note that for the two modes computed in (13), we have two different expressions: x1 2 For = , =1 = 0.618 x2 2.618 2 x1 2.6182 2 For = 2.618, =1 = 1.618 2 x2 2.618 The ratios of these eigenvector sets are plotted in Figure 3.
Mode Shape for Mode 1 2 2 Mode Shape for Mode 2

(15)

Mass Point

Mass Point

Fixed end 0.0 0.6 1.0 -1.6 0

Fixed end 1.0

Modal Amplitude for Mode 1

Modal Amplitude for Mode 2

Figure 3 Mode Shapes of 2-DOF Example Problem

Observe from Figure 3 that for the case of the rst mode when mass 1 moves in the same direction with its amplitude of 0.616 while mass 2 moves a unit amplitude. In other words, the two masses move in phase as illustrated in Figure 2. As for the second mode, mass 1 moves in the opposite direction by -1.618 while mass 2 moves a unit amplitude, which is also illustraed in Figure 2. From these observations we conclude that mode shapes indicate how the system will deform when subjected to a harmonic excitation whose frequency is close to one of the natural frequencies (or vibration modes) of the system. Thus, mode-shape information is useful in designing structures subjected to harmonic excitations. Examples of such systems include propellered airplanes, ships, motor vehicles, and many other machinery equipment.

3. Vibration analysis of 2-DOF Spring-Mass Problem by MATLAB In Matlab we invoke the following routine: [X, D] = eig(K, M); where X is the eigenvector, D is the eigenvalues. For example, the eigenproblem of the 2-DOF system studied in the preceding sections can now be analyzed by MATLAB with 2 2.618 2 K= 2.618 2 2.618 2 , 2.618 2 M= 1 0 0 1 (16)

Upon using the following MATLAB code, we nd the following results: % % 2 dof eigenvalue analysis % %stiffness matrix K = [ 2*2.618*pi^2 -2.618*pi^2; -2.618*pi^2 2.618*pi^2];

% mass matrix M = [ 1 0; 0 1]; % call eigenvalue routine [X, D] = eig(K, M);

% write eigenvector and eigenvalues lambda = diag(D); disp([Eigenvalues of 2-dof system disp( ); disp([Eigenvectors of 2-dof system disp([ % compute the frequencies

num2str(lambda) ]); num2str(X(1,:)) ] ); num2str(X(2,:)) ] );

freq = diag(D);

% extract the two diagonal entries of D matrix

disp( ); freq = sqrt(freq); disp([Frequencies of 2-dof system

num2str(freq)]);

we nd the following results: Eigenvalues of 2-dof system Eigenvectors of 2-dof system Frequencies of 2-dof system 67.6464 0.85065 -0.52573 8.2247 9.86948 0.52573 0.85065 3.1416

Note that MATLAB prints out the highest mode rst. Hence, the mode shapes are given by For the second mode with 2 = 2.618 = 8.2247 : X2 = For the rst mode with 1 = = 3.14157 : X1 = x21 x22 x11 x12 = = 0.85065 0.5257

0.525731 0.850650 (17) The mode shapes computed by MATLAB is normalized so that their vectorial length is unity. If the modal amplitude at mass 2, x22 , for the second mode shape is scaled to be unity from MATLAB computed value x22 = 0.52573, then we would have

x21 = 1.618 as given by (15). The same is true for the modal amplitude for mode 1 at mass 1 and mass 2 points. Finally, one may express X2 as For the second mode with 2 = 2.618 = 8.22474 : X2 = x21 x22 = 0.85065 0.52573 (18)

which indicates that X2 and X2 are equally valid representation. 4. Den Hartog-Ormondroyd Oscillators

An important application of the two-dof vibration model is to study shock isolation design of a suspension system as shown Figure 1. If Figure 1 is rearranged as shown in Figure 4, the same model can be used as a resonator model whose generic characteristics can be used for the design of accelerometers and more recently as lters of microelectro-mechanical systems (MEMS). In either case, the governing equation (1) is applicable. Let us parameetrize the model as 1 = K 1 /M1 , 2 = k2 /m 2 , c = 2 m 2 2 , = m 2 /M1 (19)

and normalize the frequence response X 1 () given in (6) by its static displacement with F2 = 0:

x1

B
C

k /2 2
K1 /2
2

Two DOF Spring-Mass-Damper Suspension Model

Figure 4: 2-DOF Suspension Model

xst (1) = F1 /K 2 such that from (7) we have H1 = X 1 () K 1 H22 () = 2 xst (1) H11 () H22 () H12 ()

(20)

(21)

Dividing both the nominator and the denominator by M1 m 2 and utilizing the parameters introduced in (20), the frequency response function H1 () at mass 1 can be expressed as H1 = H22 () 2 H11 () H22 () H12 () (22)

2 2 H11 () = 2 + j2 2 + 1 + 2 2 H22 () = 2 + j2 2 + 2 2 H12 () = ( j2 2 + 2 )

Frequency Response Functions for Different Damping Ratios of a 2-DOF Suspension Model 101 Invariant points
Amplitude

100

-1 10

101

Frequency (Hz)

Figure 5: Frequency Response Function at Mass 1

Figure 5 illustrates the frequency response function (FRF) for different damping ratios with 1 = 2 = 10 H z, = 0.1 (23) Note that there are two points that all FRF curves pass through two invariant points. It is these invariant points that were rst discovered by den Hartog and Ormondroyd in 1928 who subsequently utilized their properties for improved design of mechanical shock attenuation. In their design study by using Figure 1, den Hartog and Ormondroyd observed that if the magnitudes of the two invariant points are made to be the same by varying the mass ratio and the damping ratio, a near optimum design goal can be achieved. For the design of accelerometers using Figure 4, the objective is to maximize the amplitude of the frequency response of mass 2 in order to decrease the signal to noise ratio while realizing a at plateau of frequency interval of interest. For the design of resonators using Figure 4, in addition to maximizing the amplitude of FRF at mass 2, the gap between the two invariant points should be minimized. MATLAB code that produced Figure 5 islisted below.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % hartog_oscillator.m % by k c park 15 january 2002 clear % Define some useful numbers: dtr=pi/180; Hz2rps=2*pi; rps2Hz=1/2/pi; % Conversions to/from Hz and rad/s % Natural frequencies of individual masses: 10Hz w1=10*Hz2rps; w2=w1; mu = 0.1; for zeta =0:0.05:0.8 %zeta is the damping ratio

% Define a system in transfer function form a0=1; a1 =2*zeta*w2; a2 = w2^2; b0=1; b1 =2*zeta*(1+mu)*w2; b2 = w1^2+w2^2 + mu*(1+4*zeta^2)*w2^2 - 4*mu*zeta^2*w2^2;

b3 = 2*zeta*(mu*w2^3 + w2*(w1^2+mu*w2^2)) - 4*zeta*mu*w2^3; b4 = w2^2*(w1^2+mu*w2^2)- mu*w2^4; sys = tf ( [a0 a1 a2], [b0 b1 b2 b3 b4]);

% Pick a range of frequencies to look at from 0 Hz to 100 Hz wf= logspace(0.5,1.5,400)*Hz2rps; % Find the frequency response H=freqresp(sys,wf); % This ends up being a 3 dimensional array! H=reshape(H,size(wf)); % this makes it the same size as wf % Plot the amplitude as a function of frequency in Hz loglog(wf*rps2Hz,w1^2*abs(H)), grid xlabel(Frequency (Hz)) ylabel(Amplitude) hold on; end axis([5 30 0.1 30]); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

References 1. 2. 3. R. E. D. Bishop, Vibration, Cambridge University Press, 1979. J. P. Den Hartog, Mechanical Vibrations, Dover Pub., 1984. Maurice Roseau, Vibrations in Mechanical Systems, Springer-Verlag, 1984.

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 03: Solution of 1 and 2-DOF Vibrating Models

Governing equation: m x + c x + kx = f (t) p = m x = f (t) kx (c/m) p x = (1/m) p x p 0 = k 1/m c/m x p + 0 f (t)

Canonical equation: x = Ax + Bu(t) Output equation: y = Cx + Du

Solution of the canonical equation: x = eAt x(0) +

0 t

eA(t ) Bu( ) d

eAt x(0)-term: Response due to the initial conditions.

t 0

eA(t ) Bu( ) d -term: Response due to the external (applied) forcing function.

Discrete approximation: x(kT ) = eAkT x(0) + x(kT + T ) = e +

0 0 A(kT +T ) kT +T kT

eA(kT ) Bu( ) d

x(0)

eA(kT +T ) Bu( ) d

Discrete approximation-contd x(kT + T ) = eAT [eAkT x(0)] +

0 kT

eA(kT +T ) Bu( ) d eA(kT +T ) Bu( ) d

kT +T

kT

Discrete approximation -contd x(kT + T ) = eAT [eAkT x(0)] + eAT [ +

0 kT +T kT kT

eA(kT ) Bu( ) d ]

eA(kT +T ) Bu( ) d

x(kT + T ) = eAT x(kT ) +

kT +T kT

eA(kT +T ) Bu( ) d

kT +T kT

eA(kT +T ) Bu( ) d

(via change of variable: = kT + T )

T 0

eA Bu(kT ) (d ), eA Bu(kT ) (d )

with u( ) u(kT )

=
0

= A1 (eAT I) B u(kT ) = u(kT ) where = A1 (eAT I) B

Discrete approximation -concluded

x(kT + T ) = eAT x(kT ) + u(kT ) x(k + 1) = x(k) + u(k) = A1 (eAT I) B

Useful Matlab routines: sys= ss(a,b,c,d): Specify a state space model. H =tf (num, den): Convert a state space model to the corresponding transfer function. [y,t,x]=step(sys): obtain step response. [y,t,x]=lsim(sys,u,t,x0): obtain response due to arbitrary input u. [H] =freqresp(sys,w): compute frequency response over a grid of frequencies. [Hresh]=reshape(H, size(w)): make H the same size as w.

[y,t,x]=impulse(sys,t): obtain impulse response. [y,t,x]=initial(sys,xo): free response due to initial condition.

% vibex.m % Program to compute 1DOF computations % January 16, 2004 (KCP) clear % Define some useful numbers: Hz2rps=2*pi; rps2Hz=1/2/pi; % Conversions to/from Hz and rad/s

% Natural frequency of 1 Hz omega_n=1*Hz2rps; T_n = 2*pi/omega_n; % period of the response % critical damping ratio zeta= 0.05; % a=[0 %5 percent

state space model 1; -omega_n^2 -2*zeta*omega_n];

b=[0; omega_n^2]; c=[1 0; 0 1];

% a force is appled, equivalent to unit static displacement % both momentum and displacement are available

% specify state space model for this problem sys = ss (a,b,c,0); % %%%%%%%%%%%%%%%%%%%%%%%%%%%% % Time response computation %%%%%%%%%%%%%%%%%%%%%%%%%%%% % obtain the response, y1 =displ; y2=mementum % step(sys); % simplest step response computation %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % for specified computations; % % the total time of simulation

Tfinal = 6*T_n; % 6 periods %specify the output time points: dt = (1/40)*(T_n); %40 samples per period %time interval t = 0:dt:Tfinal; %obtain response [y,t,x]=step(sys,t); %plot the response figure(2); plot(t, y(:,1), k); xlabel(Time(sec)); ylabel(Displacement); title(Step Response computed by specifying the time step); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % end of response computation %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% % % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Pick a range of frequencies to look at from 1 Hz to 10 Hz wf= logspace(-1,1,200)*Hz2rps; % Find the frequency response % generte transfer function TransF = tf(sys); H=freqresp(TransF,wf); % This ends up being a 3 dimensional array! Hreshape=reshape(H(1,:,:),size(wf)); % this makes it the same size as wf % Plot the amplitude as a function of frequency in Hz figure(3) loglog(wf*rps2Hz,abs(Hreshape)); grid on; xlabel(Frequency (Hz)) ylabel(Amplitude) frequency response calculation

title(Frequency response of1DOF damped system); % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % end the frequency response %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % %harmonic response of 1dofsystem %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Pick a range of times to look at the time response t=0:0.005:5; % for 5 period of T_n

% sinusoidal forcing function u=sin(Hz2rps*2*t); % omega_forcing = 2 Hertz! [ys,ts]=lsim(sys,u,t);

figure(4) plot(ts,ys(:,1)),grid xlabel(Time (sec)) ylabel(Displacement) title(Time response due to harmonic excitation); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Step Response computed by specifying the time step 2

1.8

1.6

1.4

Displacement

1.2

0.8

0.6

0.4

0.2

3 Time(sec)

Time response due to harmonic excitation 1

0.8

0.6

0.4

Displacement

0.2

-0.2

-0.4

-0.6

-0.8

0.5

1.5

2.5 Time (sec)

3.5

4.5

10

Frequency response of1DOF damped system

10

Amplitude 10
-1

10

-2

10

-1

10

10

Frequency (Hz)

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 04: January 26 Problem Solution via Newtons 2nd Law

A Spring-Mass-Bar System:
y, j

m
oA x, i kx

mg
A o YAC N A

X AC m x N AC

x
K

m
o A

YAC XAC A

C C Mg

(ML2/12) Mg B N F

M r C

Position Vectors Position vector for mass m: r A = xi Position vector for bar M: rC = r A + r A C L r A C = (sin i cos j) 2 (2) (1)

Velocity Vectors Velocity vector for mass m: v A = r A = xi Velocity vector for bar M: vC = v A + v A C L (cos i + sin j) vA C = 2 (3)

(4)

Acceleration Vectors Acceleration vector for mass m: a A = v A = xi Acceleration vector for bar M: aC = a A + a A C L (cos i + sin j) aA C = 2 L2 (sin i cos j) 2 (5)

(6)

Equilibrium equation for mass m:

kxi m xi mgj + N A j + X AC i + Y AC j = 0 (7)

Equilibrium equation for bar M:

Fi MaC Mgj X AC i Y AC j = 0 MC +rC

B

(8)

M L 2 = k + rC 12 Fi = 0

(X AC i Y AC j) (9)

Equilibrium equation for mass m - contd Equation(7) yields: m x + kx X AC = 0 N A mg + Y AC = 0 X AC = m x + kx

(10)

Equilibrium equation for bar M: Equation(8) [sum of forces] yields: L L2 F M(x + cos sin ) 2 2 X AC = 0 L L2 Mg + M( sin + cos ) + Y AC = 0 2 2

(11)

(12)

Equilibrium equation for bar M: Equation(9) [sum of moments] yields: M L 2 k 12 L ( sin i + cos j) (X AC i + Y AC j) 2 L + (sin i cos j) Fi = 0 2 (13)

Equilibrium equation for bar M - concluded M L 2 12 L L + [ cos X AC + sin Y AC ] 2 2 L + cos F = 0 2 (14)

Equations for x By eliminating X AC from (10) and (11), we obtain ML (M + m)x + kx + cos 2 ML sin 2 = F 2

(15)

Equations for L L 1. Obtain ( cos X AC + sin Y AC ) in (14) from 2 2 (11) and (12): L L ( cos X AC + sin Y AC ) = 2 2 (16) 2 L L ML cos (F M x) sin Mg 2 2 4

Equations for - concluded 2. Substitute (16) into (14) to obtain:

ML MgL M L2 + cos x + sin 3 2 2 = L cos F

(17)

Summary: Coupled equations for x and ML (M + m)x + kx + cos 2 ML sin 2 = F 2 M L2 ML MgL x + cos x + sin 3 2 2 = L cos F

(15)

(17)

What if M A were adopted instead of MC ? M L 2 MA = k 12 (Mgj MaC ) + r A B (Fi) = 0

(18)

+r A C

which leads to (17)! Observation: There may be a preferrable choice of locations where one can sum the moment!!!

Appendix
Two independent variables in EOM: Position vector (or displacement): r Linear momentum vector: p = m v, dr v= dt (19)

Newtons second law: F= dp d(mv) = dt dt (20)

Moment of a force and angular momentum Angular momentum: Ho = r p = r (m r) (21)

Rate of angular momentum: Ho = r p + r p = r p = r F = Mo M = Ho (22)

Work and Energy Increment of work: d W = F dr = m r dr dr rdt = d( 1 m r r) = dT =m 2 dt Integrating both sides, we obtain:
r2 r2

(23)

F dr =
r1 r1

d( 1 m r r) 2

(24)

(V2 V1 ) = T2 T1

Denition of potential energy: rr e f V (r) = dW = d V (r) r F dr

Example of potential energy for a spring undergoing displacement:

0 0

V () =

Fd x =

kx d x = 1 k 2 2

(25)

Lifting gravity force from y1 to y2 : y2 y1 F dy = y2 y1 (mgj) dy(j)

= -mg (y2 y1 ) = (V2 V1 )

V2 = mgy2 ,

V1 = mgy1

Systems of discrete bodies Mass center: For a system consisting of N masses whose position vectors are {ri , 1, 2, 3, ...N } , the mass center is dened as
N

i=

m i ri = 0
i=1

(26)

Momentum of N masses: ptotal = m vC m=

N i=1

mi

Kinetic energy of N masses

ri = rc + ri ,
N

vi = vc + vi

T = = +

1 2 i=1

m i (vC + vi ) (vC + vi )
N

1 m 2

vC vC + vC m i vi vi

1 2 i=1

m i vi

(27)

N 1 2 i=1

Since we have from (8),

N i=1

m i ri = 0, which leads to
N

m i vi = 0
i=1

(28)

so that (27) reduces to

N

T = 1 m vC vC + 2

1 2 i=1

m i vi vi

(29)

Rotational Moment of Inertia HO =

m

r (dmv), v = r (30) r ( r) dm
m

= Let r and be expressed as

r = xi + yj + zk = x i + y j + z k

(31)

Step 1: Compute r i j k r = det x y z x y z = (z y yz )i + (xz zx )j + (yx x y )k

(32)

Step 2: Compute r ( r) r ( r) = [(y 2 + z 2 )x x y y x zz ]i + [yxx + (x 2 + z 2 ) y yzz ]j + [zxx zy y (x 2 + y 2 )z ]k (33)

Step 3: Compute

r ( r)dm Hx = Ix x x + Ix y y + Ix z z Hy = I yx x + I yy y + I yz z Hz = Izx x + Izy y + Izz z Ix x =

m

H = Hx i + Hy j + Hz k (34) x y dm, etc.

m

(y 2 + z 2 ) dm, Ix y =

Other quantities may be computed by changing the subscripts from Ix x and Ix y . Step 4: Compute M O = H O

 H O = ( Hx i + Hy j + Hz k) + (Hx + Hy + Hz k) i j Since we have H O becomes = i, = j, k = k i j

(35)

(36)

M O = H O = ( Hx i + Hy j + Hz k) + (Hx i + Hy j + Hz k)

(37)

Kinetic energy of general rigid-body systems Kinematics: r = rC + r v = vC + r

(38)

Kinetic energy: T = 1 m vC vC 2 +
1 2 m

( ri ) ( ri ) dm

(39)

For a general point P that is different from point C, we have

T = 1 m v P v P + v P ( r PC ) 2 +
1 2 m

( r ) dm
2

(40)

where (x, y, z)-components of

1 2

1 2

( r )2 dm are given by

( r )2 dm = I P {}
m

= {} = IP =

0 z y z 0 x y x 0 x y z Ix x Ix y Ix z Ix y I yy I yz Ix z I yz Ix z P

(41)

where subscript P denotes that the rotatioanl moments of inertia are computed with respect to point P.

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 05: January 31 Energy Principles for Dynamics

Principle of Virtual Work for Statics For statics a la Meirovitch:

N

Fi ri = 0
i=1

(1)

where {Fi } are impressed forces, and {} designates the virtual character of the instantaneous variations, as opposed to the differential symbol {d} designating actual differentials of positions {ri } taking place in the time interval {dt}.

Principle of Virtual Work for Dynamics For dynamics a la dAlembert:

N

(Fi m i ri ) ri = 0
i=1

(2)

where { i } is the acceleration of particle m i . r

Generalized Coordinates Virtual displacements ri may be expressed in terms of the generalized virtual displacements qk (k = 1, 2, 3, ..., n): ri ri = qk , i = 1, 2, 3, ..., N qk k=1 An example: r = xi + yj + L(cos i + sin j) (4)
n

(3)

Generalized Coordinates-contd Observe that there are three independent variables, (x, y, ). r r r x + y + r = x y = xi + yj + L( sin i + cos j) (5)

Note that the dimensional unit of (x, y) is meter(m) whereas that of is radian, suggesting that the units of generalized coordinates can be different.

Interpretation of Equation (2) Lets revisit the formulation of the equations of motion for the spring-mass-bar problem studied in the previous lecture. To this end, rst, we must obtain the virtual displacements.

A Spring-Mass-Bar System:
y, j

oA

mg
x, i kx A o YAC N A

X AC m x

x
K

m
o A

YAC XAC A

C C Mg

(ML2/12) Mg B N F

M r C

Virtual displacements: Virtual displacement for mass m: r A = x i Virtual displacement for bar M: rC = r A + r A C L r A C = (cos i + sin j) 2 (7) (6)

Virtual displacements-contd Virtual rotation for bar M: Here, one utilizes the angular velocity = k to obtain the virtual rotation = k In general one has = x i + y j + z k (9) (8)

Acceleration Vectors: Acceleration vector for mass m: a A = v A = xi Acceleration vector for bar M: aC = a A + a A C L (cos i + sin j) aA C = 2 L2 (sin i cos j) 2 (10)

(11)

Equilibrium equation for mass m:

mg
kx
A o YAC N A

X AC m x

f A = kxi m xi mgj + N A j + X AC i + Y AC j = 0 (12)

Equilibrium equation for bar M:

YAC XAC A

C (ML2/12) Mg

M r C

fC = Fi MaC Mgj X AC i Y AC j = 0 MC +rC

B

(13)

M L 2 = k + rC 12 Fi = 0

(X AC i Y AC j) (14)

An Application of DAlemberts Principle for the spring-mass-bar problem (Fi m i ri ) ri = f A r A + fC rC + MC C = 0 Note that f A r A = 0 and so other two terms! (15)

{kxi m xi mgj + N A j + X AC i + Y AC j} r A + {Fi MaC Mgj X AC i Y AC j} rC M L 2 k + rC B Fi { 12 + rC A (X AC i Y AC j)}

(16)

The rst term in (16) becomes f A r A ={kxi m xi mgj + N A j + X AC i + Y AC j} (xi) f A r A ={kx m x + X AC } x Remark: Observe that the reactions forces Y AC and N A have played no role in the resulting virtual work!

(17)

The second term in (16) becomes fC rC = {Fi MaC Mgj X AC i Y AC j} rC L (cos i + sin j) = {Fi M[xi + 2 L2 (sin i cos j)] 2 Mgj X AC i Y AC j} L {xi + (cos i + sin j) } 2

(18)

 L2 L cos sin ] fC rC = {F M[x + 2 2 X AC } x L2 L cos sin ] + {F M[x + 2 2 L X AC } cos 2 L L2 + {M[ sin cos ] Mg 2 2 L Y AC } sin 2

(19)

 L2 L cos sin ] fC rC = {F M[x + 2 2 X AC } x ML M L2 L cos F cos x +{ 2 2 4 L M Lg cos X AC sin 2 2 L sin Y AC } 2 (20)

In evaluating the third term, rst, carry out: rC B Fi = L (sin i cos j) Fi 2 FL = cos k 2 rC A (X AC i Y AC j) = L L { cos X AC + sin Y AC }k 2 2

(21)

(22)

Finally, we obtain: MC C = FL M L 2 + cos { 12 2 L L + [ cos X AC + sin Y AC ]} 2 2

(23)

Combine (17), (20) and (23) to obtain ML {(M + m)x + kx + cos 2 ML F} x sin 2 2 (24) ML M L2 x + cos x +{ 3 2 MgL + sin L cos F} = 0 2

Since x and are arbitrary, we obtain: ML cos (M + m)x + kx + 2 ML sin 2 = F 2 M L2 ML MgL + cos x + sin 3 2 2 = L cos F

(25)

(26)

Observations 1. When using Newtons second law, Eqs.(25) and (26) are obtained by eliminating the reaction forces X AC and Y AC . 2. On the other hand, one does not need to consider reaction forces in applying dAlemberts principle! Only apparent forces including inertia forces need to be considered. This is a major advantage of applying energy principles over Newtons method.

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 06: 02 February 2006 Hamiltons Principle and Euler-Lagranges Equations

Lets begin with dAlemberts principle:

N

(Fi m i ri ) ri = 0
i=1

T + W W =
N N i=1

d (m i ri ri ) = 0 dt

(1)

Fi ri , Ti = d( 1 m i ri r) 2
i=1

Lets integrate (1) in time:

t2 t1

{T + W

N i=1

d (m i ri ri )} dt = 0 dt

(2)

t2 t1

{T + W } dt =

t2 N t1 i=1

d (m i ri ri ) dt dt

(3)

Noting that we have

t2 N t1 i=1 N

d (m i ri ri ) dt dt

(4)

=
i=1

(m i ri ri )|t2 t1

so that, if ri (t1 ) and ri (t2 ) are specied, we have ri (t1 ) = ri (t2 ) = 0 (5)

Hence, equation (2) becomes

t2 t1

{T + W } dt = 0

(6)

which is known as extended Hamiltons principle.

Hamiltons Principle for Conservative Systems In general the work done, W , consists of two parts: W = Wcons + Wnoncons (7)

where subscripts cons and noncons designate conservative and nonconservative systems, respectively.

Hamiltons Principle for Conservative Systems For conservative systems, we have have from (6) with Wnoncons = 0:
t2 t1

{T + Wcons } dt = 0

(8)

which is known as Hamiltons principle for conservative systems.

Action Integral for Conservative Systems Observe that the work done on the conservative systems can be expressed in terms of the corresponding potential energy:
N

Wcons = V,

V =
i=1

(
r

rr e f

Fi dri )

(9)

Substituting (9) into (8), one obtains:

t2 t1

{T V } dt = 0

t2 t1

{T V } dt = 0
t2 t1

(10)

I =

{L} dt = 0,

L =T V

Euler-Lagrange Equations of Motion Lets substitute the general work expression (7) into the extended Hamiltons principle (6) to obtain:
t2 t1

{T + Wcons + Wnoncons } dt = 0

(11)

which, with (9) together with L = T V , becomes

t2 t1

{L + Wnoncons } dt = 0

(12)

Since L can be expressed in terms of the generalized coordinates as L L L = { qk + qk } qk qk k=1 one obtains
t2 t1 n

(13)

L dt =

t2 t1

L L { qk + qk } dt qk qk k=1

(14)

Integrating by part the rst term of the preceding integral, we obtain

t2 t1 n L L { qk } dt = { qk }|t2 t1 qk qk k=1 k=1 t2 t1 n

d L { ( )qk } dt dt qk k=1

(15)

Since qi (t1 ) = q(t2 ) = 0, we have

t2 t1

L { qk } dt qk k=1 d L { ( )qk } dt dt qk k=1

n

t2 t1

(16)

Substituting (16) into (14), then introducing the resulting expression into (12), we nally obtain:
t2 t1

d L L { ( )+ + Q k }qk dt dt qk qk k=1 Wnoncons =

n

(17)

Q k qk
k=1

Since qk are arbitrary, we obtain: L d L ( ) = Qk dt qk qk (18)

which is called Euler-Lagranges equations of motion.

A Spring-Mass-Bar System for Euler-Lagranges Equations

y, j

x
K
A

m
o A x, i

y C B F

Mg
B x B

Notice how simple the problem description becomes now!

Kinematics: Position vectors of mass M, bar at C, and B where the nonconservative force F is applied: r A = xi L rC = xi + (sin i cos j) 2 r B = xi + L(sin i cos j) L rC = j 2 (19)

Kinetic energy: T = 1 m( A r A ) + 1 M( C rC ) + 1 IC 2 r r 2 2 2 1 2 1 cos + M L 2 2 ] = 2 [(M + m)x + M L x 3

(20)

Potential energy: Vspring =

x 0

(kxi) dr A = 1 kx 2 2 (Mgj) drC (21)

Vgravit y =

rC

rC

L = Mg (1 cos ) 2 Total potential energy: V = Vspring + Vgravit y (22)

 Nonconservative work Wnoncons : Wnoncons = F B r B = Fi [xi + L(cos i + sin j)] = F(x + L cos ) Q x = F, Q = F L cos

(23)

Note that there are only two state variables: x and in the system kinetic energy, potential energy and the nonconservative work.

Euler-Lagranges Equations d L ( ) dt x d L ( ) dt L = Qx = F x L = Q = F L cos

(24)

Euler-Lagranges Equations d [(M + m)x + 1 M L cos ] + kx = F 2 dt ML d (2L + 3x cos ) 6 dt + 1 M L(x + g) sin = F L cos 2 (25)

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 07: 07 February The Method of Lagranges Multipliers

The Euler-Lagranges equations of motion are obtained from the system energy that consists of the kinetic energy, potential energy and virtual work due to nonconservative forces. In other words, while enabling the derivation process simpler, the Euler-Lagrange formalism eliminated an important information: the reaction forces on joints and constraints (such as boundaries). Often, both the designers and analysts need to know joint force levels so that necessary joint articulations can be carried without jeopardizing the system safety. A case in point is the human joints which, repeadedly over loaded, can casue bone deformations and arthritis. The question is: How do we re-introduce the constraint forces (reaction forces) within the Euler-Lagrange formalism? This was developed by Lagrange and described in his book, M canique Analytique, e published in 1788. This is accomplished as follows. Lets consider holonomic cases, viz., the constraint conditions that can be expicitly stated in terms of position vectors, and consider a two-dof

spring-mass system shown below. Suppose we would like to know the reaction force between mass m 1 and spring k2 , and the reaction force between spring k1 and the attached boundary. We now proceed with the procedure that leads to the determaination of the reaction forces.

x1
k1
m1

x2 k2 m2

f1
(a) Assembled system

f2
x2 x3 k2 m2

x1 xg x0
k1
m1

Subsystem 1

f1
(b) Partitioned system

Subsystem 2

f2

A procedure for the method of Lagranges multipliers Step 1: Step 2: Step 3: Step 4: Partition the system into completely free subsystems. Identify the conditions of constraints between the completely free systems. Construct the energy of each of the completely free subsystems. Obtain the total energy by summing the energy of each of the completely free subsystems. Append the conditions of constraints by multiplying each with an

Step 5:

unknown coefcient, (multiplier), to the total system energy (kinematic and potential).

Reference solution based on the assembled system: L =T V T = 1 m 1 x1 + 1 m 2 x2 2 2 2 2 V = + x1) W = f 1 (t)x1 + f 1 (t)x2

2 1 k1 x 1 2 1 2 k2 (x 2 2

(1)

Hence, we obtain via the Euler-Lagrange formalism: m 1 x1 + (k1 + k2 )x1 k2 x2 = f 1 (t) m 2 x2 + k2 x2 k2 x1 = f 2 (t) (2)

Application of the method of Lagranges multipliers Step 1: Done in the preceding gure. Step 2: Conditions of constraints: Between k1 and the xed end : 0g = x0 x g = 0 Between m 1 and k2 : 13 = x1 x3 = 0 Step 3: Energy of two completely free subsystems For subsystem 1: 2 T1 = 1 m 1 x1 2 V1 = 1 k1 (x1 x0 )2 2 W1 = f 1 (t)x1 (4) (3)

For subsystem 2: 2 T2 = 1 m 2 x2 2 V2 = 1 k2 (x2 x3 )2 2 W2 = f 2 (t)x2 Step 4: Sum the total system energy T = T1 + T2 V = V1 + V2 W = W1 + W2 (5)

(6)

Step 5: Construct the constraint functional = 0g 0g + 13 13 = (x0 x g ) 0g + (x1 x3 ) 13 Total Lagrangian of the partitioned system: L = (T1 + T2 ) (V1 + V2 ) (x g , x0 , x1 , x3 , 0g , 13 ) (8) (7)

Partitioned equations of motion: x1 term: x2 term: x3 term: x0 term: 0g term: 13 term: m 1 x1 + k1 x1 k1 x0 + 13 = f 1 (t) m 2 x2 + k2 x2 k2 x3 = f 2 (t) k2 x2 + k2 x3 13 = 0 k1 x1 + k1 x0 + 0g = 0 (x0 x g ) = 0 (x1 x3 ) = 0

(9)

Matrix form of the partitioned equations of motion m 1 D 2 + k1 0 0 k1 0 1 where D 2 =

d2 . dt 2

0 m 2 D 2 + k2 k2 0 0 0 k2 k2 0 0 1

k1 0 0 k1 1 0

|0 1 x1 f 1 |0 0 x2 f 2 x3 0 |0 1 |1 0 x0 = 0 | |0 0 og x g |0 0 13 0

(10)

Symbolically, the above equation can be expressed as A CT C 0 x = f xb (11)

Hence, solution of (11) provides both the displacement x and the reaction forces . Question: Can one obtain the equations of motion for the assembled system (2)? The answer is yes! To this end, we observe the following relation between the partitioned degrees of freedom and the assembled ones as 1 0 0 x1 x1 x2 0 1 0 = (12) x = Lxg x2 1 0 0 x3 x0 0 0 1 x0 Substituting the above assembling operator L into (11), we obtain

LT A L LT C CT L 0

xg

LT f = xb (13)

Simple matrix multiplications show LT C = K = LT AL = 0 0 0 0 1 0 m 1 D 2 + k1 + k2 k2 k1

k2 m 2 D 2 + k2 0

k1 0 k1

(14)

Therefore, (13) reduces to:

m 1 D 2 + k1 + k2 k2 k1 0

k2 m 2 D 2 + k2 0 0

k1 0 k1 1

|0 x1 f 1 |0 x2 f 2 |1 x0 = 0 | |0 og xg

(15)

Finally, invoking the bounday condition x0 = x g = 0 we arrive at m 1 D 2 + k1 + k2 k2 k2 m 2 D 2 + k2 x1 x2 = f1 f2 (17) (16)

Note that solution of (16) provides the reaction force at the left-end support whereas (17) does not!

Application of the method of Lagranges multipliers

y, j

m
o A C y C x, i

Mg
B x C

y, j A o A C C y C x, i

y, j

m
o A x, i L

mg
B Subsystem 1 x C Subsystem 2

Mg

Kinematics: Kinematics of Subsystem 1: r A = xi + yj v A = xi + y j Kinematics of Subsystem 2: rC = xC i + yC j L r A = rC (sin i cos j) 2 L r B = rC + (sin i cos j) 2 (18)

(19)

Step 1; Partitioning is done in the preceding gure! Step 2: Conditions of constraints: (1) Between mass m and the ground: (y yg ) j = 0 (2) Joint at A: rA rA = 0 L (x xC + sin ) i 2 + (y yC L cos ) j = 0 2

(20)

Step 3: Construct the energy of completely free two subsystems. Energy of Subsystem 1: T1 = 1 m(x 2 + y 2 ) 2 V1 = mgy + Energy of Subsystem 2: 2 2 T2 = 1 M(xC + yC ) + 1 IC 2 2 2 V2 = Mg yC (22)
1 kx 2 2

(21)

 Nonconservative work Wnoncons : Wnoncons = F B r B L = Fi [xC i + yC j + (cos i + sin j) ] 2 L = F(xC + cos ) 2 Q x = F, Q = F L cos
B

(23)

Step 4: The total energy of the system: T = 1 m(x 2 + y 2 ) 2

+ 1 M(xC + yC ) + 1 IC 2 2 2 2 2

(24)

V = mgy + 1 kx 2 + Mg yC 2 Step 5: Construct the constraint functional: = (y yg ) 0g + (x xC + (y yC L + sin ) Ax 2

L cos ) Ay 2

(25)

Partitioned equations of motion via the method of Lagranges multipliers

x : m x + kx + Ax = 0 M xC Ax = F M yC + Mg Ay = 0 L L IC + cos Ax + sin Ay = F L cos 2 2 (y yg ) = 0 L (x xC + sin ) = 0 2 L (y yC cos ) = 0 2 y : xC : yC : : 0g : Ax : Ay : m y + mg + 0g + Ay = 0

(26)

Note that (0g , Ax , Ay ) corrrespond to the reaction forces (N A , X AC , Y AC ) that are obtained via Newtons second law, viz., in equation (10) of Lecture 04 Slides.

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 8: 21 February Vibration of String, Bar and Shaft

z w x

f(x,t)

dx

w(x,t)

f (x, t)dx
ds
w(x,t) x

T (x) + T (x) dx x
(x + dx, t) =
w( L ,t) x

(x, t) = x

w(x,t)

dx x x+ dx

2 w(x,t) x 2

dx

T ( L)

K w(L,t)

T (x)

String in transverse vibration

Formulation via Hamiltons principle:

t2 t1

[L + Wnoncons ]dt = 0, L = T V
L 0 1 (x)w2 2

T = V =
0

dx

(1)

T (x)(ds d x) + 1 K w 2 (L , t) 2

From the preceding gure, we nd ds =

1 2 [1 + wx (x, t)] 2 2 d x [1 + 1 wx (x, t)]d x 2

(2)

Hence, substituting (2) into (1),we obtain V =

0 L 1 2 T (x) 2 wx (x, t) d x + 1 K w 2 (L , t) 2

(3)

Transverse displacement: w(x, t) (m) Time derivative: w = w (m/s) dt Spatial derivtive: wx = w (m/m) dx String tension: T (x) (N ) Mass per unit string length: (x) (kg/m) String length: ds (m) End spring : K (N /m)

Since there are impotant differences between the variations of discrete model vs. continuum model, lets perform L below. L = T V T = {
0 L 1 (x)w2 (x, t) 2 1 2 T (x)

d x} (4)

V = {
0

2 wx (x, t) d x}

+ { 1 K w 2 (L , t)} 2

Evaluation of

t2 t1

T dt
L 1 (x)w 2 (x, t) 2

T = { =
0 0 L

d x} d x} (5)

1 (x){w 2 (x, t)} 2

=
0

(x){w(x, t) w(x, t)} d x

Evaluation of

t2 t1

T dt - contd

With T , lets perform the denite time integral of T :

t2 t1

T dt =

t2 t1 0

(x){w(x, t) w(x, t)} d xdt

(6)

Evaluation of
t2 t1

t2 t1 t2 t1 L 0

T dt - contd
L 0

T dt = = =
0

(x){w(x, t) w(x, t)} d xdt (x)w(x, t) w(x, t) dt}d x (7)

t2 w(x, t)]t1

t2 t1

{ [(x)w(x, t)

t2 t1

(x)w(x, t) w(x, t) dt] }d x

Evaluation of

t2 t1

T dt - concluded

Since w(x, t1 ) and w(x, t1 ) are considered known everywhere over the spatial domain 0 x L, we have w(x, t1 ) = w(x, t1 ) = 0 so that (
t2 t1 t2 t1

(8)

T dt) reduces to
t2 t1 0 L

T dt =

(x)w(x, t) w(x, t)d xdt

(9)

Evaluation of

t2 t1

V dt

Since the potential energy V does not involve time derivative of w(x, t), it is adequate to carry out simply V , viz.: V = =
0 0 L L 1 2 T (x) 2 wx (x, t) d x + 1 K w2 (L , t) 2

T (x) wx (x, t)wx (x, t) d x

(10)

+ K w(L , t)w(L , t)

Evaluation of

t2 t1

V dt - contd

The rst term in the previous equation becomes:

L 0

T (x) wx (x, t)wx (x, t) d x

w(L ,t) wx (x, t)w(x, t)]|w(0,t)

= [T (x)
0 L

(11)

T (x) wx x (x, t)w(x, t) d x

Evaluation of
L 0

t2 t1

V dt - concluded

T (x) wx (x, t)wx (x, t) d x (12)

= [T (x) wx (x, t)w(x, t)]x=L [T (x) wx (x, t)w(x, t)]x=0

0 L

T (x) wx x (x, t)w(x, t) d x

Virtual work due to nonconservative force f (x, t) Wnoncons =

0 L

f (x, t)w(x, t)d x

(13)

Substituting (13), (12), (10) and (9) into (1), one obtains Hamiltons principle:
t2 t1

[(T V ) + Wnoncons ]dt = 0

(1)

Hamiltons principle for string

t2 t1 0

{[(x)w(x, t) (14)

+ T (x) wx x (x, t) + f (x, t)] w(x, t)}d x [(T (x) wx (x, t) + K w(x, t))w(x, t)]x=L + [T (x) wx (x, t)w(x, t)]x=0 dt = 0

In order for the above variational expression to hold, rst the term inside the brace {.} must vanish, which yields the governing equation of motion: (x)w(x, t) = T (x) wx x (x, t) + f (x, t) (15)

Similarly, the two terms in the bracket must also vanish: [(T (x) wx (x, t) + K w(x, t))w(x, t)]x=L [T (x) wx (x, t)w(x, t)]x=0 (16) (17)

Boundary conditions for continuum string Two types of boundary conditions: Essential (or geometric) boundary conditions: displacements are specied. when

Natural (or force) boundary conditions: when the boundary conditions come from force (moment) balance considerations.

Boundary conditions at x = L: From (16) we nd T (L) wx (L , t) + K w(L , t) = 0 or w(L , t)] = 0 (18) (19)

Since the right end is left to move, (19) is not appropriate. Hence, the correct boundary conditon is given by (18), which is a natural boundary condition.

Boundary conditions at x = 0: From (17) we nd T (0) wx (0, t) = 0 or w(0, t) = 0 (20) (21)

Since the right end is xed, (20) is not appropriate. Hence, the correct boundary conditon is given by (21), which is an essential or geometric boundary conditon.

Boundary conditions - concluded Observe that the variational equation, viz., Hamiltons principle (14) provides the information to determine the boundary conditions, depending on the end conguration or constraint condition. This is a distinct advantage of the variational formulation of continuum models as opposed to Newtons approach.

Analogy: Torsional shaft and axial Rod

String Variable w(x, t) (m) Stiffness T (x) (N ) Shaft (x, t) (rad) G J (x)(N m 2 ) Axial bar u(x, t) (m) E A(x) (N )

Hence, the equations of motion for bar and shaft can be derived by appropriate parameter changes.

Strings in transverse vibration

Equation (15) with f (x, t) = 0 offers the continuum eigenvalue problem for strings together the boundary conditions (18) and (21). To this end, we seek the solution of the homogeneous equation of (15) in the form of

w(x, t) = W (x) F(t)

(22)

by employing the separation of variable used for the solution of partial differential equations. Substituting (22)

into (15), (18) and (21), we obtain d dW (x) [T (x) ]F(t) dx dx d 2 F(t) = (x)W (x) , 0<x <L 2 dt dW (x) [T (x) + K W (x)]|x=L F(t) = 0 dx [W (x)]|x=0 F(t) = 0 Applying the temporal part F(t) in the form of F(t) = F e jt , F = 0 we obtain the following boundary-value problem set: (24)

(23)

d dW (x) [T (x) ] + 2 (x)W (x) = 0, 0 < x < L dx dx dW (x) + K W (x)]|x=L = 0 [T (x) dx [W (x)]|x=0 = 0

(25) (26) (27)

When the tension T (x) and mass (x) are constant along the string length, equation (25) can be rearranged as d 2 W (x) + 2 W (x) = 0, dx2 2 = 2 /T, 0 < x < L whose general solution is known as W (x) = C1 sin x + C2 cos x (29)

(28)

Substituting the preceding solution into the (x = L)boundary condition (26) yields: T{C1 cos L C2 sin L} + K {C1 sin L + C2 cos L} = 0 (30)

Similarly, the (x = 0)-boundary condition (27) yields: C1 sin( 0) + C2 cos( 0) = 0 With C2 = 0, (30) reduces to T cos L + K sin L = 0 Fixed end at x = L: When K , viz.,
T K

C2 = 0

(31)

(32)

0, we have
T

sin L = 0 =

n , n = 1, 2, ... = L

(33)

So that we obtain the frequency equation for the xed rightend case as k k = L Free end at x = L: When K 0, we have cos L = 0 =
T

T , k = 1, 2, ...

(34)

(2k 1) , k = 1, 2, ... (35) = 2L

So that we obtain the frequency equation for the xed

rightend case as (2k 1) k = 2L T , k = 1, 2, ...

(36)

Note on the derivation of the string equation from elasticity theory

Equilibrium: i j + ( fj u j ) = 0 xi Constitutive law: i j = ci jk ek

(37)

where i j is the Cauchy stress, ei j is the Eulerian strain, f j is the body force, and ci jk is the constitutive constant.

Lagrangian strain-displacement relation: u j 1 u i 1 u k u k Ei j = 2 ( + )+ 2 , k = 1, 2, 3. xj xi xj xj where x j refers to the undeformed conguration.

(38)

Hamiltons principle for continuum:

t2 t1

[L + Wnoncons ]dt = 0, L = T U
1 (u 2 2 V0 1 e 2 ij ij V

T = U= U=
V0

+ v 2 + w 2 ) d V0 (39) d V, (i, j) = 1, 2, 3. d V0 , (i, j) = 1, 2, 3.

1 S E 2 ij ij

where Si j is the second Piolar-Kirchhoff stresses; V and V0 refer to deformed and undeformed volume, respectively.

A continuum model for strings: Sx x = E E x x

2 E x x = u x + 1 (u 2 + wx ), u x = x 2

u , etc. x

T =
V0 L

1 (u 2 2

+ w 2 ) d V0 d A0
A0

=
0

(x) (u 2 + w2 ) d x, (x) =
1 S E 2 xx xx V0 L

(40)

U= =
0

d V0

1 2

2 E A[(u x + 1 (u 2 + wx )]2 d x x 2

Approximation of the string strain energy:

2 u 2 << wx x

(41)

so that one arrives at U=

0 1 2 2 E A(u x + 1 wx )2 d x 2

=
0

1 2

2 4 E A(u 2 + u x wx + 0.25wx ) d x x

(42)

1 2

2 E A(u 2 + u x wx ) d x x

Ignoring the axial part, we have

L

U
0 L

1 T (x) 2

2 wx d x,

T (x) = E Au x (43)

T
0

1 (x)w2 2

dx

which yields via Euler-Lagranges formalism the same governing equation of motion (15) together with the two boundary conditions (16) and (17).

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 9: 23 February Cables with End Masses and End Springs

z w(x,t)
M0

f(x,t)

ML

K0
O

Cables with End Masses and End Springs

Formulation via Hamiltons principle:

t2 t1 L 0

[L + Wnoncons ]dt = 0, L = T V

T = + V =
0

1 (x)w2 (x, t) 2

dx
1 2 ML

1 2 M0 L

w (0, t) +
2

w (L , t)
2

(1)

1 2 T (x)

2 wx (x, t) d x

+ 1 K 0 w2 (0, t) + 1 K L w2 (L , t) 2 2

Symbol Denitions: Transverse displacement: w(x, t) (m) Time derivative: w = w (m/s) dt Spatial derivative: wx = w (m/m) dx String tension: T (x) (N ) Mass per unit cable length: (x) (kg/m) End springs: (K 0 , K L ) (N /m) End masses: (M0 , M L ) (N /m)

Evaluation of

t2 t1 t2

T dt

T dt =
t2 t1

t1

{[
0

(x)w(x, t) w(x, t)d x]

(2)

+ M0 w(0, t) w(0, t) + M L w(L , t) w(L , t)} dt

Evaluation of V V =
0 L

T (x) wx x (x, t)w(x, t) d x (3)

+ [T (x) wx (x, t)w(x, t)]x=L [T (x) wx (x, t)w(x, t)]x=0 + [K L w(x, t)w(x, t)]x=L + [K 0 w(x, t)w(x, t)]x=0

Virtual work due to nonconservative force f (x, t) Wnoncons =

0 L

f (x, t)w(x, t)d x

(4)

Substituting (4), (3) and (2) into (1), one obtains Hamiltons principle:
t2 t1

[(T V ) + Wnoncons ]dt = 0

(1)

Hamiltons principle for cable with end masses and end springs
t2 t1 0 L

{[(x)w(x, t)

+ T (x) wx x (x, t) + f (x, t)] w(x, t)}d x {[(T (x) wx (x, t) + K L w(x, t) + M L w(x, t)]w(x, t)}x=L {[T (x) wx (x, t) + K 0 w(x, t) + M0 w(x, t)]w(x, t)}x=0 dt = 0 (5)

The governing equation of motion: (x)w(x, t) = T (x) wx x (x, t) + f (x, t) The boundary conditions: {[(T (x) wx (x, t) + K L w(x, t) + M L w(x, t)]w(x, t)}x=L = 0 {[T (x) wx (x, t) + K 0 w(x, t) + M0 w(x, t)]w(x, t)}x=0 = 0 (7) (6)

(8)

Observations: 1. The governing equation of motion for the case of end masses and end springs is the same as the case without

any end conditions. This means the governing equation of motion is the same for all possible boundary conditions. 2. Therefore, the general form of solution w(x, t) = F(t) [C1 sin x + C2 cos x] is applicable to all the cable vibration problems. (9)

Boundary conditions at x = L: From (7) we nd T (L) wx (L , t) + K L w(L , t) + M L w(L , t) = 0 or w(L , t) = 0 (10) (11)

Boundary conditions at x = 0: From (8) we nd T (0) wx (0, t) + K 0 w(0, t) + M0 w(0, t) = 0 or w(0, t) = 0 (12) (13)

From (10)-(13), one nds that the two essential boundary conditions given by (11) and (13) are special cases of the two natural boundary conditions given by (10) and (12), since the former two are obtained from the latter two if K L and K 0 , respectively. Substituting (9) into (10) and (12), together with F(t) = Fe jt , yields the following characteristic equation:

cos + ( L L 2 ) sin

sin + C1 0 ( L L 2 ) cos = 2) (0 0 C2 0

(14)

, = L T KL L K L L = , 0 = 0 T T ML M0 , 0 = L = L L =

(15)

The characteristic equation of (14) is found by taking the determinant to be zero: (0 0 2 )( L L 2 ) sin + cos [(0 0 2 ) + ( L L 2 )] 2 sin = 0

(16)

We now examine several special cases in the following.

1. Fixed-Fixed Ends (k L , k0 ) This case corresponds to (k L , k0 ) so that, by dividing (16) by (k L k0 ) yields: sin = sin L = 0 k k = L L = k, k = 1, 2, ... (17)

T , k = 1, 2, 3, ...

which was already discussed previously.

2. Fixed-Free Ends (k L 0, k0 , L = 0 = 0) This case leads to the following characteristic equation from (16): cos = sin L = 0 (2k 1) , k = 1, 2, ... L = 2 (2k 1) k = 2L T , k = 1, 2, 3, ...

(18)

which was also treated before.

3. Flexible Supports (k L = k0 = 0, L = 0 = 0) This case leads to the following characteristic equation from (16): (k 2 2 ) tan + 2k = 0, k = k L = k0 (19)

Other general cases can be evaluated using (17). This is left for exercises

Mode shape of the first natural frequency of cable with flexible supports 1

0.9

0.8

0.7 mode shape

0.6

0.5

0.4 frequency in Hertz = 0.1039 spring k = 10 beta*L = 2.6277 0.3

0.2

0.1

0.2

0.3

0.4

0.5 cable span

0.6

0.7

0.8

0.9

Mode Shapes of Cable with End Springs

Mode shape of the first natural frequency of cable with flexible supports 1

0.9

0.8

0.7

0.6 mode shape

0.5

0.4

0.3

0.2 frequency in Hertz = 0.1242 spring k = 10000 beta*L = 3.141

0.1

0.1

0.2

0.3

0.4

0.5 cable span

0.6

0.7

0.8

0.9

Mode Shapes of Cable with End Springs

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 10: 28 February Vibration of Continuum Beams

Summary of Elementary Beam Theory

Strain-displacement: +

xx

Constitutive relation: x x = E

1 w(x, t) 2 ) 2( x

u(x, t) 2 w(x, t) = z x x2 (1)

xx

2 w(x, t) Bending moment: M y = zx x d A = E Iz x2 A

Kinetic energy:
L 0

T =

1 2

w(x, t) m(x) (wt ) d x, wt = t

2

(2)

Potential energy of the beam: V = = =

1 2 1 2 1 2 L

[
0 L A

x x E(
A

x x d A]d x

[
0 L

)2 d A]d x xx

[
0 L A

2 2 E{(z 2 wx x + u x wx )

4 2 + (u 2 + 0.25wx 2zu x wx x zwx x wx )}d A]d x x

(3)

1 2 1 2

[
0 L 0 A

2 2 E(z 2 wx x + u x wx )d x

{ E I (x) (wx x )2 + P(x) (wx )2 } d x w(x, t) wx = , etc. x

P(x) = E A(x) u x ,

Hamiltons principle:
t2 t1

[T V + W ] dt = 0

(4)

External energy: W =
0 L

f(x, t) w(x, t) d x

(5)

Variation of the Kinetic energy:

t2 t1

T dt =
0 L 0 t2 t1 L 0

m(x) [w(x, t)t w(x, t)]|t2 d x t1 (6)

m(x) w(x, t)tt w(x, t) dt d x

t2 t1

m(x) w(x, t)tt w(x, t) dt d x

since we consider w(x, t1 ) and w(x, t2 ) as specied or constant at t = t1 and t = t2 .

Variation of the Potential energy:

L V = { E I w(x, t)x x w(x, t)x }|0 {[E I w(x, t)x x ]x L + [P w(x, t)x w(x, t) }|0

+
0

(7)

{ [E I w(x, t)x x ]x x

[P w(x, t)x ]x }w(x, t) d x Substituting (5)-(7) into Hamiltons principle (4), we obtain the desired variational equation.

t2 t1 L 0

{ m(x) w(x, t)tt + [E I w(x, t)x x ]x x [P w(x, t)x ]x f (x, t)}w(x, t) d x (8)

L + {E I w(x, t)x x w(x, t)x }|0 L {([E I w(x, t)x x ]x + P w(x, t)x )w(x, t) } |0 dt = 0

As in the case of string equations, the rst term provides the governing continuum partial differential equation for the beam, the last two expressions yield the boundary conditions.

Boundary-value problem for beam bending with P = 0 Governing equation of motion: m(x) w(x, t)tt + [E I w(x, t)x x ]x x = f (x, t) Moment and/or Slope Boundary Conditions:
L {M(x, t)w(x, t)x }|0

= 0, M(x, t) = E I w(x, t)x x

(9)

Shear and/or Displacement Boundary Condition:

L {V (x, t)w(x, t) } |0 = 0, V (x, t) = [E I w(x, t)x x ]x

Boundary-value problem for beam bending - contd Unlike string problems, we need two conditions at each end with x = c being the end coordinate. Free end at x = c: M(c, t) = 0 and V (c, t) = 0 Fixed end at x = c: w(c, t) = 0 and w(c, t)x = 0 Simply supported end at x = c: M(c, t) = 0 and w(c, t) = 0 (12) (11) (10)

Classical Solutions of Beam Bending Vibration Solution of the homogeneous governing equation assumes the form w(x, t) = F(t)W (x), with F(t) = Fe jt (13)

Remark (a brief tour of hyperbolic partial differential equation theory): In a rigorous treatment, one substitutes (13) into the governing equation with f (x, t) = 0 to obtain m(x)F(t)tt W (x) + E I (x)F(t)W (x)x x x x = 0 Dividing both sides by F(t)W (x) one obtains F(t)tt E I (x) W (x)x x x x = F(t) m(x) W (x) (15) (14)

Since the left-hand term is a function of time whereas the right-hand term is a function of only x, both terms must be independent of time, t, and the coordinate, x. Hence, both terms must be a constant if the equality is to hold for all time and for the entire spatial domain range:, viz., F(t)tt E I (x) W (x)x x x x = = 2 , F(t) m(x) W (x) being real (why?) (16)

which yields the following two sets of ordinary differential equations: (17) F(t)tt + 2 F(t) = 0 W (x)x x x x 4 W (x) = 0, m(x)2 4 = E I (x) (18)

Solution of (17) can be expressed as F(t) = Fest s + =0

2 2

s1 s2

+j j

(19)

Therefore, F(t) has the form of F(t) = A sin t + B cos t (20)

which implies that w(x, t) oscillates with time with the corresponding period = 2/. Hence, is its characteristic frequency. Solution of (18) assumes the following form: W (x) = C1 sin x + C2 cos x + C3 sinh x + C4 cosh x (21)

Vibration of Beam Bending 1. Free-free beam: For this the boundary condition (10) gives M(x, t) = E I (x)w(x, t)x x = 0 V (x, t) = E I (x)w(x, t)x x x = 0 at x = 0 at x = L {E I (x)W (x)x x }|x=0 = 0 {E I (x)W (x)x x x }|x=0 = 0 {E I (x)W (x)x x }|x=L = 0 {E I (x)W (x)x x x }|x=L = 0 (22)

Substituting W (x) (21) into the above four boundary conditions, we obtain with = L:

C2 + C4 = 0 C1 + C3 = 0 C1 sin C2 cos + C3 sinh + C4 cosh = 0 C1 cos + C2 sin + C3 cosh + C4 sinh = 0 AC = 0

(23)

whose characteristic equation (the determinant of A of the above equation) is given by 1 cos L cosh L = 0 (24)

0 10

Search for initial root locations

-1 10 Caracteristics roots of free-free beam

-2 10

-3 10

-4 10

-5 10 0

10 beta*L

12

14

16

18

20

Initial root nder for a free-free beam

Mode shape of free-free beam 1 2-th mode beta*L = 4.73 0.8

0.6

0.4

Mode shape

0.2

-0.2

-0.4

-0.6

-0.8

0.1

0.2

0.3

0.4

0.5 Beam span

0.6

0.7

0.8

0.9

Second mode of a free-free beam

Mode shape of free-free beam 1 3-th mode beta*L = 7.8532 0.8

0.6

0.4

0.2 Mode shape

-0.2

-0.4

-0.6

-0.8

-1

0.1

0.2

0.3

0.4

0.5 Beam span

0.6

0.7

0.8

0.9

Third mode of a free-free beam

What are the mode shapes of L = 0 ? When L = 0, one can check the rank of the characteristic matrix A reduces from three to two. This is a degenerative case. Hence, the solution form assumed by (21) is not appropriate. Therefore, one must invoke the governing equation (18), which becomes with = 0: W (x)x x x x = 0 (25)

that is subject to the boundary conditions (22). Integration of the above equation yields W (x) = c1 = c2 x + c3 x 2 + c4 x 3 which upon satisfying the boundary conditions (22), yields W (x)x = c2 W (x) = c1 + c2 x (27) (26)

Note that W (x)x = c2 is the slope of the beam and is a constant. This means the free-free beam with zero frequency (beta = 0) rotates as a straight line. This is referred to a rotational rigid-body mode. There is a second rigid body mode given by W (x) = c1 if c2 = 0 (i.e., no rotational rigid body mode). c1 represents a uniform displacement across the beam span. This is known as a translational rigid body mode. This is illustrated below.
Rotational rigid body mode W(x) _x = C2 C2 C1 Translational rigid body mode W(x) = C1

Two rigid body modes of a free-free beam

Beams with other boundary conditions There are ve additional end conditions that are possible from combinations of the three distinct boundary conditions (10) - (12). These are left for your practice.

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 11: 02 March Beams with Uncertain Boundary Conditions

Consider a beam whose two ends are constrained by transverse springs (kw1 , kw2 ) as well as rotational springs (k 1 , k 2 ) as shown in the gure below.
w z Beam with unknown boundary conditions
EI, m(x)

k1 kw1
L

k2
x

kw2

Fig. 1 Modeling of unknown boundary conditions

Before we proceed further, it should be noted that the mathematically known boundary conditions can be realized by taking the limit values of the four spring constants as follows: = kw2 and k 1 = k 2 = 0 = k 1 and kw2 = k 2 = 0 = k 1 and kw2 = k 2 = k 1 = 0 and kw2 = k 2 = 0 (1) For intermediate values of the unknown springs a convenient way of identifying the boundary conditions is to invoke a variational formulation. This is because the appropriate boundary conditions, both natural and essential, are determined as part of variational process. Simply supported : kw1 Cantilever beam x = 0 : kw1 Fixed-xed : kw1 Free-free : kw1

Kinetic energy: T =
1 2 L

m(x)
0

w(x, t)2 d x, t

w(x, t) w(x, t)t = t

(2)

Potential energy of the beam: Vb =

1 2 L 0

{ E I (x) w(x, t)2 x + P(x) w(x, t)2 } d x x x 2 w(x, t) = , 2 x w(x, t) w(x, t)x = x

(3)

w(x, t)x x

where E I (x) is the bending rigidity and P(x) is the pre-stressed axial force.

Potential energy of four unknown springs: Vs = 1 { kw1 w(0, t)2 + k 1 w(0, t)2 x 2 + kw2 w(L , t)
2

+ k 2 w(L , t)2 x

(4)

External energy due to distributed applied force f(x, t), shear forces Q (1,2) and moments M(1,2) applied at both ends: W =
0 L

f(x, t) w(x, t) d x + Q 1 w(0, t) (5)

+ M1 w(0, t)x + Q 2 w(L , t) + M2 w(L , t)x Q 1 = Q(0, t), M1 = M(0, t), Q 2 = Q(L , t) M2 = M(L , t)

(6)

Hamiltons principle:
t2 t1

[T Vb Vs + W ] dt = 0

(7)

Kinetic energy
t2 t1

T dt =
0

t2 t1

m(x) w(x, t)tt w(x, t) dt d x

(8)

Variation of the internal energy of the beam Vb

L Vb = [E I w(x, t)x x w(x, t)x ]|0 L {[E I w(x, t)x x ]x w(x, t)}|0 L 0

(9)

{ [E I w(x, t)x x ]x x [P w(x, t)x ]x }w(x, t) d x

Variation of the potential energy of the unknown boundary forces and moments Vs = { kw1 w(0, t) w(0, t) + k 1 w(0, t)x w(0, t)x + kw2 w(L , t) w(L , t) + k 2 w(L , t)x w(L , t)x (10)

t2

Hamiltons principle:
t1

[T Vb Vs + W ] dt =

+ +

t2 t1 t2 t1 t2 t1 t2 t1 t2 t1

{[E I w(L , t)x x ]x + kw2 w(L , t) Q 2 } w(L , t)dt {[E I w(0, t)x x ]x kw1 w(0, t) + Q 1 } w(0, t)dt {[E I w(L , t)x x ] + k 2 w(L , t)x M2 } w(L , t)x dt {[E I w(0, t)x x ] k 1 w(0, t)x + M1 } w(0, t)x dt
L 0

(11)

{ m(x)w(x, t)tt + [E I w(x, t)x x ]x x

f (x, t) } w(x, t) d xdt = 0

The governing equation of motion: m(x) w(x, t)tt + [E I w(x, t)x x ]x x f (x, t) = 0 The boundary conditions:
{[E I w(L , t)x x ]x + kw2 w(L , t) Q 2 } w(L , t) = 0 {[E I w(0, t)x x ]x kw1 w(0, t) + Q 1 } w(0, t) = 0 {[E I w(L , t)x x ] + k 2 w(L , t)x M2 } w(L , t)x = 0 {[E I w(0, t)x x ] k 1 w(0, t)x + M1 } w(0, t)x = 0

(12)

(13)

Free vibrations of a beam The free vibration of beams can be modeled from (12) and (13) by setting P = Q 1 = Q 2 = M1 = M2 = f (x, t) = 0 (14) so that (12) and (13) are simplied to m(x) w(x, t)tt + [E I w(x, t)x x ]x x = 0 {[E I w(L , t)x x ]x + kw2 w(L , t) } w(L , t) = 0 {[E I w(0, t)x x ]x kw1 w(0, t) } w(0, t) = 0 {[E I w(L , t)x x ] + k 2 w(L , t)x } w(L , t)x = 0 {[E I w(0, t)x x ] k 1 w(0, t)x } w(0, t)x = 0

(15)

Free Vibration Governing Equation and Four Natural Boundary Conditions: m(x) w(x, t)tt + [E I w(x, t)x x ]x x {[E I w(L , t)x x ]x + kw2 w(L , t) } {[E I w(0, t)x x ]x kw1 w(0, t) } {[E I w(L , t)x x ] + k 2 w(L , t)x } {[E I w(0, t)x x ] k 1 w(0, t)x } = 0 =0 =0 =0 =0

(16)

Four natural boundary conditions - contd w(x, t) = W (x)e jt which, when substituted into (16), yields 2 m W (x) + W (x)x x x x EI kw1 W (0) W (0)x x x EI k 1 w(0)x x W (0)x EI kw2 W (L) W (L)x x x + EI k 2 W (L)x w(L)x x + EI = 0, =0 =0 =0 =0 (18) 0x L (17)

The frequency equation of vibrations of a uniform beam can be obtained by assuming the solution of form:
W = c1 sin x + c2 cos x + c3 sinh x + c4 cosh x 2 m 4 = EI Wx = (c1 cos x c2 sin x + c3 cosh x + c4 sinh x) Wx x = 2 (c1 sin x c2 cos x + c3 sinh x + c4 cosh x) Wx x x = 3 (c1 cos x + c2 sin x + c3 cosh x + c4 sinh x) (19)

k 1 3 cos kw2 sin sin k 2 cos where

kw1 3 sin kw2 cos cos +k 2 sin

3 k1 3 cosh kw2 sinh sinh k2 cosh

c 1 c2 =0 3 sinh c3 kw2 cosh cosh c4 k2 sinh kwi = kwi /(E I /L 3 )

kw1

(20)

= L,

ki = ki /(E I /L),

(21)

Free-free beam (kw1 = kw2 = k 1 = k 2 = 0)

3 0 3

0 3 cos 3 sin
sin cos

c1 0 c2 0 =0 3 cosh 3 sinh c3 sinh cosh c4

(22)

1 0 cosh sinh

0 1

0 1 det cos sin

sin cos

=0 sinh
cosh

Cantilever beam (kw2 = k 2 = 0, kw1 , k 1 .) k 1 det cos sin 3 kw1 sin cos 3 k 1 cosh sinh =0 sinh cosh kw1

(23)

1 det cos sin

1 0 sin cos

0 1 cosh sinh

1 =0 sinh cosh

(24)

(1 + cos cosh ) = 0

which gives k = {1.875, 4.694, 7.855, ...} (25)

Fixed end rotational spring parameter for cantilever beam

120

100

rotational spring, k (theta) /(EI/L)

80

60

40

20

0 1

10

Error in fundamental frequency, percent

Effect of xed end rotational spring on fundamental frequency of a cantilever beam

Simply Supported and Fixed-Fixed Beams (w(0) = w(L) = 0 or kw1 = kw2 )

0 1 cos cos +k2 sin 0 k 1 sinh sinh k 2 cosh

k1 det sin
2

sin k cos

k2 sinh

c 1 c2 c3 = 0 cosh cosh c4
1

(26)

2 2 sin sinh + (k1 + k2 ) (sin cosh cos sinh ) + k1 k 2 (1 cos cosh ) = 0

Several ideal cases can be obtained from (26): Simply supported ends: {k1 = 0, k2 = 0} sin sinh = 0 Clamped at x = 0 and simply supported at x = L : {k1 , k2 = 0} tan tanh = 0 Clamped at both ends: {k1 , k2 } 1 cos cosh = 0 When the boundary conditions are not ideal, viz., {0 < k1 < and 0 < k2 < } and k 2 }. (28)

(27)

one needs two or more measured frequencies to determine {k 1

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 12: 07 March 2006 Finite Element Modeling of Vibration of Bars

Lets consider the modeling of a bar with general boundary conditions by the nite element modeling.
Bar with unknown boundary conditions M0
EA, m(x)

ML

k01

k02

k L1

k L2

Theoretical Basis: Same as the continuum bar model! Kinetic energy of the contimuun bar: Tbar =
1 2 L 0

m(x) u t (x, t) d x,
2

u(x, t) u t (x, t) = t

(1)

Potential energy of the continuum bar: Vbar =

1 2 L 0

{ E A(x) u x (x, t)2 } d x

u(x, t) u x (x, t) = x External energy due to distributed applied force f(x, t): Wbar =
0 L

(2)

f(x, t) u(x, t) d x

(3)

Kinetic energy of the discrete masses : 0 L Ts = 1 M0 u 2 (t) + 1 M L u 2 (t) 2 2 Potential energy of four springs: Vs = 1 { k01 u 2 (t) + k02 [u 0 (t) u(0, t)]2 0 2 + k L1 u 2 (t) + k L2 L Hamiltons principle
t2 t1

(4)

[u L (t) u(L , t)]

(5)

[Ttotal Vbar Vs + Wbar ] dt = 0,

Ttotal = Tbar + Ts (6)

Key departure in FEM modeling of beams from continuum modeling: 1. Instead of carrying out the variation (the -process of Hamiltons principle(6) in terms of the continuum variable u(x, t), the energy expressions, (T, V, W, etc.), are approximated on a completely free bar with an arbitrary length , area A, and mass density = m(x). 2. Ideally, the interpolation of the axial displacement u(x, t) over the completely free bar segment, 0 x , is chosen to satisfy the homogeneous differential beam equation, i.e., E A u(x, t)x x = 0 (7)

if possible. If not, it is chosen such that as the length of the bar gets smaller and smaller, it approximately satises the homogeneous equation.

We now consider a completely free, isolated, small segment of a bar.

A completely free element of length taken out from the continuum bar l u1 u2 2 1 an element isolated x

M0

ML

k01

k02

L Bar with unknown boundary conditions

EA, m(x)

x k L1 k L2

Approximation of u(x, t) over the element segment,

<< L: (8)

u(x, t) = c0 (t) + c1 (t)x + c2 (t)x 2 + . . . + cn (t)x n

Class of elements: 1. Linear Element or Constant Strain Element : Retain the rst two terms in (8), viz., (c0 , c1 ) to result in: u(x, t) = c0 (t) + c1 (t)x (9)

2. Quadratic Element or Linear Strain Element : Retain the rst three terms in (8), viz., (c0 , c1 , c2 ) to result in: u(x, t) = c0 (t) + c1 (t)x + c2 (t)x 2 (10)

3. Cubic Element or Quadratic Strain Element : Retain the rst four terms in (8), viz., (c0 , c1 , c2 , c3 ) to result in: u(x, t) = c0 (t) + c1 (t)x + c2 (t)x 2 + c3 (t)x 3 How does one determine the coefcients, (c0 , c1 , c2 , c3 , etc.)?
Bar Elements
Linear (2-Node) Element

(11)

u1
Node 1

u2
o

x=-l /2

l
o u3

Node 2

x= l/2

Quadratic (3-Node) Element

u1
Node 1

u2
Node 2

x=0

x=-l /2

x= l/2

The linear bar element 1. We specify the two discrete nodes, viz., nodes 1 and 2 in the previous gure. Namely, At node 1: At node 2: u( /2, t) = u 1 (t) u( /2, t) = u 2 (t) (12)

so that we have the following two equations: u 1 (t) = c0 (t) + c1 (t) ( /2) u 2 (t) = c0 (t) + c1 (t) ( /2) u(x, t) = u(, t) = ( 1 x/ ) u 1 (t) + ( 1 + x/ 2 2 1 (1 ) u 1 (t) + 1 (1 + ) 2 2 ) u 2 (t) u 2 (t), = 2x/

(13)

The Standard form of Linear (Two-Noded) Bar Element u(, t) = N1 ( ) u 1 (t) + N2 ( ) u 2 (t) N1 ( ) = 1 (1 ) 2 N2 ( ) =
1 2 (1 + )

(14)

1 1 For notational clarity, we express (14) in the form u(, t) = N( ) q(t) N( ) = N1 ( ) N2 ( ) q(t)T = u 1 (t) u 2 (t)

(15)

Construction of Linear Bar Element Elemental kinetic energy:

el Tbar = 1 2 1 2 1 2 1 1

m(x) u 2 (x, t)d x (16) m( ) u 2 (, t)( 1 d ), 2 x=

1 2

1 = 2

Elemental potential energy:

el Vbar = 1 2 /2 /2 1 1

E A(x) u 2 (x, t) d x x 2 (17)

1 = 2

E A( ) ( )2 u 2 (, t) ( 1 d ) 2

Construction of Linear Bar Element Contd

el Tbar

1 1 = m( ) u 2 (, t)( 1 d ) 2 2 1 1 1 1 A q(t)T [NT N] q(t)d = 2 2 1 1 1 T 1 = q [ A NT N d ] q(t) 2 2 1 = 1 q(t)T [mel ] q(t) 2

(18)

where we utilized (15), viz.: u(, t) = N( ) q(t) (19)

Construction of Linear Bar Element Contd

el Vbar

1 = 2 1 = 2 =

1 1 1 1

E A( ) ( )2 u 2 (, t) ( 1 d ) 2 2E A( )
1 T q(t)T [N N ] q(t)d T N N d ] q(t)

(20)

1 q(t)T 2

2E A

= 1 q(t)T [k ] q(t) 2

1 el

Elemental mass matrix, mel bar A m= 6 Elemental stiffness matrix, kel bar k= EA 1 1 1 1 (22) 2 1 1 2 (21)

Approximation of elemental external energy (3) W el =

/2 /2

f(x, t) u(x, t) d x
1 1 el

= q(t)T {

NT f(x, t)

1 2

d }

(23)

= q(t)T {f } fel = {
1 1

NT f(x, t)

1 2

d }

Summary of elemental energy expressions:

el Tbar = 1 {q(t)(el) }T [mel ] {q(t)(el) } 2 el Vbar = 1 {q(t)(el) }T [kel ] {q(t)(el) } 2 el Wbar = {q(t)(el) }T {fel }

(24)

Question: How do we model a long bar with bar elements? Answer: (1) Partition the long bar into small elements. This needs to be explained. (2) Generate the elemental energy expression. This is done (see (24)). (3) Sum up the elemental energy to form the total system energy. (4) Perform the variation of the total system energy to obtain the equations of motion! Lets now work on items (1), (3) and (4).

1.A Partition a bar into many elements.

Partition a bar into finite elements element global node 1 (1) 2 (2) 3 (3) 4 (4) 5 element 2 2 3 1 2 element 3 4 3 1 2 (. . . )

element 1 2 global node 1 elelental node 1 2

element 1

element 2

element 3

u1 q1
(1)

u2 q2
(1)

u2 q1
(2)

u3 q2
(2)

u3 q1
(3)

u4 q2
(3)

Partitioning and establishing the Boolean relation between the global degrees of freedom, u, and elemental degrees of freedom, q

1.B Establish the relation between the elemental and assembled (global) degrees of freedom.
For element 1: q
(1)

q(1) 1 q(1) 2 q(2) 1 q(2) 2 q(3) 1 q(3) 2 0 0

u1 u2 u2 u3 u3 u4 (25)

For element 2: q(2) =

For element 3: q

(2)

I 0 0 I 0 I (2) = q 0 0 q(1)
q(3) 0 0 0 0

u 2 0 0 I 0 u3
I 0 0 I u4

0 u 0 1 For a general case: q = Lc uc

where the subscript c designates the continuum ddomain.

3.A Sum up the total system kinetic energy.

n total Tbar = el=1 1 {q(1)}T [m(1) ]{q(1)} + 1 {q(1)}T [m(1) ]{q(1)}+ 2 2 ... + 1 {q(n)}T [m(n) ]{q(n)} 2 el Tbar

q(1) T m(1) q(2) 0 = 0 .. q(n) 0

total Tbar = 1 2

0 m(2) 0 0

... ... ... ..

0 q(1) T 0 q(2) 0 .. (n) q(n) m

(26)

{q }T [m ] {q }

3.B Sum up the total system strain energy.

n total Vbar = el=1 1 {q(1)}T [k(1) ]{q(1)} + 1 {q(1)}T [k(1) ]{q(1)}+ 2 2 ... + 1 {q(n)}T [k(n) ]{q(n)} 2 el Vbar b

q(1) T k(1) q(2) 0 = 0 .. q(n) 0

0 k(2) 0 0

... ... ... ..

0 q(1) T 0 q(2) 0 .. (n) q(n) k

(27)

total Vbar = 1 {q }T [k ] {q } 2

3.C Assemble the partitioned elements back into a bar. This corresponds to substituting the partitioned elemental degrees of freedom, q , by the assembled (or global) degrees of freedom, uc , via (25): Total assembled system kinetic energy:
total T T Tbar = 1 qT m q = 1 uc Lc m Lc uc 2 2

T T = 1 uc [Lc m Lc ] uc 2
total T Tbar = 1 ug Mc uc , 2 T Mc = Lc m Lc

(28)

Total assembled system potential energy:

total T T Vbar = 1 qT k q = 1 uc Lc k Lc uc 2 2

T T = 1 uc [Lc k Lc ] uc 2
total T Vbar = 1 uc Kc uc , 2 T Kc = Lc k Lc

(29)

 Total assembled system external work:

n

W total =

{q(t)(el) }T {fel } (30)

= q f =
T uc

el=1 T

{L f }
T T fc = Lc f

T W total = uc fc ,

Euler-Lagranges equations of motion for the continuum Lagrangian:

Lc

= T total V total =
1 2 uc Mc

uc

1 2 uc Kc

uc

(31)

From the generic equation of motion d L L = Qk dt k qk we obtain the FEM equation of motion from (31) and (32): Mc uc + Kc uc = fc (33) (32)

What about the two end-springs and attached masses? Answer: Simply add both the kinetic and potential energy expressions of the two discrete mass and springs. 0 T0 = 1 M0 u 2 (t), 2 TL = 1 M L u 2 (t) L 2 (34)

V0 = 1 K 01 u 2 (t) + 1 K 02 [u 0 (t) u(0, t)]2 0 2 2 VL = 1 K L1 u 2 (t) + 1 K L2 [u L (t) u(L , t)]2 0 2 2

where u 1 (t) and u 2 (t) are the displacements of mass M0 and M L ; and, u(0, T ) and u(L , t) are the displacements of the continuum bar at x = 0 and x = L, respectively.

The Lagrangian for the two end-spring and mass models are given by T T L0 = 1 q0 m0 q0 1 q0 k0 q0 2 2
LL

= 1 qT m L q L 1 qT k L q L 2 L 2 L M0 0 0 0 0 , 0 0 , ML k0 = kL = qL = k01 + k02 k02 k L1 k L2 u(L , t) u L (t) k02 k02

m0 = mL = q0 =

(35)

k L2 k L2 + k L1

u 0 (t) , u(0, t)

How do we append the two end-springs and masses?

Unassembled Model (hardware)
M0 ML

k01

k02

k L1

k L2

( m0

k0 )

Unassembled Model (software) ( Mc Kc )

( mL

kL)

1. Stack the unassembled matrices: Munassemb = m0 0 0 k0 0 0 0 Mc 0 0 Kc 0 0 0 mL (36) Kunassemb = 0 0 kL

How do we append the two end-springs and masses? Contd 2. Construct the assembling Boolean matrix (or Assemble directly element-by-element into the assembled matrices): 1 0 0 0 0 0 = ... 0 0 0 0 0 0 | 0 ... ... 0 1 | 0 ... ... 0 | 1 0 ... ... 0 | 0 1 ... ... 0 | 0 1 0 ... 0 | 0 0 1 ... ... | ... ... ... ... ... | 0 ... ... ... ... | 0 ... ... 1 0 | 0 ... ... 0 1 0 ... ... 0 | 1 0 ... ... 0 | 0 0 0 | 0 | 0 | 0 | 0 | ... | ... | 0 | 0 0 1 (37)

Lassemb

How do we append the two end-springs and masses? Contd 3. Assemble:

T Massemb = Lassemb Munassemb Lassemb T Kassemb = Lassemb Kunassemb Lassemb

Massemb uassemb + Kassemb uassemb = fassemb

T uassemb = ([Lassemb Lassemb ]1 Lassemb ) uunassemb T fassemb = Lassemb funassemb q0 uunassemb = uc , funassemb = qL

(38)

f0 fc fL

4. Apply the boundary conditions and specialization for Homework #4, we apply the following: q0 = 0 {u 0 = 0, u c (1) = 0} (39) (40)

K L2 = 0 (as an input) In Matlab, (39) can be realized as Massemb (1 : 2, :) = []; K assemb (1 : 2, :) = []; f assemb (1 : 2) = []; Massemb (:, 1 : 2) = []; K assemb (:, 1 : 2) = [];

(41)

An Example of Three Bar Elements with end masses and springs 1 E A 1 Kc = 0 0 2 A 1 Mc = 0 6 0 m0 = M0 0 1 2 1 0 1 4 1 0 0 1 4 1 0 1 2 1 0 0 , 1 1 0 0 , 1 1

= L/3

= L/3

(42)

0 , 0 0 , ML

k01 + k02 k0 = k02 k L1 kL = k L2

k02 k02

0 mL = 0

k L2 k L2 + k L1

Three Bar Elements with end masses and springs - contd Assembling Boolean matrix Lassemb : 0 0 0 0 0 Lassemb 0 0 0 0 1 (43) Use now equation(38) to assemble the mass and stiffness matrices! 0 1 1 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 1 1 0 left spring-mass 1 system 0 0 0 three element bar 0 = = 0 0 0 left spring-mass 0 system 0

ASEN 5022 - Spring 2006 Dynamics of Aerospace Structures Lecture 13: 14 March Finite Element Modeling of Beam Bending Vibration

Lets consider the modeling of a beam under general boundary conditions by the nite element modeling.
w z Beam with unknown boundary conditions
EI, m(x)

k1 kw1
L

k2
x

kw2

Theoretical Basis: Same as the continuum beam model! Kinetic energy: T =

1 2 L

m(x)
0

w(x, t)2 d x, t

w(x, t) w(x, t)t = t

(1)

Potential energy of the beam: Vb =

1 2 L 0

{ E I (x) w(x, t)2 x } d x x 2 w(x, t) x2 , w(x, t) w(x, t)x = x

(2)

w(x, t)x x =

External energy due to distributed applied force f(x, t), shear forces Q (1,2) and moments M(1,2) applied at both ends: W =
0 L

f(x, t) w(x, t) d x (3)

+ Q 1 w(0, t) + Q 2 w(L , t) + M1 w(0, t)x + M2 w(L , t)x Q 1 = Q(0, t), M1 = M(0, t), Q 2 = Q(L , t) M2 = M(L , t)

Potential energy of four unknown springs: Vs = 1 { kw1 w(0, t)2 + k 1 w(0, t)2 x 2 + kw2 w(L , t) Hamiltons principle
t2 t1 2

+ k 2 w(L , t)2 x

(4)

[T Vb Vs + W ] dt = 0

(5)

From here on, FEM modeling differs from continuum modeling!

Key departure in FEM modeling of beams from continuum modeling: 1. Instead of carrying out the variation (the -process of Hamiltons principle(5) in terms of the continuum variable w(x, t), the energy expressions, (T, V, W, etc.), are approximated on a completely free beam with an arbitrary length , area A, the bending rigidity E I and mass density .

2. The interpolation of the transverse displacement w(x, t) over the completely free beam segment 0 x is chosen to satisfy the homogeneous differential beam equation, i.e., E I w(x, t)x x x x = 0 (6)

if possible. If not, it is chosen such that as the length of the beam gets smaller and smaller, it approximately satises the homogeneous equation.

We now consider a completely free isolated small segment of a beam.

A completely free element of length taken out from the beam l

w
1 2 an element isolated x

k1 kw1
1 2

k 2
x
EI, m(x)

kw2

Approximation of w(x, t) over the element segment, w(x, t) = c0 (t) + c1 (t)x + c2 (t)x 2 + c3 (t)x 3

<< L: (7)

Observe that the above interpolation satises the homogeneous differential equation of beam (6). How does one determine the coefcients, (c0 , c1 , c2 , c3 )? The answer comes from the nite element method.

The FEM beam bending element 1. Observing that w(x, t) and w(x, t)x are linearly independent from the result of variational formulation, we specify them at nodes 1 and 2 in the previous gure. Namely, two discrete variables at node 1: w(x1 , t) and w(x1 , t)x two discrete variables at node 2: w(x2 , t) and w(x2 , t)x (8)

2. For convenience, we introduce a local coordinate system or elemental coordinate system, (x, y) and set x1 = 0 and x2 = .

Substituting (8) into the interpolation function(7), we obtain

w1 (t) = w(x1 = 0, t) = c0 (t) + c1 (t)0 + c2 (t)02 + c3 (t)03 1 (t) = w(x1 = 0, t)x = c1 (t) + 2c2 (t)0 + 3c3 (t)02 w2 (t) = w(x2 = , t) = c0 (t) + c1 (t) + c2 (t)
2

+ c3 (t)
2

(9)

2 (t) = w(x2 = , t)x = c1 (t) + c2 (t)2 + c3 (t)3

Solving for the coefcients and backsubstituting into (7), one obtains w(x, t) = H1 ( ) w1 (t) + H2 ( ) 1 (t), + H3 ( ) w2 (t) + H4 ( ) 2 (t)
2 3

= x/
2 3

H1 ( ) = (1 3 + 2 ), H2 ( ) = ( 2 + ) H3 ( ) = (3 2 2 3 ), H4 ( ) = ( 2 + 3 )

(10)

For notational clarity, we express (10) in the form w(x, t) = N( ) q(t) N( ) = H1 H2 q(t)T = w1 (t)
w1 l
1

H3

H4 w2 (t)
w2 2
2

(11) 2 (t)

1 (t)

A completely free beam element and its nodal degrees of freedom

Element Discretization Elemental kinetic energy: T el = =

1 2 1 2

0 1 0

m(x) w(x, t)2 d x t m( ) w(, t)2 d, t x=

(12)

Elemental potential energy: Vbel = =

1 2 1 2

0 1 0

E I (x) w(x, t)2 x d x x E I ( )

3

(13)

w(, t)2 d

Discretization of Elemental Kinetic Energy T el = =

1 2 1 2 1 0 1 0 1 0

m( ) w(, t)2 d t A q(t)T [NT N] q(t)d A NT N d ] q(t) (14)

= 1 qT [ 2

= 1 q(t)T [mel ] q(t) 2 where we utilized (11), viz.:

3

w(x, t) =
i=0

ci (t)x i = N( ) q(t)

(15)

Discretization of Elemental Strain (Potential) Energy Vbel = = =

1 2 1 2 1 0 1 0

E I ( )
3

w(, t)2 d
T q(t)T [N N ] q(t)d

E I ( )
3 1

(16)

1 q(t)T 2

[
0 el

EI
3

T N N d ] q(t)

= 1 q(t)T [k ] q(t) 2 where we utilized: w(, t) = N( ) q(t) (17)

Elemental mass matrix, mel

m = A
13 35 11 210 9 70 13 420 11 210
2

9 70 13 420 13 35 11 210

13 420 2 140 11 210

2

(18)

105 13 420 2 140

105

Elemental stiffness matrix, kel

12 6 12 6 6 4 2 6 2 2 EI k= 3 12 6 12 6
6 2
2

(19)

Approximation of elemental external energy (3)

1

W el =
0

f(x, t) w(x, t) d x(=

0

f(, t) N( ) q(t)

d )

(3)

+ Q 1 w(0, t) + Q 2 w(L , t) + M1 w(0, t)x + M2 w(L , t)x Lets utilize the approximation(10): w(x, t) = N( ) q(t) w(x, t)x = and observe that w(0, t) = q1 (t), w( , t) = q3 (t), so that we have W el = q(t)T {
0 el

N( ) q(t)
w(0, t)x = q2 (t), w( , t)x = q4 (t)

(20)

(21)

NT f(x, t) fel = {

d } + Q 1 q1 (t) + Q 2 q3 (t) + M1 q2 (t) + M2 q4 (t) (22)

= q(t)T {f },

NT f(x, t)
0

d } + Q 1

M1

Q2

M2 T

Summary of elemental energy expressions: T el = 1 {q(t)(el) }T [mel ] {q(t)(el) } 2 Vbel = 1 {q(t)(el) }T [kel ] {q(t)(el) } 2 W el = {q(t)(el) }T {fel } (23)

Question: How do we model a long beam with beam elements? Answer: (1) (2) (3) (4) Partition the long beam into small elements. This needs to be explained. Generate the elemental energy expression. This is done (see (23)). Sum up the elemental energy to form the total system energy. Perform the variation of the total system energy to obtain the equations of motion!

Lets now work on items (1), (3) and (4).

1.A Partition a beam into many elements.

Partition a beam into finite elements element global node 1 (1) 2 (2) 3 (3) 4 (4) 5 element 2 2 3 1 2 element 3 4 3 1 2 (. . . )

element 1 2 global node 1 elelental node 1 2

element 1

element 2

element 3

u1 q1
(1)

u2 q2
(1)

u2 q1
(2)

u3 q2
(2)

u3 q1
(3)

u4 q2
(3)

Partitioning and establishing the Boolean relation between the global degrees of freedom, u, and elemental degrees of freedom, q

1.B Establish the relation between the elemental and assembled(global) degrees of freedom.
For element 1:

(1) =

(1) q1

For element 2:

q(2) = q(2) =

q q q q q

(1) 2 (2) 1 (2) 2 (3) 1 (3) 2

u1 u2 u2 u3 u3 u4
(24)

For element 3:

q(1) I 0 0 0 u 1 0 I 0 0 u2 0 q(2) = 0 I I 0 0 0 0 u3 0 0 I 0 q(3) u4 0 0 0 I

For general case: q

= L ug

3.A Sum up the total system kinetic energy.

n

T total =
el=1

T el

1 {q(1)}T [m(1) ]{q(1)} + 1 {q(1)}T [m(1) ]{q(1)}+ 2 2 ... + 1 {q(n)}T [m(n) ]{q(n)} 2

q(1) T m(1) q(2) 0 = 0 .. q(n) 0 T total = 1 q m q 2

0 m(2) 0 0

... ... ... ..

0 q(1) T 0 q(2) 0 .. (n) q(n) m

(25)

3.B Sum up the total system strain energy.

n

V total =
el=1

Vbel

1 {q(1)}T [k(1) ]{q(1)} + 1 {q(1)}T [k(1) ]{q(1)}+ 2 2 ... + 1 {q(n)}T [k(n) ]{q(n)} 2

q(1) T k(1) q(2) 0 = 0 .. q(n) 0 V total = 1 qT k q 2

0 k(2) 0 0

... ... ... ..

0 q(1) T 0 q(2) 0 .. (n) q(n) k

(26)

3.C Assemble the partitioned elements back into a beam. this corresponds to substituting the partitioned elemental degrees of freedom, q , by the assembled (or global) degrees of freedom, ug , via (24): Total assembled system kinetic energy: T T total = 1 qT m q = 1 ug LT m Lug 2 2 T = 1 ug [LT m L] ug 2 T T total = 1 ug Mg ug , 2 Mg = LT mL (27)

 Total assembled system external work:

n

W total =

{q(t)(el) }T {fel } (28)

= q f
T = ub {LT f }

el=1 T

T W total = ug fg

Euler-Lagranges equations of motion Lagrangian:

L

= T total V total =
1 2 ug Mg

ug

1 2 ug Kg

ug

(29)

From the generic equation of motion d L L = Qk dt k qk we obtain the FEM equation of motion from (29) and (30): Mg ug + Kg ug = fg (31) (30)

Modeling for Structural Vibrations: FEM Models, Damping, Similarity Laws

14

141

Lecture 14: MODELING FOR STRUCTURAL

142 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

14.1 FINITE ELEMENT MODELING OF VIBRATION PROBLEMS

In modeling of cable vibration problems by linear elements, it was observed that one needs to model the cable by more than 50 elements if the fundamental frequency is to be computed within 4 digit accuracy. What was not discussed therein is the accuracy of mode shapes. To gain further insight into nite element modeling of vibration problems, let us consider the modeling of plane beam vibration problems. For simplicity, a beam with simple-simple supports is used to model 5th modes. Classical theory tells us that we have the following solution:

k-th mode: k-th mode shape:

k =

k L

EI

(14.1)

k x W (x) = sin L

where E I is the bending rigidity, is the mass per unit beam length, and L is the beam span.

1 10

Frequency error vs. elements for simple support beam

0 10

6-th mode

beta*L = 18.8497

Frequency Error (%)

-1 10

-2 10

-3 10

-4 10 5

10

15

20 25 30 35 s Number of Beam Element

40

45

50

Figure 14.1 Frequency error vs. number of elements for simply-simply supported beam 142

143

14.1 FINITE ELEMENT MODELING OF VIBRATION PROBLEMS

Frequency error vs. elements for fixed-simple support beam

1 10

6-th mode 0 10 Frequency Error (%)

beta*L = 19.6351

-1 10

-2 10

-3 10 5

10

15

20

25 30 35 Number of Beam Element s

40

45

50

Figure 14.2 Frequency error vs. number of elements for xed-simply supported beam In using the nite element method for modeling of beam vibrations, the question arises: how many elements does one need for an accurate computation of its k-th mode and mode shape. While there exists a vast amount of literature on the accuracy and convergence properties of various elements which one may employ to answer the question, we will adopt a posteriori assessment approach. To this end, let us take a simply supported beam and concentrate on its 6th modes. The frequency error vs. the number of beam elements used are shown in Fig. 14.1. As can be seen, ve-digit accuracy is achieved with about 45 elements. If all one needs is the frequency with one percent accuracy, one could use only 10 elements. In Figure 14.2 the frequency error of a xed-simply supported beam is illustrated. Although the error for this case is a little higher than that of the simply supported case, the frequency error converges with the same trend. lets now focus on the mode shape accuracy.
Simple - Simple Support Beam 1.5 5 elements 6-th mode 1 beta*L = 19.8826 Freq(Hertz) = 918.1601

Simple - Simple Support Beam 1.5 10 elements 6-th mode 1 beta*L = 18.9243 Freq(Hertz) = 831.7825

0.5

0.5 Mode shape

converged mode shape computed mode shape spline fit

Mode shape

-0.5

-0.5

-1

-1

-1.5

0.1

0.2

0.3

0.4

0.5 0.6 Beam span

0.7

0.8

0.9

-1.5 0

0.1

0.2

0.3

0.4

0.5 0.6 Beam span

0.7

0.8

0.9

Figure 14.3 Frequency error vs. number of elements for xed-simply supported beam 143

Lecture 14: MODELING FOR STRUCTURAL

Simple - Simple Support Beam 1.5 15 elements 6-th mode 1 beta*L = 18.8652 Freq(Hertz) = 826.5961

144 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

Simple - Simple Support Beam 1.5 30 elements 6-th mode 1 beta*L = 18.8506 Freq(Hertz) = 825.3172

0.5 Mode shape Mode shape 0.1 0.2 0.3 0.4 0.5 0.6 Beam span 0.7 0.8 0.9 1

0.5

-0.5

-0.5

-1

-1

-1.5 0

-1.5 0

0.1

0.2

0.3

0.4

0.5 0.6 Beam span

0.7

0.8

0.9

Figure 14.4 Frequency error vs. number of elements for xed-simply supported beam

Simple - Simple Support Beam 1.5 40 elements 6-th mode 1 beta*L = 18.8499 Freq(Hertz) = 825.257 1 1.5

Simple - Simple Support Beam 50 elements 6-th mode beta*L = 18.8497 Freq(Hertz) = 825.2404

0.5 Mode shape Mode shape 0.1 0.2 0.3 0.4 0.5 0.6 Beam span 0.7 0.8 0.9 1

0.5

-0.5

-0.5

-1

-1

-1.5 0

-1.5 0

0.1

0.2

0.3

0.4

0.5 0.6 Beam span

0.7

0.8

0.9

Figure 14.5 Frequency error vs. number of elements for xed-simply supported beam Figures 14.3-5 illustrate how the mode shapes converge to the exact solution as the number of elements increases. Also plotted is spline-tted curves that utilize only the sample points (or computed discrete mode shape points). It is clear that curve tting in general enhances the discrete raw data points, especially for the case of crude models. Scanning over the six mode shape plots (Fig. 14.3-5) vs. the increasing number of elements, an acceptable number of element for capturing the sixth mode shape appears to be 30. Note that there are six half sine waves in the 6th mode shape. Hence, for the 30-element model, each half sine wave is sampled by ve elements or six nodal points, meaning that ten elements span one full sine wave. To conclude, ten elements captures the sixth mode frequency with less than one percent error, whereas for an adequate mode shape capturing three time of that elements (30 elements) are needed. This is often referred to as three-to-one rule. Finally, the case of ten-element model satises the so-called Shannons sampling criterion which state that for the minimum sampling number for a sinusoidal signal is three.

144

145

14.2 CHARACTERIZATION OF LINEAR STRUCTURAL DYNAMICS EQUATIONS

14.2 CHARACTERIZATION OF LINEAR STRUCTURAL DYNAMICS EQUATIONS

Let us now study the characteristics of the second-order damped system Mu + Du + Ku = f(t), D = (M + K) (14.2)

where and are constants. The damping matrix, D = M + K, is called a Rayleigh damping as it is proportional both to mass and stiffness of the system. The coupled equations of motion for a linear structure(14.2) can be decoupled by the (n n) eigenvector matrix T, which relates the displacement vector u to a generalized solution vector q via u = Tq Substituting (14.3) into (14.2) and premultiplying the resulting equations by TT one obtains I q + (I + ) q + in which TT MT = I TT KT =
2 = diag(1 , ..., 2 ) N

(14.3)

q = fq ,

fq = TT f( f )

(14.4)

(14.5) (14.6)

where j is the j-th undamped frequency component of (14.4). The solution of (14.4) for its homegeneous part can be expressed as q = cest where s is in general complex constants. Substitution of (14.7) into (14.4) with fq = 0 yields {s2 I + ( I + )s + Equation (14.8) has a nontrivial solution only if det |s2 I + ( I + )s + from which the k-th solution component can be expressed as qk = ak esk t + ak esk t where sk and sk are the k-th complex conjugate pairs of (14.9) and ak and ak are arbitrary constants. Two characterizations of (14.9) are possible: frequency characterization, viz., by xing the damping parameters and and varying the undamped frequency ; and, damping characterization, viz., by xing the undamped frequency and varying the damping parameters and . 145

(14.7)

}c = 0

(14.8)

|=0

(14.9)

(14.10)

Lecture 14: MODELING FOR STRUCTURAL

146 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

Root loci of proportionally damped structural dynamics equation 4 B

Imaginary part of normalized frequency by sampling rate h

Increasing frequency

1 C

( 1 1/2 ) ____________

(1/, 0)

s1 o

s2

-1

-2 Increasing frequency B -4 -8 -7 -6 -5 -4 -3 -2 -1 Real part of normalized frequency by sampling rate h 0 1

-3

Fig. 14.6 Root Loci of Proportionally Damped Structural Dynamics System

14.2.1 Frequency Characterization Based on the Root Locus Method The characteristic equation (14.9) represents n roots: s j2 + ( + 2 )s j + 2 = 0, j j j = {1, 2, . . . , n} (14.11)

Since not only the distribution but also the maximum and minimum frequencies are not known a-priori, a complete range of the Rayleigh-damped system can be characterized by the following expression: s 2 + ( + 2 )s + 2 = 0 0 (14.12)

In order for the subsequent characterization to remain valid for all sampling rates, we normalize (14.12) by the sampling rate h such that s 2 + (h + (h)2 ) + (h)2 = 0 s h 0 h (14.13) = , h = h

s s 2 + ( + 2 ) + 2 = 0,

= h,

Note rst that the rigid-body motion (i.e., = h = 0) corresponds from (14.13) to: s1 = {0, 0} and s2 = {h, 0} = {, 0} 146 (14.14)

147

14.2 CHARACTERIZATION OF LINEAR STRUCTURAL DYNAMICS EQUATIONS

These two roots are marked as s1 and s2 in Fig. 14.6. As the undamped frequency is increased, the root locus approaches the branch point at A in Fig. 18.6: s A = {h(1 1 )/, 0} = {(1 1 )/, 0} (14.15)

As is further increased, the root locus follows the half circle with its center at {1/, 0} and with its radius , viz.: 1 / s B = {1/, 1 /} The two complex roots merge at sC = { h(1 + 1 ) , 0} (14.16)

(14.17)

For > |sC |, one locus branches out toward the negative innite axis while the other approaches the origin {0, 0}. This is illustrated in Fig. 14.6. For the special case = 0, viz., mass proportional damping, the branch occurs at s = ( h , 0) when 2 = /2. Hence, the solution components with frequency < exhibit over-critically damped responses 2 as the locus becomes a straight line () = h . In particular, if = 0, the root loci coincide with the s 2 imaginary axis, whose response is characterized by purely oscillatory components. From the physical viewpoint, the case of mass-proportional damping introduces higher modal damping for lower frequency solution components and the degree of damping decreases as the frequency increases. This does not, however, necessarily mean that the response components of the high-frequency modes will decay slower than those of the low-frequency modes within a time period. As a matter of fact, the decay rate is uniform for all frequency components since we have |est | = et , for all (14.18)

The root locus of the stiffness-proportional damping case, = 0, touches the origin and forms the half circle as is increased. As is further increased, the locus branches out into the negative real axis. Therefore, the decay rate increases as increases. As such, this representation of system damping is often used in the modeling of structural damping due to joint effects, acoustic noise and internal material friction. 14.2.2 Damping Characterization for Constant Frequency The conventional characterization of damping into the equations of motion for structures is to express each component of (14.9) (14.19) s2 + 2 sk + 2 = 0 k so that the characteristic roots can be expressed as sk = 1 2 where is termed the damping ratio. The equivalent damping ratio for (14.11), therefore, is eq = 1 ( + ) 2 147 (14.21) (14.20)

Lecture 14: MODELING FOR STRUCTURAL

148 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

Now, we like to examine for a xed what happens when is varied. For the undamped case, viz., eq = 0, the roots lie on the imaginary axis, s = {0, }. As is increased, the roots rotate toward the left-hand complex plane with the constant magnitude of = h since || = h s (14.22)

and the shift angle being = tan1 1 2 / . The two complex roots merge on the real axis when = 1, thus the root locus is a half-circle with radius h. For > 1, one root branches out to the innite negative axis while the other approaches the origin (see Figure 18.6). This invariance property, that is, the magnitude of the complex root remains the same for all damping ratios, plays an important role both for controller synthesis and computational algorithms.

14.3 VARIOUS DAMPING MODELS Modeling of damping remains a challenge in structural dynamics. Over the years various damping models have been proposed. Below lists a sample of parameterized damping models. 14.3.1 Mass-Proportional Damping Model If the damping can be made proportional to the mass such as dynamic friction cases, the damping matrix can be parameterized according to D = M as already discussed in the previous section. The root loci are a special case of (14.2) governed by s 2 + s + 2 = 0 (14.24) (14.23)

which is plotted in Fig. 18.7 as method A with its parameter = 1. Note that the root loci form a straight vertical line when ( > /2). 14.3.2 Viscous Damping Model One of the most widely used damping models is the viscous damping characterization. This has been adopted for modeling of coated damping layers, of lubricants in rotating machines, and of a plethora of unknown sources. Mathematically, the viscous damping model can be expressed as D = 2T [ p ] TT , TT MT = I, [ p ] = diag (1 1 , 2 2 , ..., N N ) TT KT =
p p p

(14.25)

Therefore, the characteristic equation of a viscous damped system is given by j s j2 + 2 j j s j + 2 = 0

p

(14.26)

Note that the two cases { p = 0, 1 = 2 = . . . = N = } and { p = 2, 1 = 2 = . . . = N = } have been studied in the preceding section. 148

149

14.3 VARIOUS DAMPING MODELS

Root loci of various proportional damping models Imaginary part of normalized frequency by sampling rate h
25

20

model D1

model C model D2

15

10

model D3 model B model A

-5

-10

-15

-20

-25 -2

-1.8

-1.6

-1.4

-1.2

-1

-0.8

-0.6

-0.4

-0.2

Real part of normalized frequency by sampling rate h

Fig. 14.7 Root Loci of Proportionally Damped Structural Dynamics System The root loci of a viscous damped case, when combined with a mass-proportional damping, are obtained by { p = 1, 1 = 2 = . . . = N = } s 2 + ( + 2 ) s + 2 = 0 (14.27)

which is plotted in Fig. 14.7 labeled as model C with its parameters { = 1.0, = 0.025}. Note that in this model the modal damping ratio is the same ( = const) for all the frequencies greater than { > /2(1 ), < 1.0 }. 14.3.3 Intermediate-frequency damped case In practice such as acoustically treated structural systems, both the low and high frequencies are associated with very low damping while intermediate frequencies are damped depending on damping treatment employed. Such damping models may be parameterized according to D = M + T [ (1 tanh h) p ] TT which results in the following root loci equation: s 2 + ( + (1 tanh h) p ) s + 2 = 0 (14.29) (14.28)

The root loci of this model are plotted in Fig. 14.7 as method D1, D2, D3 with its parameters p = 2, = 1.0, h = 1.0, = 0.025, = {0.025, 0.05, 0.1} 149 (14.30)

Lecture 14: MODELING FOR STRUCTURAL

1410 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

Note that as becomes large, the real part of the root loci approaches s as (14.31)

Hence, if = 0 the root loci will lie on the imaginary axis, indicating no damping for those frequencies.

14.4 USE OF SIMILARITY LAWS IN MODELING Dimensional analysis has been widely used in experiment design and scale-model construction. The task in dimensional analysis is to identify physical quantities that inuence the physical phenomena on hand, then deduce a independent set of dimensionless products. These dimensionless products are then utilized for experiment design and scale-model construction. The underlying principle is Buckinghams theorem that enables a systematic construction of linearly independent dimensionless products. It should be pointed out that dimensional analysis is applicable both to statics and dynamics. In dynamics, a useful dimensional analysis may be carried out, which can provide insight into dynamic behavior. It is referred to mechanical similarity after Landau and Lifshits. The basic idea is as follow. The starting point of mechanical similarity is that multiplication of kinetic and potential energy expression or the Lagrangian by any constant does not affect the equations of motion. It is this observation that can yield the dynamic behavior of a system without actually solving the equations of motion. For example, we may pose the question: Does the amplitude of string vibration affect the frequency, assuming its potential energy is a quadratic function of its amplitudes? To answer this question, let us consider the simplest case, i.e., a single mass and a single spring case given by T = 1 m x 2, 2 U = 1 kx 2 2 L= T Y (14.32) x(0) = x0 , x(0) = x0

m x + kx = 0,

where m and k are the mass and spring constant, x is the vibration amplitude, and x0 and x0 are the initial amplitude and velocity of the mass. Of course, we all know that the frequency of the above system is given by = k/m (14.33)

which clearly shows that the frequency is independent of the amplitude x(t). Let us scale the new amplitude x by = so that the potential energy U changes to U (x ) = 1 k(x )2 = 1 2 x 2 2 2 Now we introduce the time scale = t t (14.36) (14.35) x x (14.34)

where characterizes the ratio of periods of motion or time durations of the two systems. 1410

1411 The kinetic energy fo the system is now changed to K (x , t ) = 1 m( 2

14.4 USE OF SIMILARITY LAWS IN MODELING

dx 2 ) = 1 ( )2 m x 2 2 t

(14.37)

Therefore, the Lagrangian of the new system becomes 1 L = T U = 1 ( )2 m x 2 1 2 x 2 = 2 ( 2 1 m x 2 1 kx 2 ) 2 2 2 2 (14.38)

The above equation states that the only way the resulting equation of mition is unaltered is to choose = L = 2 ( 1 m x 2 1 kx 2 ) = 2 L 2 2 Since the time scale ratio is unity, it means that the frequencies for both systems are the same. In other words, the frequencies of a lumped spring-mass system is unaffected by their vibration amplitudes. Comparing (14.33) and (14.39) one may argue that the mechanical similarity method is somewhat more complicated than what could be obtained by the frequency equation (14.33). Its chief advantage is that one needs not solve the governing equation (14.32) for its frequency expression. Keplers Third Law: Let us now apply the mechanical similarity to the motions of two satellites around a heavenly body. Newtons law of universal gravitation states G Mm G Mm U = (14.40) 2 r r where G is the universal gravitational constant, M and m are the masses of the heavenly body and the satellite, and r is the distance between the two bodies. The Lagrangian of the two satellites may be expressed as F= G Mm 1 , |r1 | G Mm 2 , L2 = 1 m 2 v2 v2 + 2 |r2 | L1 = 1 m 1 v1 v1 + 2 Let us transform L2 by introducing to obtain L2 = c( 1 m 1 v1 v1 + 2 t2 = t1 , r2 = r1 c= m2 2 ( ) m1 dr1 dt1 dr2 v2 = dt2 v1 = t =1 t

(14.39)

(14.41)

(14.42) (14.43)

2 G Mm 1 ), 3 r1

Note that the above scaling on L2 amounts to a thought experiment in which the satellite m 2 is brought to the orbit of satellite m 1 . For this thought experiment to be true, its corresponding equation of motion should be the same as that obtained by L1 . This can be the case only if the following condition is satised: L2 = c L1 2 =1 3 t2 r2 ( )2 = ( )3 t1 r1 (14.44)

which states that square of the revolution ratio of the two satellites is proportional to the cube of the ratio of the orbital sizes, Keplers third law. 1411

Lecture 14: MODELING FOR STRUCTURAL

1412 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

The Race between a Sliding Block and a Rolling Cylinder on an Inclined Slope Consider a cube sliding on an inclined plane without friction, whose Lagrangian can be expressed as Lcube = 1 m s 2 mgs sin 2 (14.45)

where m is the mass of the cube, s is the distance along the slope measured from the bottom of the slope, g is the gravity, is the inclined angle of the slope, and s is the speed of the cube along the slope. If a cylinder of the same mass m is to roll along the slope without slip, its Lagrangian can be obtained as s Lcyl = 1 m( )2 + 1 ( 1 mr 2 )2 mgs sin 2 2 2 (14.46)

where r is the radius of the cylinder, and is the angular velocity of the cylinder during rolling without slip. Since the angular velocity can be expressed in terms of s as = (14.46) can be simplied to Lcyl = Transforming the above by the scaling t /t = , yields s /s = (14.49)
1 2

s , r

s =

s t

(14.47)

3m ( )2 mgs sin s 2

(14.48)

3m 2 2 3 2 1 2 2 2 mgs sin ] (14.50) ( ) mgs sin = s [ ms 2 2 2 2 2 3 Comparing the above with (14.45), we nd for the resulting equation of motion from (14.50) to be the same as obtained by (14.45) we must have Lcyl =
1 2

2 2 =1 3
3 2

t = t

3 2

(14.51)

Now if the distance is the same, i.e., s = s, we have = 1. Therefore, the cylinder will arrive at the bottom of the inclined slope by a factor of longer over the arrival time of the cube. In other words, the cube will arrive at the bottom faster than the cylinder, and the arrival time ratio of the two bodies will be proportional to the ratio of the square root of the kinetic energies. Physically, the reason for the slower average speed of the cylinder is because for the case of cylinder, the same potential energy change is translated into both the translational and rotational kinetic energy while the entire change energy gets into the translational kinetic energy for the cube. 14.4.1 Mechanical Similarity in String, Rod and Shaft Vibrations The Lagrangian of strings, rods and shafts can be expressed as L = T U T = U=
1 2 1 2 0

w(x, t) 2 ] dx t w(x, t) 2 ] dx k(x) [ x m(x) [ 1412

(14.52)

1413

14.4 USE OF SIMILARITY LAWS IN MODELING

where m(x) is the mass per unit length for string and rods, and the mass moment of inertia for shafts, respectively; and k(x) denotes the tension force, the product of Youngs modulus and the cross section area E A(x), and the shear rigidity G I (x) for strings, rods and shafts, respectively. Let us consider the following scaling transformations: m = m k = , k x = = x , t p = = t p (14.53)

which, when substituted into (14.52), yields L =( 2 ) T ( ) U = ( 2 ) [T ( 2 ) U ] 2 (14.54)

In order for the transformations not to affect the resulting equations of motion, we must have L = CE = ( 2 ) L Period ratio = p = p k/m k /m 2 =1 2 = (14.55)

where C E is the energy ratio whose role we will discuss shortly. The preceding similarity law is specialized for strings, rods and shafts in Table 14.1. Note that if the frequency to be the same from one structure to another, e.g., in laboratory experiment compared to the real system, then the experiment design must employ = (14.56)

k /m

k/m

Table 14.1 Similarty Laws in Strings, Rods, and Shafts Strings Period Ratio (p/p)
T / A T / A

Rods
E/ E /

Shafts
G I / A G I / A

On the other hand, if the frequency ratio is to be proportional to the dimension ratio, then one must see to it that the following is observed: p = p k k = m m (14.57)

Other scaling can be similarly considered from Table 14.1. It should be noted that the preceding similarity laws are valid when the scaled and original systems have the same boundary conditions. 1413

Lecture 14: MODELING FOR STRUCTURAL

1414 VIBRATIONS: FEM MODELS, DAMPING, SIMILARITY LAWS

Observe that the similarity laws summarized in Table 14.1 can be derived if one solves for each of the three vibration problems. The emphasis is to utilize the respective energy expressions rather than the solutions of the governing differential equations. This is because it is often considerably easier to obtain the energy expressions of a system than obtain the vibration characteristics or quasi-static deformations. Finally, let us consider the role of the energy ratio ) (14.58) 2 which plays a pivotal role in sizing up the experiment design as it offers the experiment energy requirements relative to the actual system. In practice it is often the energy requirement that dictates the scaling than the geometrical considerations. Notice that since we have only one constraint for four parameters, the remaining three parameters can be used in meeting practical considerations in experiment design. CE = ( 14.4.2 Mechanical Similarity in Euler-Bernoulli Beam The Lagrangian of a continuous Euler-Bernoulli beam can be expressed as L = T (Um + Ug ) T = Um = Ug =
1 2 1 2 1 2

A(x) [
0

w(x, t) 2 ] dx t 2 w(x, t) 2 E I (x) [ ] dx x2 w(x, t) 2 ] dx P(x) [ x

(14.59)

where w(x, t) is the transverse displacement of the neutral axis of the beam, m(x) is the mass per unit beam length, E I (x) denotes the bending rigidity of the beam, and P(x) denotes the pre-stressed axial force. Introducing the scaling transformations given by (14.53) as modied to the case of beam, the Lagrangian of the new system can be expressed as L =( f E I ) T 3 Um U g , = , 2 EI 2 f 2 ) Ug ] = ( 2 ) [T ( 4 ) [Um + ( ) L, 2 if ( 2 ) =1 4 and ( f = P P (14.60)

L =(

f 2 ) =1

Therefore, the similarity law for a beam is expressed as p = p E I / A ( )2 E I / A provided EI P (x) = P(x) EI
2 2

if P(x) = 0

(14.61)

Notice that the appendage arms for a spinning satellite, helicopter rotor blades and turbine blades all experience axial tensions. Hence, the scaling of beam sizes must consider the axial force scaling as well. One important application of the preceding scaling law is for the experiment design of rotating members for micro-machine members as the rotating speed is very high and experiment design often necessitates a scaling up rather than scaling down. 1414

Solution of Vibration and Transient Problems

15

151

Chapter 15: SOLUTION OF

VIBRATION AND

TRANSIENT PROBLEMS

152

15.1 MODAL APPROACH TO TRANSIENT ANALYSIS Consider the following large-order nite element model equations of motion for linear structures: Mu(t) + Du(t) + Ku(t) = f(t), D = (M + K), (, ) are constant. (15.1)

where the size of the displacement vector, n, ranges from several thousands to several millions. Now suppose we would like to obtain the displacement response, u(t), for expected applied force, f(t). There are two approaches: direct time integration and modal superposition. We will differ direct time integration techniques for transient response analysis to the latter part of the course, and concentrate on modal superposition techniques. To this end, we rst decompose the displacement vector, u(t), in terms of its modal components by u(t) = q(t) (15.2) where is the mode shapes of the free-vibration modes, and q(t) is the generalized modal displacement. The mode shape matrix has the property of simultaneously diagonalizing both the mass and stiffness matrices. That is, it is obtained from the following eigenvalue problem: K=M
2 = diag(1 , ..., 2 ) N

(15.3)

where j is the j-th undamped frequency component of (15.1). In structural dynamics, one often employs the following special form of mode shapes (eigenvector): T K = , T M = I = diag(1, 1, ..., 1) (15.4)

Substituting (15.2) into (15.1) and pre-multiplying the resulting equation by T results in the following uncoupled modal equation: qi (t) + ( + i2 ) qi (t) + i2 qi (t) = pi (t), pi (t) = (i, :)T f(t) The above equation can be cast into a canonical form xi (t) = Ai xi (t) + b pi (t), Ai = whose solution is given by xi (t) = eAi t xi (0) +
0 t

i = 1, 2, 3, ..., n.

(15.5)

xi = [ qi (t) qi (t) ]T b= 0 1 (15.6)

0 i2

1 , ( + i2 )

eAi

(t )

b pi ( ) d

(15.7)

It should be noted that the above solution provides only for one of the n-vector generalized modal coordinates, q(t) (n 1). Carrying out for the entire n-modal vector, the physical displacement, u(t) (n 1), can be obtained from (15.2) by (15.8) u(t) = (1 : n, 1 : n) q(t) (1 : n, 1), q = [ q1 (t), q2 (t), ..., qn (t) ]T 152

153

15.2 SOLUTION BY DIRECT INTEGRATION METHODS

While the solution method described in (15.2) - (15.8) appears to be straightfowrad, its practical implementation needs to overcome several computational challenges, which include: (a) When the size of discrete nite element model increases, the task for obtaining a large number of modes (m), if not all, m << n, becomes computationally expensive. In practice, it is customary to truncate only part of the modes and obtain an approximate solution of the form u(t) (1 : n, 1 : m) q(t) (1 : m, 1), q = [ q1 (t), q2 (t), ..., qm (t) ]T , m << n (15.9)

It is not uncommon to have m/n < (1/100 1/1000) . (b) A typical vehicle consists of many substructures whose structural characteristics are distinctly different from one to another. For example, a fuselage has different structural characteristics from wing structures. Likewise, engine blocks are considerably stiffer than the car frame structure. The impact of stiffness differences on the computed modes and mode shapes can lead to accuracy loss, and frequently to an unacceptable level. (c) In modern manufacturing arrangements, rarely an aerospace company or automobile company designs, manufactures, assembles and tests the entire vehicle system. This means, except for the nal performance evaluation, each substructure can be modeled, analyzed and tested before it can be assembled, as a separate and independent structure. An alternative approach is to numerically integrate the equations of motion (15.1). We will discuss computational procedures of two direction algorithms in the next section. Their algorithmic properties will be examined later in the course.

15.2 SOLUTION BY DIRECT INTEGRATION METHODS There are two distinct direct time integration methods: explicit and implicit integration formulas. We summarize their computational sequences below. 15.2.1 Central Difference Method for Undamped Case (D = 0) First, we express the acceleration vector u from(15.1) as u(t) = M1 (f K u(t)) (15.10)

Hence, it is clear that if the mass matrix is diagonal, the computation for obtaining the acceleration vector would be greatly simplied. We now describe direction time integration by the central difference method: Initial step: Given the initial conditions, {u(0), u(0), f(t)}, obtain the velocity at the half step {t = h, h = u(0) = M1 (f(0) K u(0)) 2 u( 1 h) = u(0) + 1 h u(0) 2 u(h) = u(0) + h Subsequent steps 153 2 u( 1 h) (15.11) t} by

Chapter 15: SOLUTION OF

VIBRATION AND

TRANSIENT PROBLEMS

154

Ttotal = hn max for n = 1 : n max u(n) = M1 (f(n) K u(n)) u(n + 1 ) = u(n 1 ) + h u(n) 2 2 u(n + 1) = u(n) + h u(n + 1 ) 2 end 15.2.2 The Trapezoidal Rule for Undamped Case (D = 0) This method is also referred to Newmarks implicit rule with its free parameter chosen to be ( = 1 , = 1/4). 2 Among several ways of implementing the trapezoidal integration rule, we will employ a summed form or half-interval rule given as follows. u(n + 1 ) = u(n) + 1 h u(n + 1 ) 2 2 2 u(n + 1 ) = u(n) + 1 h u(n + 1 ) 2 2 2 (15.13) u(n + 1) = 2u(n + u(n + 1) = 2u(n +
1 ) 2 1 ) 2

(15.12)

u(n) u(n)

In using the preceding formula, one multiply the rst of (15.13) by M to yield M u(n + 1 ) = M u(n) + 1 h M u(n + 1 ) 2 2 2 The term M u(n + 1 ) in the above equation is obtained from (15.1) as 2 Mu(n + 1 ) = f(n + 1 ) Du(n + 1 ) K u(n + 1 ) 2 2 2 2 which, when substituted into (15.14), results in M u(n + 1 ) = M u(n) + 1 h {f(n + 1 ) Du(n + 1 ) K u(n + 1 )} 2 2 2 2 2 [M +
1 hD] 2

(15.14)

(15.15)

(15.16) u(n +
1 ) 2

= M u(n) +

1 h 2

{f(n +

1 ) 2

K u(n +

1 )} 2

Now multiply the second of (15.13) by [M + 1 hD] to obtain 2 [M + 1 hD] u(n + 1 ) = [M + 1 hD] u(n) + 1 h [M + 1 hD] u(n + 1 ) 2 2 2 2 2 2 third, substitute the second term in the righthand side of (15.17) by (15.16), one obtains [M + 1 hD] u(n + 1 ) = [M + 1 hD] u(n) + 1 h {M u(n) + 1 h {f(n + 1 ) K u(n + 1 )} 2 2 2 2 2 2 2 [M + 1 hD + ( 1 h)2 K] u(n + 1 ) = M {u(n) + 1 h u(n)} + 1 h Du(n) + ( 1 h)2 f(n + 1 ) 2 2 2 2 2 2 2 154 (15.18) (15.17)

155 Implicit integration steps Assemble: A = [M + 1 hD + ( 1 h)2 K] 2 2 Factor: A = LU for n = 0 : n max

15.4 ILLUSTRATIVE PROBLEMS

b(n) = M {u(n) + 1 h u(n)} + 1 h Du(n) + ( 1 h)2 f(n + 1 ) 2 2 2 2 u(n + 1 ) = A1 b(n), where A1 = U1 L1 2 u(n + 1 ) = {u(n + 1 ) u(n)}/( 1 h) 2 2 2 u(n + 1) = 2 u(n + 1 ) u(n) 2 u(n + 1) = 2 u(n + 1 ) u(n) 2 end (15.19)

15.3 DISCRETE APPROXIMATION OF MODAL SOLUTION The modal-form solution of the equations of motion for linear structures given by (15.7) and (15.8) involves the convolution integral of the applied force. For general applied forces an exact evaluation of the convolution integral can involve a considerable effort. To this end, an approximate solution is utilized in practice. To this end, (15.7) is expressed in discrete form at time t = nh: xi (nh) = eAi
nh

xi (0) +
0

nh

eAi

(nh )

b pi ( ) d

(15.20)

Likewise, at time t = nh + h, xi (nh + h) is given by xi (nh + h) = eAi

nh+h

xi (0) +
0 nh

nh+h

eAi

(nh+h )

b pi ( ) d (15.21)

= eAi h [eAi +
nh+h nh

xi (0) +
0

nh

eAi

(nh )

b pi ( ) d ]

eAi

(nh+h )

b pi ( ) d

The bracketed term in the above equation is xi (nh) in view of (15.20) and the second term is approximated as nh+h nh+h eAi (nh+h ) b pi ( ) d [ eAi (nh+h ) d ] b pi (nh) (15.22) nh nh A1 (eAi h I) b p (nh)
i i

Substiutting this together the bracketed term by xi (nh) into (15.21), xi (nh + h) is approximated as xi (nh + h) = eAi h xi (nh) + Ai1 (eAi h I) b pi (nh), xi = [ qi (nh + h), qi (nh + h) ]T (15.23)

Once xi (nh + h) is computed, the physical displacement u(nh + h) is obtained via the modal summation expression (15.8).

155

Chapter 15: SOLUTION OF

VIBRATION AND

TRANSIENT PROBLEMS

156

15.4 ILLUSTRATIVE PROBLEMS Consider a beam with boundary constraints as shown in Figure 15.1. For illustrative purposes, two beams will be considered: a xed-simply supported beam and a beam with boundary constraints with the following specic boundary conditions: Fixed and simply supported ends: w(0, t) = Beam with boundary springs: w(0, t) = w(L , t) = 0, k 1 = (1.e + 5) E I , L k 2 = EI L (15.25) w(0, t) = 0, x w(L , t) = 0 (15.24)

w z Beam with boundary constraints

EI, m(x)

k1 kw1
L

k 2
x

kw2

Fig. 15.1 Beam with boundary constraints It should be noted that the end condition, (w(0, t) = w(L , t) = 0), is equivalent to (kw1 , kw2 ). However, in computer implementation it is impractical to use (kw1 , kw2 ) due to limited oating point precision. The applied force chosen are Step load: f (L/2, t) = 1.00, 0 t Sinusoidal load: f (L/2, t) = sin(2 f f t) (15.26)

where the forcing frequency is set to f f = 1.5(1 /2/ ), with 1 being the fundamental frequency of the model problems. Figures 15.2-3 illustrate time responses of a beam with boundary springs subject to unit mid-span step load. The responses by solid red lines are those obtained by using the central difference method, the ones with + are by the trapezoidal rule, and the blue lines by the canonical formula(15.23). The step increments used for the three methods are h= 1.8928E 7, for the central difference method 1.5861E 5, for the canonical formula 1.6.3445E 5, for the trapezoidal rule (15.27)

It should be noted that the stepsie for the central difference method is dictated by the computational stability whereas taht of the canonical formula and teh trapezoidal rule by accurcy considerations. We hope to revisit this issue later in the course. 156

157
x 10
-6

15.4 ILLUSTRATIVE PROBLEMS

Beam under midspan step load with 4 beam elements

Vertical displacement at the beam center

3 time (sec.)

6 x 10

7
-3

Fig. 15.2 Beam Midspan Vertical Time Response

2.5

x 10

-5

Beam under midspan step load with 4 beam elements

2 Rotational displacement at the beam center

1.5

0.5

-0.5

-1

3 time (sec.)

6 x 10

7
-3

Fig. 15.3 Beam Midspan Rotational Time Response

157

Methods for Vibration Analysis

16

161

Chapter 16: METHODS FOR VIBRATION ANALYSIS

162

16.1 PROBLEM CLASSIFICATION According to S. H. Krandall (1956), engineering problems can be classied into three categories: equilibrium problems eigenvalue problems propagation problems Equilibrium problems are characterized by the structural or mechanical deformations due to quasi-static or repetitive loadings. In other words, in structural and mechanical systems the solution of equilibrium problems is a stress or deformation state under a given load. The modeling and analysis tasks are thus to obtain the system stiffness or exibility so that the stresses or displacements computed accurately match the observed ones. Eigenvalue problems can be considered as extentions of equilibrium problems in that their solutions are dictated by the same equilibrium states. There is an additional distinct feature in eigenvalue problems: their solutions are characterized by a unique set of system congurations such as resonance and buckling. Propagation problems are to predict the subsequent stresses or deformation states of a system under the time-varying loading and deformation states. It is called initialvalue problems in mathematics or disturbance transmissions in wave propagation. Modal testing is perhaps the most widely accepted words for activities involving the characterization of mechanical and structural vibrations through testing and measurements. It is primarily concerned with the determination of mode shapes (eigenvevtors) and modes (eigenvalues), and to the extent possible the damping ratios of a vibrating system. Therefore, modal testing can be viewed as experimental solutions of eigenvalue problems. There is one important distinction between eigenvalue analysis and modal testing. Eigenvalue analysis is to obtain the eignvalues and eigenvectors from the analytically constructed governing equations or from a given set of mass and stiffness properties. There is no disturbance or excitation in the problem description. On the other hand, modal testing is to seek after the same eigenvalues and eigenvectors 162

163

16.2 STRUCTURAL MODELING BY SYSTEM IDENTIFICATION

by injecting disturbances into the system and by measuring the system response. However, modal testing in earlier days tried to measure the so-called free-decay responses to mimick the steady-state responses of equilibrium problems.

Table 1: Comparison of Engineering Analysis and System Identication Engineering Analysis Equilibrium Construct the model rst, then obtain deformations under any given load. System Identication Measure the dynamic input/output rst, then obtain the exibility.

Eigenvalue

Construct the model rst, then obtain eigenvalues without any specied load.

Measure the dynamic input/output rst, then obtain eigenvalues that corresponds to the specic excitation.

Propagation

Construct the model rst, then obtain responses for time-varying loads.

Measure the dynamic input/output rst, then obtain the model corresponds to the specic load

Observe from the above Table that the models are rst constructed in engineering analysis. In system identication the models are constructed only after the appropriate input and output are measured. Nevertheless, for both engineering analysis and system identication, modeling is a central activity. Observe also that, in engineering analysis, once the model is constructed it can be used for all of the three problems. On the other hand, the models obtained by system identication are usually valid only under the specic set of input and output pairs. The extent to which a model obtained through system identication can be applicable to dynamic loading and transient response measurements depends greatly upon the input characteristics and the measurement setup and accuracy.

163

Chapter 16: METHODS FOR VIBRATION ANALYSIS

164

16.2 STRUCTURAL MODELING BY SYSTEM IDENTIFICATION As noted in the previous section, modeling constitutes a key activity in engineering analysis. For example, the nite element method is a discrete structural modeling methodology. Structural system identication is thus a complementary endeavor to discrete modeling techniques. A comprehensive modeling of structural systems is shown in Fig. 1. The entire process of structural modeling is thus made of seven blocks and seven information transmission arrows (except the feedback loop). Testing consists of the rst two blocks, Structures and Signal Conditioning along with three actions, the application of disturbances as input to the structures, the collection of sensor output, and the processing of the sensor output via ltering for noise and aliasing treatment. FFT and Wavelets Transforms are software interface with the signal conditioniners. From the viewpoint of system identication, its primary role is to produce as accurately as possible impulse response functions either in frequency domain or in time domain variables. It is perhaps the most important software task because all the subsequent system realizations and the determination of structural model parameters do depend on the extracted impulse response data. About a fourth of this course will be devoted to learn methods and techniques for extracting the impulse response functions. System realization performs the following task: For the model problem of plant: x = A x + B u Given measurements of output: y = C x + D u input: u Determine system characteristics: A, B C and D Structural modeling block is to extract physical structural quantities from the system characteristics or realization parameters (A, B, C, D). This is because 164

165

Disturbances

Structures

Sensors
Signal Conditioning

Filtered

Data

FFTs Wavelets

Impulse

Response

Realization x = Ax + Bu y = Cx + Du

Realization Parameter

?
Monitoring

16.2 STRUCTURAL MODELING BY SYSTEM IDENTIFICATION

165
Instrumentation
Repair Design modification Accident investigation

Model
Update

Model-based Diagnostics/ Health monitoring Active tayloring, ...

FEM/ Active Control Models

Structural Models, M (mass) D (damping) K (stiffness) Nonlinearities

Control, vibration isolation design, damage detection, ...

Chapter 16: METHODS FOR VIBRATION ANALYSIS

166

realization characteristics still consist of abstract mathematical models, not necessarily in terms of the desired structural quantities. Specically, one obtains

Given realization parameters: A, B, C, and D Determine either modal quantities: modes() and mode shapes () or physical matrices: mass (M), stiffness(K) and damping(D)

Finite element model updating, active controls and health monitoring are the beneciaries of the preceding four activities. Hence, we will try to touch upon these topics, perhaps as term projects, depending on how this course progresses itself before the Thanksgiving recess.

Finally, if necessary, one may have to repeat testing, hopefully this time utilizing the experience gained from the rst set of activities. Even experienced experimentalists often must repeat testing. A good experimentalist rarely believes his/her initial results whereas a typical analyst almost always thinks his/her initial results are valid! 16.3 ANALYTICAL SOLUTION OF VIBRATING STRUCTURES

This section is a starting point of a guided tour, though an incomplete one at best, of modeling, analysis and structural system identication. To this end, we introduce a reference problem, which for our case is an analytically known model so that when we are astray from the tour path, we can all look up the map and hopefully steer ourselves back to the reference point and continue our tour.

166

167

16.3 ANALYTICAL SOLUTION OF VIBRATING STRUCTURES

u1 y1 m1 = 1 k1 = 1 k2 = 10 m2 = 1 k3 = 100 u2 y2 m3 = 1 k4 = 100 u3 y3

Figure 2. Three DOF Spring-Mass System 16.3.1 Three Degrees of Freedom Model Problem The model problem we are going to walk through is a 3-DOF (three degrees of freedom), undamped oscillator given by The system mass and stiffness matrices, M and K are given by M= 1 0 0 0 1 0 0 0 1 (16.1) K= 11 10 0 10 110 100 0 100 200

whose frequencies (the square root of the eigenvalues) are given by = [ 2.991168982 6.875901898 16.271904658 ] rad/sec. (16.2)

and the eigenvectors are given by = 0.974189634 0.224712761 0.199992181 0.815213561 0.104678951 0.533789306 0.021417986 0.543534706 0.839113397 (16.3)

The mode shapes (eigenvectors) are plotted below in Fig. 3. Let us now compute the analytical impulse response functions.

167

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1 0.8 0.6 0.4 Modal Amplitude 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 Third Mode First Mode

168

Second Mode

1.2

1.4

1.6

1.8

2.2

2.4

2.6

2.8

Coordinate (x)

Figure 3. Mode Shapes of Three DOF Spring-Mass System 16.3.2 Impulse Response Functions For a given forcing function f(t), the model equations are given by M q + K q = f(t) (16.4)

where q is the displacement vector of its dimension (3 1). Let us consider the following special matrix forcing function f= (t) 0 0 0 0 (t) 0 0 (t) (16.5)

where the unit impulse function (t) is dened by

(t) dt = 1

(16.6)

Notice we have introduced a matrix-valued forcing function instead of the customary vector-valued forcing function. Hence, the response or output q should be a 168

169

16.3 ANALYTICAL SOLUTION OF VIBRATING STRUCTURES

matrix-valued function. The forcing fucntion or input consists of three separate unit impulse loadings, each applied at one of the three distinct mass locations. This is in fact the most desired testing condition called single input multiple output (SIMO) testing procedure. There are two ways of characterizing the impulse response functions: frequency domain and time doamin characterizations. As we are planning to study both characterizations, we will describe them for the example 3-DOF problem. 16.3.3 Frequency Response Functions In order to obtain the frequency response functions of the model 3-DOF problem (16.3), we rst seek a solution of the form q = q0 e jt Upon substituting into (1.4) one obtains q(t) = (2 M + K)1 f(t) Fourier transformations of both sides of the above expression yield Q() =

(16.7)

(16.8)

q(t)e

jt

dt =

(2 M + K)1 f(t)e jt dt

(16.9)

Since a convolution of any function with the impulse response function (t) is the function itself, we have Q() = H() = (2 M + K)1 (16.10)

There are a total of nine components in the impulse response function H() for this example problem. This can be seen by expanding H() H11 H21 H31 169 H12 H22 H32 H13 H23 H33

H() =

(16.11)

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1 10 1 10 0 10 -1 10 -1 10 -2 10 10-3 -2 10 -3 10 -4 10 -3 10 10-4 10-5 100 10-1 10-2

1610

0 10

H11
-4 10 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

-5 10 -6 10 0

H12
0.5 1 1.5 2 2.5 3 3.5 4 4.5

H13
0.5 1 1.5 2 2.5 3 3.5 4 4.5

10-6 0 101

101 100 10-1 10-2 10-3 10-4 10-5 10-6 0 0 10 -1 10 -2 10 -3 10 -4 10 -5 10 -6 10 0

101 100 10-1 10-2

100

10-1

10-2 10-3 10-4 10-3

H21
0.5 1 1.5 2 2.5 3 3.5 4 4.5 10-5 0 101 0.5 1 1.5

H22
2 2.5 3 3.5 4 4.5

10-4 0 101 100

H23
0.5 1 1.5 2 2.5 3 3.5 4 4.5

100

10-1 10-2 10-3

10-1

10-2

10-4 10-5

10-3

H31
0.5 1 1.5 2 2.5 3 3.5 4 4.5

10-4 0

H32
0.5 1 1.5 2 2.5 3 3.5 4 4.5

10-6 10-7 0 0.5 1 1.5

H33
2 2.5 3 3.5 4 4.5

Figure 4. Frequency Response Functions, H() For example, H11 is the frequency response function at the mass point 1 where the unit impulse is applied. On the other hand, H12 and H13 are due to the unit impulse load at mass points 2 and 3, respectively. In general, Hr s is the frequency response of the r th degrees freedom due to the load applied at the mass point s. Figure 4 shows all of the nine components vs. frequency ( f = /2 ). Notice that we have the symmetry of the following three components: H12 = H21 H23 = H32 H13 = H31 1610

(16.12)

1611

16.3 ANALYTICAL SOLUTION OF VIBRATING STRUCTURES

which can be veried from Fig. 4. This is due to the fact that the unit impulse loads applied at the three mass points are the same, and both the mass and stiffness matrices are symmetric. This property plays an important role in determining the frequency response functions from measured data.

16.3.4 Time-Domain Impulse Response Functions Theoretically, the impulse response function h(t) has to be determined from the following convolution integral q=
t

h(t ) f( ) d

(16.13)

Luckily for the impulse load given by (16.5), one can obtain the time-domain impuse response function h(t) either by an analytical approach such as the transition method and modal superposition technique using the system eigenvalues and their eigenvectors, or by a direct time integration method. If one chooses the analytical transition matrix technique, one rst transforms the second-order systems into a rst-order equation: x=Ax+Bu A= B= u=f x = [v q ]T , v = Mq Solution of the above equation is given by x=e
A(t

0 M1 I 0

K 0 (16.14)

t0 ) x(0) +
t0

eA(t ) B u( ) d

(16.15)

Alternatively, one may invoke the classical modal superposition method. To this end, we introduce q= (16.16) 1611

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1612

Substituting this into the coupled dynamic equation (16.4) gives T M + T K = T f Using the identities T M = I, T K = diag (2 ) (16.18) (16.17)

the modal equation can be expressed as + 2 = 0 with the following modal initial conditions: (0) = 0, (0) = T M1 (16.20) (16.19)

Note that the impulse loads T f is replaced by the equivalent initial velocity condition by the momentum conservation relation q(0) = q(0 ) + tq=M
1 0 t

f d t,

q(0 ) = 0 (16.21)

(0) = 1 q(0) = T q(0) = T M1 Solution of (16.19) with the initial condition (16.20) is given by

sin q(t) = h(t) =

1 t/

0 0

0 sin 2 t/ 0

0 0 sin 3 t/

T
3

(16.22)

Finally, a direct time integration, e.g., by using the trapezoidal rule can give an approximate solution in the form 1612

1613

16.4 DISCRETE IMPULSE RESPONSE FUNCTIONS

Given the initial condition and: = dt/2, 2 = S = M + 2 K Start integration: t = t + 2 g = M(qn + qn ) qn+ 2 = S1 g qn+1 = 2qn+ 2 qn qn+1 = M1 Kqn+1 qn+1 = qn + (qn+1 + qn ) End the integration
1 1

(16.23)

The time-domain impulse response functions are given by q: h11 h21 h31 h12 h22 h32 h13 h23 h33

h(t) = q(t) =

(16.24)

Notice also that we have again the following symmetry from Fig. 5 that corresponds to the result obtained by the frequency method (16.12): h12 = h21 h23 = h32 h13 = h31

(16.25)

16.4 DISCRETE IMPULSE RESPONSE FUNCTIONS In the preceding section we have obtained the reference impulse response functions both in frequency and time domains when the Fourier transforms can be carried 1613

Chapter 16: METHODS FOR VIBRATION ANALYSIS

0.4

1614
0.05

0.1

0.3

0.2

0.05

0.1

-0.1

-0.2

-0.05

-0.3

h11
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

-0.4

h12
-0.1 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

h13
-0.05 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

0.1

0.15

0.1

0.1
0.05

0.05
0.05

-0.05
-0.05

-0.05
-0.1

h21
-0.1 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

h22
-0.15 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

h23
-0.1 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

0.05

0.1

0.1

0.05

0.05

-0.05

-0.05

h31
-0.05 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

h32
-0.1 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

h33
-0.1 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Figure 5. Time-Domain Impulse Rsponse Functions , h(t) out exactly. In practice both the forward and inverse Fourier transforms are caried out by Fast Fourier Transform (FFT) procedure. In order to appreciate the impact of discrete approximations on the solution accuracy, let us consider the rst solution component of the modal response equation (16.19) given by (t)1 = sin(2 f 1 t)/(2 f 1 ), f 1 = 4.760593E-01 (16.26)

1614

1615 16.5 IMPULSE RESPONSES WHEN LOADINGS ARE ARBITRARY

0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 6 5 4 3 2 1 0 0 5 10 15 20 25 30 35 Incorrectly Sampled Case 0 5 10 15 20 25 30 35

0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 6 5 4 3 2 1 0 0 5 10 15 20 25 30 35 Correctly Sampled Case 0 5 10 15 20 25 30 35

Figure 6. Fourier Transforms of sin(2 f t) The only difference between the incorrectly sampled vs. correctly cases is the number of samples used, 32 vs. 34. The nonzero magnitude in the incorrectly samped case is known as leakage phenomenon in FFT-based data processing. This and related ltering techniques become an integral part of system identication activities, which we plan to cover in the subsequent sections. 16.5 IMPULSE RESPONSES WHEN LOADINGS ARE ARBITRARY In the preceding sections we have obtained the reference impulse response functions both in frequency and time domains when the loadings are the exact unit impulse 1615

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1616

function. This is indeed a luxuary that is difcult to realize in practice. In a laboratory testing environment (we will address the in-situ loads later), one creates a series of ramdom bursts that are rich in frequency contents. To simulate such loadings, fortuntely using MATLAB c , one can generate them by invoking randn command: f1 = 5 * randn(1,512) (16.27) f2 = 10 * randn(1,512) f3 = 100* randn(1,512) It should be noted that the three random number sets in the above command would in principle be different, since at each time Matlab generates a new set of random numbers! Applying the above three forcing functions, the transient response of the three degrees of freedom oscillator can be obtained as shown in Figs. 7 and 8.

The frequency domain system description of the 3dof oscillator for this forcing function case can be obtained by using (16.12):
jt

Q() =

q(t)e

dt =

(2 M + K)1 f(t)e jt dt (16.28)

Q() = H() f()

In the above expressions, is the frequency-domain counterpart of the time convolution integral. The impulse response function H() can thus be obtained, only symbolically, by H() = [Q() fT ()] [f() fT ()]1 = conjugate of (16.29)

This task, especially with experimental data, remains a challenge. A good portion of this course will be devoted to the extraction of accurate frequency response functions from multi-input and multi-output (MIMO) systems.

1616

1617
15 10 5 0 -5 -10 -15 -20 0 30 20 10

16.6 A CLASSICAL IDENTIFICATION METHOD

f1

100

200

300

400

500

600

Time

f2

-10 -20 -30 -40 0 100 200 300 400 500 600

Time
400 300 200 100

f3

-100 -200 -300 0 100 200 300 400 500 600

Time

Figure 7. Randomly Generated Forcing Functions 16.6 A CLASSICAL IDENTIFICATION METHOD Suppose that we have obtained the impulse response functions either in frequency domain (Fig. 4) or in time domain (Fig. 5) from measurement data. It should be noted that the transient response histories under random forcing functions shown in Fig. 8 are typical of measured data obtained by vibration tests. We now want to obtain the modes and the mode shapes from the impulse response function H() (16.11) or h(t) (16.22).

1617

Chapter 16: METHODS FOR VIBRATION ANALYSIS

0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 -0.5 0 0.15 0.1 0.05 0 -0.05 -0.1 -0.15 0 0.06 0.04 0.02 0 0 -0.02 -0.04 -0.06 -0.08 0 1 2 3 4 5 6 -0.05 -0.1 -0.15 0 1 2 3 4 5 6 0 -0.5 -1 -1.5 0 1 2 3 4 5 1 2 3 4 5 6 0.15 0.1 0.05 1 2 3 4 5 6 0.15 0.1 0.05 0 -0.05 -0.1 -0.15 0 1 2 3 4 5 6 1.5 1 0.5 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 -0.1 -0.12 0 1 2 3 4 5 6 1.5 1 0.5 0 -0.5 -1 -1.5 0 1 2 3 4 5 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 -0.5 0 1 2 3 4 5

1618

Figure 8.Transient Rsponse of 3-DOF System under Random Forcing Functions 16.6.1 Frequency Method The frequency response function (16.10) can be reexpressed by using (16.18) as H() = (2 M + K)1 = (2 T + 2 T )1 = ( 2 + 2 )1 T Therefore, the component-by-component frequency response function H()i j is given by 1618 (16.30)

1619

16.6 A CLASSICAL IDENTIFICATION METHOD

Hi j () =
k=1 N

ik jk ( (
2 k

2 )

=
k=1

k Ai j 2 2 k )

(16.31)

The denominators in the above equation play an important role in the identication of vibrating structures. In this course it will be called a modal constant or modal residual constant, given by Modal (Residual) Constant:
k Ai j

= ik jk

(16.32)

Observe that, for multi-degrees of freedom systems, each modal constant magnitude in the modal series expansion of H()i j dictates the inuence of that mode. For example, one may truncate certain modal contributions if their modal constants are sufciently small. As an example, let us compute the three modal constants of H11 () from the 3-DOF example problem given by
N

H11 () =
k=1 N

1k 1k ( 2 2 ) k (
k A11 2 2 k )

(16.33)

=
k=1

where k Ai j can be obtained from (16.3) as

1 A11 2 A11 3 A11 2 = 11 = 0.9741896342 = 9.490454429E-01 2 = 21 = 0.2247127612 = 5.049582495E-02 2 = 31 = 0.0214179862 = 4.587301242E-04

(16.34)

For convenience, Fig. 4a is reproduced below for discussion. 1619

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1620

101 100

Amplitude (H ) 11

-1 10

-2 10 -3 10

H11
10-4 0 0.5 1 1.5 2 2.5 Frequency(Hz) 3 3.5 4 4.5

Figure 9. Impulse Frequency Function H()11 Notice that the fact that 3 A11 = 4.587301242E-04 is small compared with the other two modal constants. This is reected as a small blip in Fig. 9 at the frequency 3 = 16.2719/2 = 2.5898 H z. Now suppose that we have obtained the impulse frequency response curves H()i j from measurements. The rst step in constructing the experimentally determined model is to obtain the modal constants k A11 . To determine the three modal constants 1 A11 , 2 A11 and 3 A11 , one constructs the following matrix relation for all the discrete points min max : 1620

1621

16.6 A CLASSICAL IDENTIFICATION METHOD 2 min )1 2 ( 2 2 )1 3 2 ( 2 3 )1 3

2 3 2 3 2 max )1

2 2 ( 1 min )1 H11 (min ) 2 H11 (2 ) ( 2 2 )1 1 2 H11 (3 ) ( 2 3 )1 1 = . . . . 2 2 H11 (max ) ( 1 max )1 Hv (n 1) = R(n 3) A(3 1) 11 where Hv 11 denotes that {H11 (),

2 min )1 2 ( 2 2 )1 2 2 ( 2 3 )1 2 2 2

1 A11 2 A11 3 A11

2 2

2 max )1

min

(16.35) max } is arranged as a vector.

A least squares solution of the above overdetermined equation would give the sought-after modal constants for H()11
1 A11 2 A11 3 A11

= [RT R]1 [RT Hv ] 11

(16.36)

The modal constants for the remaining eight frequency response components can be similarly determined. Finally, the mode shape which has nine unknowns can be solved by a nonlinear programming method once the eighteen modal constants are determined (Remember we have six distinct frequency response functions Hi j , namely, H11 , H12 , H13 , H22 , H23 and H33 . And each has three modal constants). 16.6.2 A Classical Time Domain Method There exist several classical time domain methods for determining the modes and mode shapes. Among a plethora of classical methods, we will study Ibrahims method as it can be viewed as a precursor to modern realization algorithm based on Kalmans minimum realization procedure. The transient response for a rst-order system x(t), see (16.14), can be expressed from (16.15) 0 K (16.37) x(t) = eAt x(0), A= 1 0 M 1621

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1622

Since we have A = 1 , e1 . = . e . . . . . . . . . . . . eA = e 1 . . . . . (16.38)

e2 . . . .

. . .

e n

the coupled transient response x(t) can be expressed in terms of the rst-order eigenvalue and the associated eigenvector : x(t) = et 1 x(0) = et z(0), which for the example 3-DOF system becomes x(t) = et (6 1) (6 6) (6 6) z(0) (6 1) (16.40) x(t) = z(t) (16.39)

Therefore, for n-discrete response points one has the following relation:

[ x(t1 )

x(t2 ) . . . x(tn ) ] = [et1 ] [et2 ] X(t1 ) = (t1 ) z(0)

. . . [etn ]

z(0) (16.41)

Now if we shift the starting sampling event from t = t1 to t2 = t1 + t, we have a time-shifted equation [ x(t2 ) x(t3 ) . . . x(tn+1 ) ] = [et2 ] [et3 ] X(t2 ) = (t2 ) z(0) . . . [etn+1 ] z(0)

(16.42)

1622

1623 Now observe that [eti+1 ] can be decomposed as [eti+1 ] = E( t) [eti ] e1 . E( t) = . . .

t

16.7 DISCUSSIONS

e 2 . . .

. . . . . . . . . . . . . . .

e 6

. . . .
t

(16.43)

Using this decomposition and (16.41), X(t2 ) in (16.42) can be written as

X(t2 ) = (t2 ) z(0) = E( t) (t1 ) z(0) = E( t) 1 X(t1 ) (16.44) X(t2 ) = S( t) X(t1 ) S( t) = E( t) 1

Finally, one obtains S( t) = [X(t2 ) XT (t1 )] [X(t1 ) XT (t1 ) ]1 (16.45)

Once S( t) is determined, the system eigenvalues and eigenvectors can be obtained from the following eigenproblem: S( t) = E( t) (16.46)

It should be noted that the resulting eigenvectors are not scaled in general, especially with respect to the system mass matrix. This and other issues will be discussed in the following chapters. 1623

Chapter 16: METHODS FOR VIBRATION ANALYSIS

1624

16.7 DISCUSSIONS As stated in Section 16.2, structural analysis is rst to construct the structural models, viz., the mass, the damping, and stiffness operators. Then by using known forcing functions or best possible forcing function models, it is to determine the response of the structure. Structural system identication on the other hand is to determine the structural model or model parameters based on the measured forcing functions and the measured structural response data. More precisely, let us consider a general system model given by (we will derive them in the next chapter) x(t) = A x(t) + B u(t) y(t) = C x(t) + D u(t) (16.47)

Structural analysis is to obtain x for a given u knowing that A and B are known by modeling work, usually by the nite element method or the boundary element method. Structural system identication is to obtain A along with the remaining three opeartors B, C and D, when u and the sensor output y are available. In other words, structural analysis is to obtain vectorial quantities from known matrix quantities, whereas structural system identication is to obtain matrix quantities from measured vectorial quantities. Therefore, the structural response can be obtained uniquely once the structural model (structural mass, stiffness and damping matrices) and the forcing function are given. This is not the case in general in determining the structural model from measured forcing function and measured response data. This is because the information (usually frequency contents) contained in the measured response can vary widely, depending on the sensor types, sensor locations, and forcing function characteristics. For example, if the forcing function consists of two tuned harmonic frequencies, the measured response would capture at most two modes by tuning the forcing function frequencies to match two of the system modes. This is why one prefers to utilize forcing functions that contain rich enough frequency contents. This and other related issues will be discussed throughout the course.

1624

A Solution of Two-Cable Systems: Continuum Models

17

171

Chapter 17: A SOLUTION OF

TWO-CABLE SYSTEMS:

CONTINUUM MODELS

172

17.1 INTRODUCTION Lets consider the following scenario: A construction company has been designing a cable that will run between two towers in a mountain. After a preliminary analysis and a scale model test, a technical evaluation team concluded that a third exible tower may have to be placed. To this end, we will need to formulate the equations of motion for the cable system complete with the appropriate boundary conditions. Assume that the original two supporting towers are substantially stiffer than the new one to be placed between the two end-supports of the original cable system. The design team has concluded that they should model not only the exibility but also the inertia effect of the middle tower. In addition, the new system is to have its fundamental frequency about 50% higher than the original system (which was the reason why a third tower is needed). Observe that the cable tension, T , of the original system could not be increased any further as that will result in an unacceptable bending on the original two towers. We will now proceed the formulation steps as follows.
z w(x,t)
M0

L /2

z
f(x,t)
Mm

L /2
ML

K0
O

x
O

Km

Figure 17.1 Cable with a middle support There are two ways of selecting the coordinate system. Once is to use one coordinate system for the entire system. An alternative is to employ two independent coordinate systems, one for the left cable and another for the right cable as shown in Figure 17.1 above. In what follow, two independent coordinate systems will be adopted.

17.2 GOVERNING EQUATIONS OF MOTION. Introduce w(x, t) and w(x, t) whose domains are dened by For the left cable: w(x, t), 0 x For the right cable: w(x, t), x L , 172 (17.1)

= L/2

173 Using these distinct displacements, we have

t2 t1

17.2

GOVERNING EQUATIONS OF MOTION.

[L + Wnoncons ]dt = 0, L = T V

T =
0

1 (x)w2 (x, t) 2

dx +

1 (x)w2 (x, t) 2

dx (17.2)

+ 1 M0 w 2 (0, t) + 1 M L w2 (L , t) + 1 Mm w 2 ( , t) 2 2 2 V =
0 1 T (x) 2 2 wx (x, t)

dx +

1 T (x) 2

2 wx (x, t) d x

+ 1 K 0 w2 (0, t) + 1 K L w2 (L , t) + + 1 K m w2 ( , t) 2 2 2
t2 t1

Evaluation of
t2 t1

T dt
t2 t1

T dt =

{[
0

(x)w(x, t) w(x, t)d x ] + [

(x)w(x, t) w(x, t)d x ]

(17.3)

+ M0 w(0, t) w(0, t) + M L w(L , t) w(L , t) + Mm w( , t) w( , t)} dt

Evaluation of V V =
0

T (x) wx x (x, t)w(x, t) d x

T (x) wx x (x, t)w(x, t) d x (17.4)

+ { [T (x) wx (x, t)w(x, t)]x= [T (x) wx (x, t)w(x, t)]x=0 } + { [T (x) wx (x, t)w(x, t)]x=L [T (x) wx (x, t)w(x, t)]x= } + [K 0 w(x, t)w(x, t)]x=0 + [K L w(x, t)w(x, t)]x=L + [K m w(x, t)w(x, t)]x= Virtual work due to nonconservative force f (x, t) Wnoncons = 0 due to free vibration problem of interest. Substituting (17.5), (17.4) and (17.3) into (17.1), one obtains the following Hamiltons principle: 173 (17.5)

Chapter 17: A SOLUTION OF

t2 t1 0 L

TWO-CABLE SYSTEMS:

CONTINUUM MODELS

174

{[(x)w(x, t) + T (x) wx x (x, t)] w(x, t)}d x {[(x)w(x, t) + T (x) wx x (x, t)] w(x, t)}d x

{[T (x) wx (x, t) + K 0 w(x, t) + M0 w(x, t)]w(x, t)}x=0 {[(T (x) wx (x, t) + K m w(x, t) + Mm w(x, t)]w(x, t)}x= {[T (x) wx (x, t)]w(x, t)}x= {[(T (x) wx (x, t) + K L w(x, t) + M L w(x, t)]w(x, t)}x=L dt = 0 Since w(x, t)|x= and w(x, t)|x= are the same, viz., w(x, t)|x= w(x, t)|x= = 0 the fourth and fth rows in (17.6) must be concolidated to read: {[(T (x) wx (x, t) + K m w(x, t) + Mm w(x, t)]w(x, t)}x= {[T (x) wx (x, t)]w(x, t)}x= = {[(T (x) wx (x, t) + K m w(x, t) + Mm w(x, t) T (x) wx (x, t)]w(x, t)}x= = {[(T (x) wx (x, t) T (x) wx (x, t) + K m w(x, t) + Mm w(x, t)]w(x, t)}x= Substituting this into (17.6) leads to the following variation equation:
t2 t1 0 L

(17.6)

(17.7)

(17.8)

{[(x)w(x, t) + T (x) wx x (x, t)] w(x, t)}d x {[(x)w(x, t) + T (x) wx x (x, t)] w(x, t)}d x

{[T (x) wx (x, t) + K 0 w(x, t) + M0 w(x, t)]w(x, t)}x=0 {[(T (x) wx (x, t) T (x) wx (x, t) + K m w(x, t) + Mm w(x, t)]w(x, t)}x= {[(T (x) wx (x, t) + K L w(x, t) + M L w(x, t)]w(x, t)}x=L dt = 0 The preceding Hamiltons equation yields two governing differential equations: 174

(17.9)

175 The governing equation of motion:

17.3

VIBRATION PROBLEM

(x)w(x, t) = T (x) wx x (x, t), 0 x , (x)w(x, t) = T (x) wx x (x, t), The three natural boundary conditions: xL

= L/2

(17.10)

{[T (x) wx (x, t) + K 0 w(x, t) + M0 w(x, t)]w(x, t)}x=0 = 0 {[(T (x) wx (x, t) T (x) wx (x, t) + K m w(x, t) + Mm w(x, t)]w(x, t)}x= = 0 {[T (x) wx (x, t) + K L w(x, t) + M L w(x, t)]w(x, t)}x=L = 0

(17.11) (17.12) (17.13)

Together with the above three natural boundary conditions, one must augment them with the continuity condition at the middle support given by (17.7), which provide the critical fourth constraint condition. The three natural boundary conditions plus the continuity condition at the middle support: [T wx (x, t) + K 0 w(x, t) + M0 w(x, t)]x=0 = 0 [T wx (x, t) T wx (x, t) + K m w(x, t) + Mm w(x, t)]x= = 0, [T wx (x, t) + K L w(x, t) + M L w(x, t)]x=L = 0 [w(x, t)|x= w(x, t)]x= = 0 with T = T (x) = T (x) 17.3 VIBRATION PROBLEM The general form of solution from the governing two equations of motion (17.10): w(x, t) = F(t) W (x), W (x) = [C1 sin x + C2 cos x], 0 x w(x, t) = F(t) W (x), W (x) = [C3 sin x + C4 cos x], x L F(t) = fe jt = L/2 (17.14)

(17.15)

When (K 0 , K L ), by substituting this condition and the general solution form (17.15), we nd: 175

Chapter 17: A SOLUTION OF

TWO-CABLE SYSTEMS:

CONTINUUM MODELS

176

W (0) = 0 [T W (x)x T W (x)x + K m W (x) 2 Mm W (x)]x= = 0, W (L) = W (2 ) = 0 [W (x)|x= W (x)]x= = 0 Lets focus on the second equation of the above equation set. First, dividing by the tension T and multiplying by , we have [ W (x)x Second, we note that with 2 = W (x)x +
2 T 2

= L/2 (17.16)

Km 2 Mm W (x) W (x)]x= = 0 T T

(17.17)

2 Mm 2 Mm = Mm = T T 2 = m Mm m = , = , Third, introducing

2 Mm = T = L/2

2 =

Mm ( )2 (17.18)

m = equation (17.17) can be written as [ W (x)x Therefore, (17.16) simplies to W (0) = 0 [ W (x)x

Km T

(17.19)

W (x)x + m W (x) m 2 W (x)]x= = 0

(17.20)

W (x)x + m W (x) m 2 W (x)]x= = 0 (17.21)

W (L) = W (2 ) = 0 [W (x)|x= W (x)]x= = 0 176

177 Finally, using

17.3

VIBRATION PROBLEM

W (x) = C1 sin(x) + C2 cos(x), 0 x W (x) = C3 sin(x) + C4 cos(x), x L

(17.22)

the preceding boundary and constraint set (17.21) leads to the following characteristic equation:
0 1 [ cos() + (m m 2 ) sin()] [ sin() + (m m 2 ) cos()] 0 0 sin() cos() 0 0 C1 0 cos() sin() C2 0 = sin(2) cos(2) 0 C3 sin() cos() 0 C4 (17.23)

By successive manipulations of the third and fourth columns, it can be shown that the above equation simplies to
0 1 [ cos() + (m m 2 ) sin()] [ sin() + (m m 2 ) cos()] 0 0 sin() cos() AC = 0 (17.24) 0 sin() 0 0 C1 0 0 C2 0 = cos() 0 C3 1 0 C4

This simplication is equivalent to taking the left and right cable coordinates independently for both cases as (0 x , 0 x ). Setting det(A) = 0, one nds with m = m =
Mm :

Km cos( ) = ( )2 2( ) T sin( )

(17.25)

Since the desired frequency is L = 1.5 we nd the spring constant to be m = Km = 10.26404145599745 T (17.27) = 1.5/2 (17.26)

The fundamental mode shape is plotted in Figure 17.2 using the computation routines listed below. 177

Chapter 17: A SOLUTION OF

TWO-CABLE SYSTEMS:

CONTINUUM MODELS

178

Two cables supported by a flexible middle suppor t 0.4

0.2 middle spring (km) =10.264 0 beta*L = 4.7124

Mode shape

-0.2

-0.4

-0.6

-0.8

-1

0.1

0.2

0.3

0.4

0.5 0.6 Beam span

0.7

0.8

0.9

Figure 17.2 Fundamental mode of a cable with a middle support

% Computations of two cable vibrations % using a general characteristic equation % clear all; % % Km determined from the det(C_cable) % km=10.26404145599745; % % compute mode shape for this frequency % beta_ell = 1.5*pi/2; % target frequency % compute C_cable (4x4) matrix Ccable = CmatrixTwoCable(km, beta_ell); % it is assumed that the minor Cbeam(3x3) matrix is non-singular Cminor = Ccable(2:4, 2:4); b=-Ccable(2:4,1); % mode shape coefficients assuming c_4 =1 coef = Cminor\b; xL_coord=zeros(1,0); yL_coord =zeros(1,0); xR_coord=zeros(1,0); yR_coord =zeros(1,0);

178

179
for i=0:100, span = i/100; arg =beta_ell*span; sx = sin(arg); cx = cos(arg); WL = sx + coef(1)*cx; WR =coef(2)*sx + coef(3)*cx; xL_coord = [xL_coord span]; yL_coord =[yL_coord WL]; xR_coord = [xR_coord span]; yR_coord =[yR_coord WR]; end; x_coord=0.0:0.005:1.00; y_coord=[yL_coord(:,1:100)

17.4

A COUNTRYSIDE ENGINEERS MODEL

yR_coord ];

%normalize y_coord y_coord = -y_coord/(max(abs(y_coord))); figure(1); plot(x_coord, y_coord); xlabel(Beam span); ylabel(Mode shape ); legend([ middle spring (km) =, num2str(km), beta*L = , num2str(2*beta_ell)]); title(Two cables supported by a flexible middle support); % end of the program % % compute the determinant of two-cable with middle support % function [C_cable] = CmatrixTwoCable(km, beta_bar); % characteristic matrix for two cable with middle flexible support %vibration problem; % the two end supports are fixed %input beta = beta*L/2; km = middle spring, m is assumed %to be half of the cable weight. % %output : C_cable (4x4) %%%%%%%%%%%%%%%%%%% % data preparation %%%%%%%%%%%%%%%%%%% c = cos(beta_bar); s = sin(beta_bar); % C_cable =[ 0 1 0 0; beta_bar*c+km*s-beta_bar^2*s -beta_bar*s+km*c-beta_bar^2*c -beta_bar 0; 0 0 s c; s c 0 -1]; return;

179

Chapter 17: A SOLUTION OF

TWO-CABLE SYSTEMS:

CONTINUUM MODELS

1710

17.4 A COUNTRYSIDE ENGINEERS MODEL Suppose that you are in the countryside without access to advanced modeling tools and you are asked to design the same task. If that individual happens to be this instructor, here is an approach. Step 1: Model each half of the cable a lumped mass-spring system (M, K ) such that the by fundamental frequency is 1 = (2) = K /M. This is illustrated in Figure 17.3.
Left cable Equivalent simple model
Mm

Right cable Equivalent simple model

Km

K
M

K
M

Left cable

x1
(1-a) K

x2
(1-a) K

Mm

xm

Km

The spring distribution factor, a, is chosen such that when Mm = Km = 0, the fundamental frequency is half of the individual frequency, viz., from 2pi to pi. Thus, a is found to be a = 0.75.

Figure 17.3 A simple equivalent model of a cable with a middle support Step 2: Form the left and right cables to be in parallel and attach them to the support system characterized by (Mm , K m ). Step 3: In doing so, distribute the cable springs such that if the middle support is removed, the fundamental frequency of the system would be half of the individual frequency. This is because for cables the fundamental frequency is given by 1 = n L T (17.28)

which states that if the cable length is doubled, the fundamental frequency (n = 1) will be halved. A simple weighted distribution of the cable springs indicates that the distribution factor should be a = 3/4 1710 (17.29)

Right cable

aK

aK

1711

17.4

A COUNTRYSIDE ENGINEERS MODEL

Step 4: Form the model equation set given by: M x1 + K x1 K xm = f 1 4 K M x2 + K x2 xm = f 2 4 K K Mm xm + (K m + )xm (x1 + x2 ) = f m 2 4 0 K 4 =0 (K M 2 ) K 4 K K 2 4 (K m + 2 Mm )

(17.30)

whose characteristic equation is given by (K M 2 ) det 0 K 4

(17.31)

For computational expediency, divide each of the three rows of the above characteristic matrix by M and using n = K /M, we have det
2 (n 2 )

0
2 (n 2 ) 2 4n

0
2 4n

(K m /M +

2 n 4 2 4n 2 n 2

=0 Mm /M )
2

(17.32)

Introducing the following non-dimensional parameters, Km Km K 2 = = m n , M K M equation(17.32) becomes det

2 (n 2 )

m =

Mm M

(17.33)

0
2 (n 2 )

0
2 n

2 n

2 n 4 2 4n 2 2 (m n + 2n

=0 m 2 ) (17.34)

For this characteristic equation, one nds m for the case of = 1.5, with m = 1, to be m = 13.747 which, when compared with the exact solution given by (17.27), yields about 34% off. One possible improvement for estimating the middle support spring is to distribute the spring constant K m as shown in Figure 17.4. 1711 (17.35)

Chapter 17: A SOLUTION OF

TWO-CABLE SYSTEMS:

CONTINUUM MODELS

1712

Left cable

Right cable

Left cable

x1
(1-a) K

x2
(1-a) K

x1
(1-a) K

M
b _ 2

M
Km

x2
(1-a) K

Mm

xm

Mm

xm

Km

(1-b)Km

Possible Improvement of estimating the middle support spring Km by distributing Km into three ways as shown on the right in the above figure.

Figure 17.4 Possible Improvement in modeling the support spring Whether this conjecture will be valid or not is left as an exercise to curious minds. Better yet, one may employ each half of the cable by a two-DOF model, instead of the one-DOF model I used herein. In any case, you now have a glimpse of old-fashioned modeling approaches. Let me know if you would like to learn more about old-fashioned modeling practices.

1712

Right cable

aK

aK

aK

aK

Modeling of Shock Absorbers

18

181

Chapter 18: MODELING OF SHOCK ABSORBERS

182

18.1 INTRODUCTION We have studied the so-called Den Hartog-Ormondroyd 2-DOF shock isolation model (Lecture 02) and discrete modeling of cable models (Lectures 19 and 20). In this lecture we revisit the shock isolation problem in more detail, which illustrates the importance of discrete modeling in the age of FEM world. 18.2 INVARIANT POINTS OF THE DEN HARTOG-ORMONDROYD MODEL Consider a Two-DOF system as shown Figure 18.1. The governing equations for the system is given by M1 x1 + K 1 x1 + k2 (x1 x2 ) + c2 (x1 x2 ) = f 1 (t) m 2 x2 + k2 (x2 x1 ) + c2 (x2 x1 ) = f 2 (t)
B
1

(18.1)

x1

k /2 2
K1 /2
2

Two DOF Spring-Mass-Damper Suspension Model

Figure 18.1: 2-DOF Shock Isolation Model Assume that the excitation acts directly to M1 in the form f 1 (t) = F1 sin(t), The solution of (18.1) assumes x1 (t) = I m(X 1 e jt ), x2 (t) = I m(X 2 e jt ) (18.3) f 2 (t) = 0 (18.2)

Substituting these into (18.1) yields the coupled equations in the frequency domain as (M1 2 + K 1 + k2 + ic)X 1 (k2 + jc)X 2 = F1 (k2 + jc)X 1 + (m 2 2 + k2 + jc)X 2 = 0 182 (18.4)

183 X 1 obtained from the above equation reads:

18.3

SHOCK ABSORBERS

X 1 () (m 2 2 + k2 + jc) = F2 [(M1 2 + K 1 )(m 2 2 + k2 ) m 2 2 k2 ] + jc[K 1 (M1 + m 2 )2 ] Let us parameterize the model as

1

(18.5)

K 1 /M1 , 2 =
1,

k2 /m 2 , c = 2 m 2
1,

1,

= m 2 /M1 ,
1

s = /

xst = F1 /K 1 , f = 2 /

cc = 2m 2

(18.6)

The ratio of X 1 () to xst is the amplication factor from the static to the dynamic response, which can be expressed as (2 ccc s)2 + (s 2 f 2 )2 1 X 1 () =[ c 2 2 ]2 xst (2 cc s) (s 1 + s 2 )2 + [f 2 s 2 (s 2 1)(s 2 f 2 )]2
Frequency Response Functions for Different Damping Ratios of a 2-DOF Suspension Model

(18.7)

101

Invariant points
Amplitude

100

-1 10

101 Frequency (Hz)

Figure 18.2: Frequency Response Function at Mass 1

1 Figure 18.2 plots X x() vs. the driving (or exciting) frequency for various damping coefcients, st i.e., by varying while keeping other parameters constants. Note that there are two points, marked by red dots in the gures, the curves pass through for all damping ratios.

We will call these two points as den Hartogs invariant points and will exploit its properties in two ways: one is to minimize the maximum amplitude resulting in good shock absorbers; and, the other is to maximize its peak amplitude so that the resulting resonators possesses a minimum loss or high-Q performance. 183

Chapter 18: MODELING OF SHOCK ABSORBERS

184

18.3 SHOCK ABSORBERS A good shock absorber should have the following characteristics: (a) The peak amplitude should be insensitive over a wide range of excitation frequencies; (b) The cost for implementing damping should be affordable; (c) The auxiliary mass (m 2 ) should be as light as possible; (d) The response time duration to reach its steady state should be short. We now proceeds with the task of designing a good shock absorber. 18.3.1 Determination of Undamped Natural Frequencies the natural frequencies are determined by setting c = 0 from the denominator of (18.7): [ f 2 s 2 (s 2 1)(s 2 f 2 ) ] = 0 s [1 + (1 + ) f ]s + f = 0
4 2 2 2 2 2 The two distinct roots of this equation will be designated as s1n and s2n .

(18.8)

18.3.2 Determination of Invariant Points Since the two invariant points would play a key role, it is necessary to nd the two coordinates. Of several approaches, we utilize the fact that they are independent of system damping, c. To this end, we rearrange the frequency response function or the magnication factor (18.8) as (2 ccc s)2 A + B 1 X 1 () ] 2 , s = / = H (s, c, f, ) = [ c 2 xst (2 cc s) C + D A=1 B = (s 2 f 2 )2 C = (s 2 1 + s 2 )2 D = [f 2 s 2 (s 2 1)(s 2 f 2 )]2 The damping-independency of the two invariant points implies that at the invariant points the magnication factor satises: C A = B D (18.10) AD = BC 184

(18.9)

185 for which the amplication factor becomes: X 1 () A 1 = H (s, c, f, ) = ( ) 2 xst C

18.3

SHOCK ABSORBERS

(18.11)

The equation for determining the invariant points (18.10) yields two equations: (s 2 1 + s 2 ) 1 = (s 2 f 2 ) [f 2 s 2 (s 2 1)(s 2 f 2 )] 1 (s 2 1 + s 2 ) = (s 2 f 2 ) [f 2 s 2 (s 2 1)(s 2 f 2 )] The rst of the above relation yields the trivial solution, viz., s = /
1

(18.12)

=0

(18.13)

a static case which ifs not of interest for the present discussion. The second of (18.13) can be rearranged as (2 + )s 4 2(1 + f 2 + f 2 )s 2 + 2 f 2 = 0 The two invariant points, s1 and s2 can be shown to satisfy
2 s1 <

(18.14)

1 2 < s2 1+

(18.15)

and the corresponding amplication factors are given by X 1 () 1 = 2 xst |1 (1 + )s(1,2) | 18.3.3 Determination of Amplication Factor at the Invariant Points The foregoing analysis now supplies the coordinates of the two invariant points as P = (x1 , y1 ) = (s1 , 1 ), 2 |1 (1 + )s1 | Q = (x2 , y2 ) = (s2 , 1 ) 2 |1 (1 + )s2 | (18.17) (18.16)

In order to see succinctly the magnication factor, we arrange (18.14) as 1 (1 + )2 f 2 2 z 2 z 1 = 0, 2+ (2 + ) z= 1 s 2 (1 + ) 1 (18.18)

so that the product of the two roots of the above equation, (z 1 z 2 ) is given by 185

Chapter 18: MODELING OF SHOCK ABSORBERS

186

z1 z2 =

2+ 2 =1+
1.

(18.19)

which is independent of the frequency ratio f = 2 /

It should be noted that for a shock absorber the magnication factor should remain small for a wide range of frequencies, which can be realized if we set the amplication factors to attain their maximum at the two invariant points and at the same time their peaks to be the same. The latter e condition can be realized if the coefcient associated with z-terms in (18.18) vanishes: 1 (1 + )2 f 2 = 0 f = 2
1

1 , = m 2 /M1 1+

(18.20)

The amplication factor, (X 1 ()/xst ), for this particular choice at the two invariant points becomes X 1 () 1 = = 2 xst |1 (1 + )s(1,2) | 18.3.4 Determination of Absorber Damping In the previous section, we imposed the condition on the amplication factor at the two invariant points to be the same. However, it is possible that the amplication factor may be larger at other exciting frequencies. Of several possible choices, we impose that the maximum amplication factors for the frequency ranges of interest would not exceed that at the invariant points. This can be carried out as follows. First, we differentiate (18.9) with respect to s = / 1 set the resulting expression to zero: X 1 () ( )[ ]=0 (18.22) s xst for the two invariant points obtained from (18.14).
3 ( +2 ) 2 c 2 2 =( ) = cc 8(1 + )3 1

1+

(18.21)

For s = s1 : For s = s2 :

2 = (

c 2 ) = cc

1 3 + ( +2 ) 2 8(1 + )3

(18.23)

In practice, we must have one damping value for all the frequency range. To this end, we take the average value to be: = 3 8(1 + )3 (18.24)

186

187 18.3.5 Applications

18.3

SHOCK ABSORBERS

First, in order to utilize Matlab capability, we express (18.5) in its transfer function form: (m 2 s 2 + k2 + cs) X 1 (s) = F2 [(M1 s 2 + K 1 )(m 2 s 2 + k2 ) + m 2 s 2 k2 ] + sc[K 1 + (M1 + m 2 )s 2 ] where s is now the Laplace Transform variable. The preceding equation can be arranged to read: (a0 s 2 + a1 s + a2 ) X 1 (s) = 2 1 xst (b0 s 4 + b1 s 3 + b2 s 2 + b3 s + b4 ) 2 a0 = 1, a1 = 2 1 , a2 = 2 b0 = 1, b1 = 2(1 + )
1,

(18.25)

(18.26)
3 1

b2 =

2 1

2 + (1 + )2 , b3 = 2

Proof of two invariant points for den Hartog oscillator

Q
Amplitude Ratio (X1 ()/xst )

1=10, 2 = 9.5, =0.01

10
1

Invariant points P

10

s1
Frequency ( ) (Hz) 10
1

s2

Figure 18.3: Invariant Points of a Den Hartog Oscillator

1 (s) The amplication factor ( Xxst ) vs. the driving frequency () is illustrated in Fig. 18.3, with ( 1 = 10(H z), 2 = 9.5(H z), = 0.01). Observe the two invariant points at which the amplication factors are different, namely, at s1 and s2 points determined by (18.14) and its magnitudes given by (18.17).

187

Chapter 18: MODELING OF SHOCK ABSORBERS

188

Optimized Amplitude for den Hartog oscillator

= 0.1206
Amplitude Ratio (X1 ()/xst )

= 0.03015
10
1

opt = 0.0603

1=10, 2 =9.5, =0.01

10

Frequency ( ) (Hz) 10

Figure 18.4: Den Hartog suspension model optimized for a wider range of frequencies

When the amplication factor is optimized to have its peak at the invariant points by selecting the frequency ratio according to (18.20) and the damping ratio according to (18.24) (zeta = 0.0603), its amplication attains its maximum at the invariant points. This is illustrated along with two non-optimal damping coefcients in Fig. 18.4. Observe that the optimum amplication factor at the invariant points are about 13.5, which may not be acceptable. Figure 18.5 illustrates the reduction of the peak amplication factor by about half by increasing the small mass from m 2 = 0.01M1 to ve times (m 2 = 0.05M1 ). In practice, a suspension system is designed to operate far smaller then the invariant points or sufciently larger then the invariant point frequencies. Under this scenario the infrequent disturbances close to the invariant points can still be bounded. 188

189

18.3

SHOCK ABSORBERS

Optimized Amplitude for den Hartog oscillator

Amplitude Ratio (X1 ()/xst )

10

1=10, 2 =9.5, =0.05

10

Frequency ( ) (Hz)

10

Figure 18.5: Den Hartog suspension model optimized for a wider range of frequencies
Two-DOF Vehicle Shock Isolation Model x1
M1

x2

C 1

K /2 1

m2
K2

Ground input : xg (t)

Figure 18.6: A Two-DOF Vehicle Suspension Model The two-DOF suspension model that we have been studying so far is applicable to machinery shock isolation design. For vehicles subject to road roughness conditions, a modied model is more appropriate as shown in Fig. 18.6. An analysis of this model for achieving an optimum shock isolation is left for an exercise. 189

Modeling of Resonators

19

191

Chapter 19: MODELING OF RESONATORS

192

19.1 A GENERIC RESONATOR A second example where simplied discrete modeling has been found valuable is in the assessment of the performance of micro-electro-mechanical system (MEMS) resonators. Of several available resonator models, we will consider a double-beam resonator. When the input beam which acts as a conductor is placed under alternating voltage eld, the input beam vibrates with the exciting voltage frequency, possibly with all the frequency ranges of the input frequency. The motion of the input beam then triggers motion of the output beam through the link element as shown in Fig. 19.1. In doing so, the link acts as a lter and triggers the output beam to resonate with its fundamental frequency, thus extracting an almost single-tone frequency. Consequently, the double beam resonator has an inherent frequency ltering capability.
Input (Receiving) Resonator

Substrate

Ls Thinkness: h bs b

Output (Transmitting) Resonator L

Figure 19.1: A Double Beam Resonator 19.2 ONE-DOF MODELING OF A SINGLE BEAM Consider one of the resonator with xed-xed ends. As we are interested in modeling the beam with one degree of freedom, we sample the displacement at the beam center and represent the motion of the beam center with an equivalent mass and stiffness. There are two ways of reducing the beam dynamics: order reduction via the nite element method and mass and stiffness lumping from the continuum beam equation. In this section we will employ the continuum beam solution and adapt it to the present task. To this end, we recall that the transverse beam displacement, w(x, t), is expressed as w(x, t) = C1 (t) sin x + C1 (t) cos x + C3 (t) sinh x + C4 (t) cosh x ( L) = 4.71 (19.1)

The three unknown coefcients (C2 , C3 , C4 ) can be expressed in terms of C1 by applying the bounday conditions w(L , t) w(0, t) = w(L , t) = =0 (19.2) w(0, t) = x x 192

Substrate

193

19.3

MODELING OF THE LINKING BEAM

When this is carried out, the transverse displacement reads: w(x, t) = (x) C1 (t) (sin L sinh L) (cos x cosh x) (cos L cosh L) (19.3)

(x) = (sin x sinh x) +

Observe that C1 (t) can be related to the mid-span displacement, w(L/2, t) from the above equation: w(t) = w(L/2, t) = (L/2)C1 (t) N ( L , x) = (x) (L/2) (19.4)

w(x, t) = N ( L , x) w(t),

The above equation can now be used to obtain equivalent stiffness and mass properties for a single degree of freedom model as follows. Observe that the kinetic and potential energy can be obtained by T = 1 m w2 = 1 w[ 2 2 V =
1 k 2 L 0 L 0 2 N x x ( L , x)

A N 2 ( L , x) d x] w (19.5) d x] w

w =
2

1 w[E I 2

Upon evaluating the integrals in the above equation, we nd that kb = 192E I , L3 m b 0.3836 ( AL) (19.6)

It should be noted that the preceding (m b , kb ) closely satises the classical frequency equation. To see this we rst compute the natural frequency of the single degree of freedom model:
2 n = kb /m b = (

192 EI ) 0.3836 AL 4 A 4 L EI

(19.7)

Observing the classical frequency equation ( L)4 = (4.73)4 = 2 (19.8)

192 it is clear the constant ( 0.39 ) in the one degree-of-freedom case should be as close as possible to the continuum case ( L)4 .

192 ) = 500.5214 0.3836

vs.

( L)4 = 500.5467

(19.9)

yielding 0.005 % error, which is quite adequate. It should be emphasized that the mass and stiffness properties obtained in (19.6) are applicable to both resonating beam as they are for most cases identical. The remainder is to model the linking beam and a procedure to couple the linking beam to the two resonating beams. 193

Chapter 19: MODELING OF RESONATORS

194

19.3 MODELING OF THE LINKING BEAM For this case we rst treat the beam ends with the following boundary conditions: 3 w(L , t) w(0, t) w(L , t) 3 w(0, t) = =0 = = 3 3 x x x x (19.10)

which allow a free transverse movement at both ends while the rotations are constrained as shown in Fig. 19.2. The fundamental frequency for this case can be obtained by the continuum beam frequency solution method described in Lecture 13 by setting the end springs as kw1 = kw2 = 0, which yields L = for its fundamental frequency. (19.12) k 1 = k 2 = (19.11)

w x Fundamental Mode Shape of the Link

Figure 19.2: Fundamental Mode Shape of the Link Element It should be noted that the other mode is a translational rigid-body mode. In other words, when the link beam undergoes a rigid-body mode, the two resonating beams move in phase. On the other hand, when the link beam deforms in its fundamental frequency mode, the two resonating beams will move out of phase. Of course for this case, the link beam will experience twisting, which is not considered in the present model. We will revisit this phenomenon later in the analysis of the complete coupled system. One important difference between the linking beam model and the resonator is that the linking beam must be modeled in terms of its two-end displacements without any rotational motion degrees of freedom. This can be done as klink = ks m link = m s 1 1 1 0 1 , 1 0 , 1 ks = 12E Is L3 s = 0.2464

(19.13)

m s = As L s ,

where is a correction factor to t the fundamental frequency of the linking beam, viz., L s = .

194

195

19.4

COUPLING OF THE LINK BEAM TO THE TWO RESONATING BEAMS

19.4 COUPLING OF THE LINK BEAM TO THE TWO RESONATING BEAMS We have completed the individual modeling of the two resonator beams and the linking beam. For illustrative purposes the in-phase and out-of-phase motions of the system are shown in Figs. 19.3 and 19.4. Observe that for the case of in-phase mode the linking beam acts as a rigid link, whereas of out-of-phase mode the linking beam undergoes its fundamental mode shape.

e
(m,k)

Input beam
x

Link beam Output beam

(m,k)
x L

In-Phase Deformed State of Resonator System

Figure 19.3: In-Phase Deformed State of Double Beam Resonator

e
(m,k)

Input beam
x

Link beam e Output beam

(m,k)
L

Out-of-Phase Deformed State of Resonator System

Figure 19.4: Out-of-Phase Deformed State of Double Beam Resonator

An equivalent mass-spring model that accounts for both motions may be represented as shown in Fig. 19.5. The starting point for the development of an equivalent mass-damper-sparing model from the skeletal model is to obtain the kinetic and potential energies of the two systems: 195

Chapter 19: MODELING OF RESONATORS

196

x3

x1 f 1
M1
C 1

f1
C 12 C 2

x4 e L/2

x2
M2
K1

M1
K12

M2
K2

x1 Model Transformtion

x2

Skeletal Physical Model

Equivalent Mass-Damper-Spring Model

Figure 19.5: Physical and Equivalent Mass-Damper-Spring Model

Skeletal Model: Ts = 1 m b x1 + 1 m b x2 + 2 2 2 2
2 2 K s = 1 kb x 1 + 1 kb x 2 + 2 2 1 2 1 2

x3 x4 x3 x4

[m link ]
T

x3 x4 x3 x4

(19.14)

[klink ]

Mass-Damper-Spring Model: 2 2 2 Tmdk = 1 M1 x1 + 1 M2 x2 2 K mdk =

1 K x2 2 1 1

(19.15) +
1 K (x 2 12 1

1 K x2 2 2 2

x2 )

where (m b , kb ) are given by (19.7) and (m link , klink ) are given by (19.14), respectively. Comparing the energy expressions of the two models, it is clear that the degrees of freedom (x3 , x4 ) which represent the transverse displacement at the junctions of the linking beam and the two resonator beams must be substituted by the mid-span displacements of the two beams, (x1 , x2 ). This can be accomplished by using the assumed displacement relation for the resonator beam given by(19.5): x3 = a x1 x4 = a x2 a = N ( L , e) (e) , N ( L , e) = (L/2)

L = 4.73

(L/2) = (sin L/2 sinh L/2) +

(sin L sinh L) (cos L/2 cosh L/2) (cos L cosh L) (sin L sinh L) (cos e cosh e) (e) = (sin e sinh e) + (cos L cosh L) (19.16) 196

197 19.5

A TWO-DOF MODEL AND EVALUATION OF RESONATOR PERFORMANCE

Table 19.1 Link offset amount (e) vs. Link factor (a) Link Offset (e) 0.0625 0.1250 0.1875 0.2500 0.3125 0.3750 0.4375 0.5000 Link Factor (a) 0.00906327886493 0.03946457838609 0.09611178177002 0.18424759908145 0.30988889279730 0.48050748799022 0.70599928850552 1.00000000000000 a2 0.00008214302378 0.00155745294719 0.00923747459501 0.03394717776728 0.09603112587914 0.23088744601468 0.49843499537030 1.00000000000000

The offset amount (e) vs. the link factor (a) for representative ranges are tabulated in Table 19.1 below. Substituting (x3 = ax1 , x4 = ax2 )obtained in the above equation into the kinetic and potential energy expression of the skeletal model (19.14), we obtain Ts =
1 m x2 2 b 1

1 m x2 2 b 2

1 2

x1 x2 x1 x2

a 2 [m link ]
T

x1 x2

x1 2 2 a 2 [klink ] K s = 1 kb x 1 + 1 kb x 2 + 1 2 2 2 x2 The above equation can be simplied by using (m link , klink ) derived in (19.13) as Ts = 1 (m b + a 2 m s )x1 + 1 (m b + a 2 m s )x2 2 2 2 2
2 2 K s = 1 kb x1 + 1 kb x2 + 1 a 2 ks (x1 x2 )2 2 2 2

(19.17)

(19.18)

Comparing (19.18) with the equivalent mass-damper-spring model (19.15), we nd the following model parameters: M1 = M2 = m b + a 2 m s = 0.3836 ( AL) + 0.24638 (a 2 As L s ) K 1 = K 2 = kb = 192E I L3 12E Is ) L3 s 197

(19.19)

K 12 = a 2 ks = a 2 (

Chapter 19: MODELING OF RESONATORS

198

19.5 A TWO-DOF MODEL AND EVALUATION OF RESONATOR PERFORMANCE The quality of a resonator is quantied by the Quality Factor, Q, as dened by 1 2system

Q-Factor =

(19.20)

where system is the damping ratio with respect to the center frequency that is the average of the two peak resonances. It has been found that a major loss source is the energy transmission is from the resonator to the substrate, which is akin to the damping of machinery equipment to the ground due to vibrations discussed in Lecture 22. The equations of motion for the two-DOF mass-damper-model shown in Fig. 22.11 can be derived as M1 0 0 M2 x1 c + c12 + 1 x2 c12 c12 c2 + c12 x1 K 1 + K 12 + x2 K 12 K 12 K 2 + K 12 x1 x2 = f1 0 (19.21)

A word for the physical meaning of the two damping parameters: c1 and c2 act as if the damping is independent of each other, while c12 produces energy loss that is proportional to (x1 x2 ). Experiments of a series of double beam resonators indicate that the system loss is higher for the in-phase mode than the out-of-phase mode. Hence, we will drop c12 in subsequent analyses and set the two damping parameters to be the same

c1 = c2 ,

c12 = 0

(19.22)

Table 19.2 Resonator Model Parameters L b h = hs Ls bs e E

40 m 8 m 2 m 11.8 m 2.7 m 0.25 L 190 GPa 2330 kg/m 3

Computations of the two degrees of freedom model using the model parameters listed in Table 19.2 198

199 19.5 give

A TWO-DOF MODEL AND EVALUATION OF RESONATOR PERFORMANCE

m b = 5.720 1013 ,

m s = 3.658 1014 ,

ks = 2235 N /m

M1 = M2 = 5.7451 1013 kg K 1 = K 2 = kb = 2720.0 N /m K 12 = a 2 ks = 75.868 N /m, a 2 = 0.03395 with e = 0.25L 1 kb /m b = 10.975 (M H z) beam = 2 1 ks /m s = 39.340 (M H z) link = 2 (19.23)

Note that the frequency of the linking beam is about three and half times larger than that of the two beam resonators, thus effectively avoiding mode coupling between them. Figure 19.6 plots the transfer function (or FRF) of the output port, (X 2 ()/(x1 )st ), that is the ratio of the input signal with respect to the output signal ratio vs. the input signal frequency. For illustrative purposes, the damping c1 = c2 is parameterized by c1 = c2 == 2 m b K 1 /m b = [ 0, 0.001, 0.005, 0.01, (19.24)

0.05 ]

Note that the model has two resonant peaks, (10.963, 11.264) M H z . Ideally the closer the frequency separation between the two peaks, the better the ability ofthe resonator to pick up the receiving frequency. A closer-up view of the two peak frequencies is shown in Fig. 19.7.

10

Double Beam Resonator

10

Amplitude Ratio (X2 ()/xst)

10

10

10

10

10

10

0.9

0.95

1.05 1.1 1.15 7 Driving Frequency (Hz) x 10

1.2

1.25

1.3

Figure 19.6: FRF from Input to the Output Signal, H21 () 199

Chapter 19: MODELING OF RESONATORS

1910

10

Double Beam Resonator

10

Amplitude Ratio (X2 ()/xst)

10

10

10

10

10

10

1.07

1.08

1.09

1.1 1.11 1.12 7 Driving Frequency (Hz) x 10

1.13

Figure 19.7: Closer-Up view of H21 () Plot In more realistic high-delity modeling, the lower-frequency peak is a little lower than that of the higher resonance. This is because the lower peak is associated with the in-phase mode and the higher with the out-of-phase mode, which causes less loss through the substrate. This is shown in an elaborate simulation that has been correlated with experimental data as shown in Fig.19.8. It should be noted that in the high-delity simulation, the electrostatic effects, the interaction with the substrate which absorbs the energy as it acts and exible ground, etc. have been modeled in an elaborate way. The high-delity model and its performance assessment is shown in Fig.19.9. In conclusion, a simple two-DOF model can be used to size up the design, peak frequencies, and the position to how to anchor the beams, and the amount of energy loss. It turns out that resonators can also be designed by exploiting the den Hartog invariant points, this time by maximizing its two peaks rather than by minimizing its peak. This is beyond the scope of this course and left as research topics to those interested in the resonator design.

1910

1911 19.5

A TWO-DOF MODEL AND EVALUATION OF RESONATOR PERFORMANCE

Figure 19.8: High-delity model of a resonator and its correlation with experimental data

1911

Chapter 19: MODELING OF RESONATORS

1912

Figure 19.9: A rened model and the performance curves

1912

Reduced-Order Modeling of Vibrating Structures: Introduction

20

201

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

202 INTRODUCTION

20.1 ASSEMBLING SUBSTRUCTURES TO MODEL A TOTAL STRUCTURE Consider a toy airplane model shown in Figure 20.1. Suppose that your job is to develop a structural model by utilizing the substructural models. Specically, manufacturers of the substructures have provided only the dominant modes and their mode shapes, and maybe their modal damping properties.

Fuselage

Wing Tail

Assembled
Fuselage Wings Tail

Figure 20.1 A toy strucuture

2 Mathematically, each of the reduced-order substructural models provided, ((:, 1 : m), {k , k = 1, 2, ...m}), is an approximation of the large-order model given by

K = M for each of the substructures.

(20.1)

In order for us to utilize the reduced-order substructural models for the development of the total system model, it is critical to understand the nature of the reduced-order models. This is addressed below. 20.2 TYPES OF REDUCED-ORDER MODELS The reduced-order substructural models may be categorize according to how the interface degrees of freedom are treated. The modes are called: xed-interface normal modes if all of the interface degrees of freedom u are restrained, free-interface normal modes if none of u is restrained, and hybrid-interface normal modes if part of u are restrained where subscript { } denotes the substructural boundaries. Hence, a typical substructure vibration problem may be expressed as( see Figure 20.2) as KI I KT I KI K
I I

MI I MT I

MI M

(20.2)

where subscript {I } denotes the interior nodes, and the physical substructural displacement,u, is related to the modal displacement,q, according to 202

203 20.3 REDUCTION OF INTERIOR DEGREES OF FREEDOM AND INTERFACE MODELING uI u qI q

u=

(20.3)

uI

Interior nodes

Interface node

Figure 20.2 Substructural DOF Classication The preceding distinction is essential in interpreting the results of each substructural vibrations. For some applications, one invokes loaded-interface normal modes (pre-stressed concrete columns, cables, prebuckled space trusses), which means one augments the interface stiffness K with pre-stress stiffness. In what follows, we will rst discuss various xed-interface modeling, which is followed by free-interfce modeling approach.

20.3 REDUCTION OF INTERIOR DEGREES OF FREEDOM AND INTERFACE MODELING Often, the size of the interior degrees of freedom far exceeds those of the substructural interface degreesof freedom. For this reason, we begin with the reduction of the interior degrees of freedom from the eigenvalue problem: [ KI I ] [ I] = (n I n I ) (n I m I ) m I << n I I [ MI I ] [ I] (n I n I ) (n I m I ) (m I m I )

(20.4)

Referring to (20.3), it is clear that I is only one of the four submatric elements that are needed to relate the modal displacement, q, to the physical displacement, u. We present two techniques to construct the remainder submatrices. It is important to observe that the above process of mode truncation,viz., retaining only m I -modes from n I -modes, implies (20.5) I 0

20.3.1 Quasistatic Constraint Interface Modes We consider a quasistatic equilibrium state in which the reaction force that will produce unit desplacement at the interface u = I is applied. For this case,one has the following equilibrium state: KI I KT I KI K I
I

0 = . (20.6)

Note that the unit interface displacement constitute a unitary matrix, I 203

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

204 INTRODUCTION

Equation(20.6) enables us to obtain

I

= K1 K I I II

= K1 K I II

(20.7)

Finally, since we retain the physical boundary displacemen, u , we have =I Thus, using (20.4) - (20.8), the substructural displacement u can be approximated by uI u
I

(20.8)

K1 K I II I

qI u

(20.9)

Total Structure

A Interface nodes u = u = u B

Substructure A

Substructure B

fixed-interface normal modes, three modes are selected

fixed-interface normal modes, three modes are selected

Constraint modes where unit displacement and rotation are imposed at interfaces

Constraint modes where unit displacement and rotation are imposed at interfaces

Fig. 20.3 Free-free beam partitioned into two substructures (Note the xed-interface modes I and the constrained modes c ) Figure 20.3 illustrates schematically the xed-interface normal modes of each substructure and the constraint modes for a free-free beam partitioned into two substructures. Note that the xed-interface modal amplitudes are zero at the interface while those of the constraint modes are unity, respectively.

204

205

20.4 ENERGY CONSIDERATIONS OF REDUCED-ORDER SUBSTRUCTURAL MODEL

20.3.2 Attachment Interface Modes Instead of applying a unit displacement on , if a unit force is applied to and no force is applied throughout the rest of the structure, we have teh following static equilibrium condition: KI I KT I from which one obtains
I

KI K

0 = I (20.10)

K1 K I II K I

KT K1 K I I II

(20.11)

The substructural displacement u can now be approximated by u uI u =

I

K1 K I KS II KS

qI u

KS = K

KT K1 K I I II

(20.12)

Comparing the constraint mode (20.9) with the above attachment mose (20.12), it is seen that the interface constraint modes are far less expensive to generate than the interface attachment modes.

20.4 ENERGY CONSIDERATIONS OF REDUCED-ORDER SUBSTRUCTURAL MODEL Before we launch on reducing the substructures using the constraint interface modes (20.9) or the attachment interface modes (20.12), it would be instructive to examine the resulting substructural energy expressions due to the approximations. 20.4.1 Approximate substructural energy when using the interface constraint modes The kinetic energy is obtained by T = 1 uT 2
1 2

MI I MT I T

MI M
I

u K1 K I II I Mc I
c T

qI u qI u

MI I MT I qI u

MI M

K1 K I II I

qI u (20.13)

0
T

I MT I

1 2

Mc

Mc = M I I Mc = M +

T I KT I

M I I K1 K I II K1 M I I K1 K I KT K1 M I M I II II II
I

K1 K I II

where u is approximated by the interface constraint modes given by (20.9). 205

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

206 INTRODUCTION

Similarly, the strain energy is obtained by

U = 1 uT 2
1 2

KI I KT I T

KI u K
I

qI u qI u

K1 K I II I 0 KS qI u

KI I KT I

KI K

K1 K I II I

qI u (20.14)

0 I 0

1 2

KS = K

KT K1 K I I II

Observe that the stiffness matrix in the strain energy given by (20.14) is diagonal corresponding to the interior modes, I , and block diagonal corresponding to the interfce degrees of freedom. However, the resulting mass matrix corresponding to the interface degrees of freedom is full as can be seen in (20.13). This indicates that, unless the interface degrees of freedom is relatively small, considerable computations are required to generate the approximate reduced-order substructural kinetic energy expression.

20.4.2 Approximate substructural energy when using the interface attachment modes The kinetic energy is obtained by

T = 1 uT 2
1 2

MI I MT I qI u qI u
T

MI M
I

u K1 K I KS II KS Ma I M
a T

MI I MT I

MI M

K1 K I KS II KS

qI u (20.15)

0
T

I MT a I

1 2

qI u

S T a I [M I I I + M I K ] S aT a M I I KS Mc = I M I +K 1 S a I = KI I K I K

Mc = I

aT I

M I KS KS MT I

a I

where u is approximated by the interface constraint modes given by (20.12). Similarly, the strain energy is obtained by 206

207

20.5 ASSEMBLING REDUCED-ORDER SUBSTRUCTURAL MODELS

U = 1 uT 2
1 2

KI I KT I qI u qI u
T

KI u K
I

K1 K I KS II KS qI u

KI I KT I

KI K

K1 K I KS II KS

qI u (20.16)

0 I 0

0 KS

1 2

KS = K

KT K1 K I I II

Observe that, while the form of its mass matrix is the same as in teh case ofthe interface constraint modes, more computations are required as the terms involve the Schur complement, K S .

20.5 ASSEMBLING REDUCED-ORDER SUBSTRUCTURAL MODELS The previous chapter has devoted to the reduced-order modeling of a single vibrating substructure. Once all of the substructures in a total system are approximated by their corresponding reduced-order models, the next task is to assemble the reduced-order substructural models. Third, the assembled reduced-order total system is either used for performance evaluation and/or design improvements. In practice, the size of the assembled total structural model that consists of reduced-order substructural models is often considdred too large. For such a case, it is customary to carry out additional reduction via total system modal analysis. Figure 20.4 illustrates the sequence of model development in large-scale vibrating structural systems.

Substructure 1 Total structural system Substructure Substructure 21 . . . Partition into substructures Reduce each substructure

Reduced-Order Substructure-1 Model Reduced-Order Substructure-2 Model Assemble into Totoal ReducedOrder System . . . Model development, Performance analysis, Shock isolation design, Model updates, . . . Control synthesis

Substructure s-1 Substructure s

Reduced-Order Substructure-s Model

Fig. 20.4 Sequence of Reduced-Order Modeling and Applications In the following, the variational formulation of partitioned equations of motion will be discussed rst. In particular, two treatment of interface constraints will be discussed: classical (or global) -method (which reads as Lagrange multiplier method), and localized -method. The partitioned equations of motion employing the two -methods are then derived. we will then focus on one of the most widely used component mode 207

Reduced-Order Total structural system

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

208 INTRODUCTION

synthesis method, the Craig-Bampton method. Finally, a component mode synthesis technique based on the localized -methoid will be described.

20.6 VARIATIONAL FORMULATION OF PARTITIONED STRUCTURAL SYSTEMS Consider a structure that consists of two substructures as shown in Fig. 20.5. When the structure is partitioned into two structures, (1) and (2) , interactions forces, (1) and (2) (or (12) ), are developed along the interface boundaries of (1) and (1) . In addition, the displacement for substructure 1 consists of the interior ones u(1) and along the partition boundary u(1) . Similarly, for substructure 2 we have u(2) and u(2) . These can be I I expressed as u(1) u(2) I I u(1) = u(2) = (20.17) (1) , u u(2)

(a) Total Structural System

(1) (2)

Partition

(1)

(2)

(1)

(1)

(12)
1 2

(2)

u(2)
Substructure 2

u(1)
1 Substructure 1

(1)

uf
(1)

(2)

u
2

(2)

Substructure 1

Substructure 2
(1) (2) Constraint: c12 = u u = 0

Constraints:

c1 = u u f = 0 c2 = u(2) u f = 0

(b) Partition modeled by classical -method

(c) Partition modeled by localized -method

Fig. 20.5 Partitioning of a Structure into Two Substructures Using these notations, the energy functionals for substructures 1 and 2 may be written as Substructure 1: Substructure 2:
(2) (1)

= (u(1) )T {K(1) u(1) (f(1) M(1) u(1) )} (20.18) = (u(2) )T {K(2) u(2) (f(2) M(2) u(2) )}

where M and K are mass and stiffness matrix, resprectively, for a substructure, and the superscripts, (1, 2), denote substructure. 208

209

20.7 PARTITIONED EQUATIONS OF MOTION EMPLOYING CLASSICAL -METHOD

While the virtual energy is completely contained in the preceding energy expressions, the interface conditions between these two substructures are needed for partitioning as well as assembly. The kinematic interface compatibility conditionmay be described inone of the two possible ways:

Classical (or Global) form: u(1) u(2) = 0 Localized form: u u(2)

(1)

(20.19) uf = 0

L(1) f L(2) f

which states that the interface displacement along substructure 1, u(1) , must be equal to that of substructure 2, u(2) ; and, L f is the interface displacement displacement operator. The constraint functional that incorporates the above constraints canbe expressed as

Classical (or Global) form: classical = ((12) )T (u(1) u(2) ) Localized form: locali zed = (1) (2)
T

(20.20) { L(1) u(1) f uf } u(2) L(2) f

Finally, the total energy functional is simply the sume of two substructural energy expressions, (20.18), plus one of the the constraint functionals, (20.20):

Classical interface form: Localized interface form:

system system

= =

(1)

+ +

(2)

+ classical + locali zed

(20.21)

(1)

(2)

We now derive the partitioned equations of motion for the two interface treatment cases.

20.7 PARTITIONED EQUATIONS OF MOTION EMPLOYING CLASSICAL -METHOD The total energy of the system for this case which is the one given by the rst of (20.21), consists of the two substructural energy expressions (20.18) plus the interface constraint functional, viz., the rst expression in (20.20), as 209

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

2010 INTRODUCTION

total

= (u(1) )T {K(1) u(1) (f(1) M(1) u(1) )} + (u(2) )T {K(2) u(2) (f(2) M(2) u(2) )} + ((12) )T (u(1) u(2) ) + (u(1) u(2) )T (12) = (u(1) )T {K(1) u(1) (f(1) M(1) u(1) ) + (B(1) )T u(1) } + (u(2) )T {K(2) u(2) (f(2) M(2) u(2) ) (B(2) )T u(2) } + ((12) )T (B(1) u(1) B(2) u(2) ) u(1) = B(1) u(1) , u(2) = B(2) u(2) (20.22)

where B(k) is the Boolean matrix that extracts the interface degrees of freedom at the interface of substructure k. The stationarity of the above variational equation, viz., of motion: M(1) 0 0 M 0 0 0 0 M(2) 0 u (12) K + Bcl
T Bcl total

= 0, yields the following partitioned equations f(1) f(2) 0 (20.23)

0 0 0

u(1) u(2) (12)

K(1) 0 B(1)

0 K(2) B(2)

(B(1) )T (B(2) )T 0

u(1) u(2) (12)

u (12) =

f 0

In the above equation set, the rst row is the equations of motion for substructure 1, the second row for substructure 2, and the third is the interface constraint equation. To illustrate the compositions of the partitioned equation further, we express each of the three equations in terms of the interior degrees of freedom, u I , and the interafce degrees of freedom, u as follows: For substructure 1: MI I M For substructure 2: MI I M
I I

MI M

(1)

uI u

(1)

KI I KI

KI K

(1)

uI u

(1)

= f

f I(1)
(1)

(12)

(20.24)

MI M

(2)

uI u

(2)

KI I KI

KI K

(2)

uI u

(1)

= f

f I(2)
(2)

(12)

(20.25)

If one is interested in constructing the equations of motion for the entire system, then all one has to do is to assemble the coefcient matrices that are associated with u(1) and u(2) into the same rows and the 2010

2011

20.8 PARTITIONED EQUATIONS OF MOTION EMPLOYING LOCALIZED -METHOD

(1)

(2)
Partitioning

(1)

uI

(1)

(2) u (1) u (1)

(2)

u(2) I

(2)

Fig. 20.6 Partitioning of a Structure into Two Substructures columns. This corresponds to an explicit enforcement of the second of the above constraint condition, e.g., u = u(1) = u(1) . The assembled equations of motion therefore becomes (1) (1) (1) (1) (1) (1) (1) MI I KI I fI uI uI MI 0 KI 0 (1) (2) (2) M (1) + M (2) M I u + K (1) I K (1) + K (2) K I u = f (20.26) M I 0 M (2) I
(2) MI I

I u(2)

K (2) I

(2) KI I

u(2) I

f(2) I

This assembly process is precisely the assembly procedure of a typical nite element software system, except it is repeated several hundreds or thousands times. The modes and mode shapes of the total system can, in principle, be extracted from the eigenproblem associated with the above assembled equations of motion. 20.8 PARTITIONED EQUATIONS OF MOTION EMPLOYING LOCALIZED -METHOD The total energy of the system for this case which is the one given by the second of (20.21), consists of the two substructural energy expressions (20.18) plus the interface constraint functional, viz., the second expression in (20.20), as
total

= (u(1) )T {K(1) u(1) (f(1) M(1) u(1) )} + (u(2) )T {K(2) u(2) (f(2) M(2) u(2) )} + (1) (2) u(1) u(2)
T

+ {

L(1) u(1) f uf } u(2) L(2) f L(1) (1) f T (2) u f } (2) Lf { (20.27)

= (u(1) )T {K(1) u(1) (f(1) M(1) u(1) ) + (B(1) )T (1) } + (u(2) )T {K(2) u(2) (f(2) M(2) u(2) ) + (B(2) )T (2) } + (1) (2) uT f
T

{
T

u(1) = B(1) u(1) ,

L(1) f (2) Lf

L(1) B(1) u(1) f uf } B(2) u(2) L(2) f (1) (2) u(2) = B(2) u(2) 2011

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

2012 INTRODUCTION

The stationarity of the above variational equation, viz., of motion: M(1) 0 0

0 0 0 M(2) 0 0 0 M 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 u uf + K B 0 BT 0 LT f 0 L f 0 u uf = K 0 u(1) (2) 0 0 u (1) (1) 0 + B (2) 0 0 uf 0 0

= 0, yields the following partitioned equations u(1) f(1) 0 (2) 0 u(2) f L(1) (1) = 0 f 0 L(2) (2) f 0 uf 0 (20.28)

(1)

0 K(2) 0 B(2) 0

(B(1) )T 0 0 0 (L(1) )T f f 0 0

(B ) 0 0 (L(2) )T f

(2) T

20.9 EIGENVALUE PROBLEM USING REDUCED-ORDER PARTITIONED EQUATIONS OF MOTION Model reduction of a substructure has been presented in the previous chapter. In this section we will use the reduction procedure based on the constrained interface modes for assembling the substructures into a total system. To this end, lets express the reduction form for substructure k as u(k) =
(k)

q(k)

(20.29)

where u(k) is approximated from the previous chapter as: u

(k)

uI u

K1 K I II I

qI u

(20.6)

Substituting the above reduction formula into (20.23), one obtains (1) q K(1) 0 0 0 (B (1) )T M(1) (2) (2) q(2) + 0 M 0 0 K (B (2) )T (12) (1) (2) B B 0 0 0 0 M(1) = ( K(1) = ( B (1) = ( p(1) = (
(1) T (1) T (1) T (1) T

q(1) q(2) (12)

p(1) p(2) 0 (20.30)

) M(1)

(1) (1)

M(2) = ( K(2) = (
(2) T (2) T

(2) T

) M(2)
(1)

(1)

) K(1)

(2) T

) K(2)

) B(1) ,

B (2) = ( p(2) = (

) B(2)

) f(1) ,

) f(2)

Vibration analysis of the total system based on the above reduced-order model is carried using the following equation: total total = Mtotal total total K 0 M(1) 0 M(2) 0 0 (1) 0 K = 0 K(2) B (1) B (2) 2012 0 0 0 (B (1) )T (B (2) )T 0

total = K total M

(20.31)

20.10 EIGENVALUE PROBLEM USING REDUCED-ORDER ASSEMBLED EQUATIONS OF MOTION 2013 It is noted that total is not the eigenvectors of the assembled model. The correct eigenvectors (mode shapes) of the assembled model are obtained by u(1) u(2) 12
(1)

0 0 0 0 0 0

= =

0 0
(1)

(2)

0 0
(1)

(2)

0 0 I 0 0 I 0 0 I

q(1) q(2) 12 total qtotal (20.32)

total

0 0

(2)

total

which includes not only the modes that span the substructures but also the interface modes pertaining to the interface force (12) . However, the eigenvalues total represent the assembled structural system. In other words, (total , total ) constitute the mode shapes and modes of the assembled system even though we have obtained them from the partitioned equations of motion.

20.10 EIGENVALUE PROBLEM USING REDUCED-ORDER ASSEMBLED EQUATIONS OF MOTION In the preceding section the reduced-order partitioned equations of motion has been directly utilized for the formulation of eigenvalue problem. While computationally equivalent, the resulting eigenvalue problem given by (20.31) involves non-denite matrices, thus requiring a special care. One way to circumvent the non-denite matrices is to assemble the partitioned equations of motion into the assembled form akin to the equation given in (20.26). This can be accomplished in the following way. First, we note that the constraint condition u(1) u(2) = 0 implies that the interface Boolean matrices B(1) and B(2) yield the following relation: B(1) u(1) = [ 0 I(1) ] u(1) I u(1) (20.33) B(2) u(2) = [ 0 I(2) ] u(2) I (2) u (21.3)

This means that the assembled degrees of freedom, (q(1) , u , q(2) ), can be related to the partitioned degrees I I of freedom, (q(1) , u , q(2) , u(2) ),according to I I q(1)
I

I(1) I u 0 q(2) = 0 I (2) u 0

I(2) I 0

0 0

(1) q(2) I I qI 0 u I(2)

(1)

q par t = La qa

(20.34)

2013

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

2014 INTRODUCTION

where the superscripts, ( par t, a), denote the partitioned and assembled degrees of freedom, and La is an assembly Boolean matrix. Therefore, teh complete transformation relation can be expressed as q(1) I u q par t q(2) I = (12) (2) u (12)

= Ttotal

qa , (12)

Ttotal =

La 0

0 I(12)

(20.35)

Substituting the above assembly transformation into (20.30) and after some simplications, one arrives at the following reduced-order assembled equations of motion: Mq+Kq=p M = (La )T M(1) 0 K(1) 0 p(1) p(2) q(1) q(2) 0 M(2) 0 La K(2) La

K = (La )T

(20.36)

p = (La )T

q = (La )T

As one can see, the above reduced-order assembled equations of motion is difcult to follow through. We will examine a step-by-step derivation of the above equation below, which is konwn in the literature as the Craig-Bampton component mode synthesis or subtructuring method method.

20.11 THE CRAIG-BAMPTON METHOD Equation(20.36) may be considered a generic component mode synthesis as it can accommodate several possible substructural reduction methods. One of its specializations was proposed by R.R Craig and M.C.C. Bampton in 1968. As the Craig-Bampton component mode synthesis technique is perhaps the most widely used substructuring method, we present a step-by-step formulation of their method below. 20.11.1 Step 1: Approximate the substructural displacements It approximates the displacement of each substructure by a set of xed-interface normal modes plus a set of constraint modes. Specically, for substructure 1, u(1) is approximated by (see Eq. (20.9)) u(1) u(1) I u
(1)

(1) I

I(1) I
(1)

q(1) I u
(1)

(1) I(1) = (K(1) )1 K I II

(20.37)

2014

2015

20.11 THE CRAIG-BAMPTON METHOD

Similarly, the substructural displacement u(2) for substructure is approximated by u

(2)

u(2) I u(2)

(2) I

I(2) I(2)

q(2) I u(2)

(2) I(2) = (K(2) )1 K I II

(20.38)

20.11.2 Step 2: Obtain the approximate substructural kinetic and strain energy The approximate substructural strain energy and kinetic energy derived in (20.13) and (20.14) are restated below. U (1) = 1 (u(1) )T K(1) u(1) 2 q(1) I u(1) I q(2) u(2)
T (2) I T (1) I (1) I T

K(1) II K(1) I M(2) II M(2) I

K(1) I K(1) M(2) I M(2)

(1) I

(1) I

q(1) I u(1) (20.39) I q(1) u(1)

I(1)
(2) I T

0
(2) I

I(1)
(2) I

T (2) = 1 u(2) M(2) u(2) 2

I(2)

I(1)

The reduced-order mass for substructure 1, M(1) , is thus given by (1) T M(1) M(1) 0 II I I (1) M = (1) T M(1) M(1) I(1) I I M(1) II M(1) I
(1) T I (1) T I (1) T I

(1) I

(1) I

I(1) (20.40)

= where

M(1) I M(1)
(1) I

M(1) = II M(1) = I M(1) =

M(1) II (M(1) II

= I I I (due to massnormalization)
(1) I ) (1) I

(M(1) + M(1) I II
(1) I

+ M(1) ) + M(1) I I K(1) II K(1) I 2

1

+ M(1)

The reduced-order stiffness for substructure 1, K A , is thus given by K(1) = where K(1) = II K(1) = I
(1) T I (1) T I

K(1) I K(1) (1) .. . 2 I

(20.41)

K(1) II

(1) I

= (1) = II
(1) I )

(K(1) + K(1) I II

(1) T I

(K(1) K(1) (K(1) )1 K(1) ) = 0 I II II I

(1) I

K(1) = K(1) + = K(1)

(1) T (1) (K(1) I + I II K(1) (K(1) )1 K(1) I II I

K(1) ) + K(1) I I

2015

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

2016 INTRODUCTION

Observe that K(1) can be written as K

(1)

(1) II 0

0 K(1) . (20.42)

which consists of the diagonal interior substructural modes and the Guyan-reduced matrix K For substructure 2, a similar procedure employed for substructure A can be repeated to yield: M(2) = M(2) II M(2) I (2) II 0 M(2) I M(2)

(20.43) 0 K(2)

K(2) =

where it is understood that the substructural displacement u(2) is approximated by the xed-interface interior modes plus the constrained modes given by (20.9). 20.11.3 Step 3: Sum up the substructural kinetic and strain energy expressions The strain energy and the kinetic energy of the total structure can be approximated by T = T (1) + T (2) U = U (1) + U (2) U (1)
(1) 1 2

q(1) I u(1) I q(1) u(1) q(2) I u(2) I q(2) u(2)

(1) II

q(1) I

1 2

u(1) 0 K(1) I M(1) M(1) q(1) II I M(1) I M(1) 0 u(1) q(2) I (20.44)

U (2) T (2)

1 2

(2) II

1 2

0 K(2) u(2) I M(2) M(2) q(1) II I M(2) I M(2) u(2)

20.11.4 Step 4: Derive the reduced equations of motion for the total system The Lagrangian of the total system is given by L = T U + ((12) )T (u(1) u(2) ) (20.45)

where the last term involving the Lagrange multiplier (12) is introduced to enforce the interface displacement compatibility constraint u(1) u(2) = 0 (20.46) 2016

2017

20.11 THE CRAIG-BAMPTON METHOD

The equations of motion for free vibration (f(1) = 0, f(2) = 0) can be derived from (20.45) as M(1) 0 q
(1)

u(1) CT = [ 0 I 0

q(1) I

0 M(2) ,

q(1) K(1) + B q 0 (2) qI (2) q = u(2) I ]

0 K(2)

q(1) q(2)

= CT (12) (20.47)

If desired, the interface force (12) can be eliminated by expressing the interface displacement u(1) in terms of u(2) or vice versa. This can be accomplished by the reduction q(1) (1) I qI u(1) (2) = La q(2) , I q I u (2) u I 0 La = 0 0 0 0 I 0 0 I , 0 I

u = u(1) = u(2)

(20.48)

Substituting (20.48) into (20.47) and premultiplying the resulting equation by (La )T we obtain the following reduced-order free vibration equation: q(1) I q = q(2) I u M(1) I M = M(1) + M(2)

Mq+Kq=0 , M=

M(1) II 0 M
(1) I

0 M(2) II M(2) I 0 (2) II 0

M(2) , I M 0

(20.49)

K= 0

(1) II

0 , K

= K(1) + K(2)

20.11.5 Step 5: Perform eigenanalysis of the total system First, we perform an eigenanalysis of (20.49): K = M g , = (2) I Second, once (, g ) are obtained, the global eigenvector g is obtained by the following expression 2017 (1) I (20.50)

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:
(1) I (2) I

2018 INTRODUCTION

g = 0 0

(1) I

0
(2) I

(2) I

(1) I

(20.51)

Observe that the eigenvalues are preserved under a similarity transformation. Thus, the global eigenvalues and eigenvector pairs are given by (g , g ). The component mode synthesis for other techniques due to Beneld and Hruda, Hurty, MacNeal, Rubin, Hintz, Dowell and Klein, and Craig and Chang may be similarly constructed. Remark 1: Note that from (20.49) equation, in carrying out the component mode synthesis by the CraigBampton method, viz., 2 M = K (20.52) the mass matrix M becomes dense even if the original substructural-level mass matrices are diagonal. Remark 2: The stiffness matrix at the interface is given by K = K(1) + K(2) (20.53)

K(1) = K(1) K(1) (K(1) )1 K(1) I II I K(2) = K(2) K(2) (K(2) )1 K(2) I II I

Note that both K(1) and K(2) are the Schur complements (or in structural mechanics known as Guyanreduced matrices) that preserve the strain energy content of each substructure. Hence, no approximation is introduced at the interface strain energy contents. On the other hand, the same cannot be said regarding the kinetic energy. This can be seen by examing the interface mass traix M : M = M(1) + M(2)
(1) T I (2) T I

M(1) = M(2) =

(M(1) II (M(2) II

(1) I (2) I

+ M(1) ) + M(1) I I + M(2) ) + M(2) I I

(1) I (2) I

+ M(1) + M(2)

(20.54)

In other words, the constraint modes I play the role of augmenting the interface kinetic energy by infusing the interior masses M I I unto the interface nodes.

2018

2019
References 1. 2.

20.11 THE CRAIG-BAMPTON METHOD

Craig, Jr., R.R., Structural Dynamics: An Introduction to Computer Methods, John Wiley & Sons (1981). Hurty, W.C., Dynamic Analysis of Structural Systems Using Component Modes, AIAA Journal, v.3, 678-685 (1965). Craig, Jr., R.R. and M.C.C. Bampton, Coupling of Substructures for Dynamic Analysis, AIAA Journal, v. 6, 1313-1319 (1968). MacNeal, R.H., A Hybrid Method of Component Mode Synthesis, Comp. and Struct., v. 1, 581-601 (1971). Beneld, W.A. and R.F. Hruda, Vibration Analysis of Structures by Component Mode Substitution, AIAA Journal, v. 9, 1255-1261 (1971). Rubin, S., Improved Component-Mode Representation for Structural Dynamic Analysis, AIAA Journal, v. 13, 995-1006 (1975). Klein, L.R. and E.H. Dowell, Analysis of Modal Damping by Component Modes Method Using Lagrange Multipliers, J. Appl. Mech. Trans. ASME, v. 41, 527-528 (1974). Hintz, R.M., Analytical Methods in Component Modal Synthesis,AIAA Journal, v. 13, 1007-1016 (1975). Craig, Jr., R.R. and C-J. Chang, A review of Substructure Coupling Methods for Dynamic Analysis, NASA CP-2001, National Aeronautics and Space Admin., Washington, DC, v. 2, 393-408 (1976).

3.

4. 5.

6.

7.

8. 9.

10. Craig, Jr., R.R., Methods of Component Mode Synthesis, Shock and Vib. Digest, Naval Research Lab., Washington, DC, v. 9, 3-10 (1977). 11. Craig, Jr., R.R. and C-J. Chang, On the Use of Attachment Modes in Substructure Coupling for Dynamic Analysis, Paper 77-405, AIAA/ASME 18th Struct., Struct. Dyn, and Materials Conf., San Diego, CA (1977). 12. M. Baruch, Optimization Procedure to Correct Stiffness and Flexibility Matrices Using Vibration Tests, AIAA J., 16 (11), 1209-1210 (1978) 13. B. Caesar, Update and Identication of Dynamic Mathematical Models, Proc. 1st Intl. Modal Anal. Conf., 394-401 (1983) 14. J.C. Chen, C.P. Kuo, and J.A. Garba, Direct Structural Parameter Identication by Modal Test Results, AIAA/ASME/ASCE/AMS Proc. 24th Struc. Dynam. and Materials Conf., 44-49 (1983) 15. J.D. Collins, G.C. Hart, T.K. Hasselman, and B. Kennedy, Statistical Identication of Structures, AIAA J., 12 (2), 185-190 (1974) 16. J.C. Chen and B.K. Wada, Criteria for Analysis-Test Correlation of Structural Dynamics Systems, J. Appl. Mech., 471-477 (1975) 17. J.E. Mottershead, Theory for the Estimation of Structural Vibration Parameters from Incomplete Data, AIAA J., 28 (3), 559-561 (1990) 18. N.G. Creamer and J.C. Junkins, Identication Method for Lightly Damped Structures, AIAA J. Guidance, Control, and Dynamics, 11 (6), 571-576 (1988) 19. A.M. Kabe, Stiffness Matrix Adjustment Using Mode Data, AIAA J., 23 (9), 1431-1436 (1985)

2019

Chapter 20: REDUCED-ORDER MODELING OF

VIBRATING STRUCTURES:

2020 INTRODUCTION

20. H. Berger, R. Ohayon, L. Barthe, and J.P. Chaquin, Parametric Updating of FE Model Using Experimental Simulation: A Dynamic Reaction Approach, Proc. 8th Intl. Modal Anal. Conf., 180-186 (1990) 21. S.R. Ibrahim and A.A. Saafan, Correlation of Analysis and Test in Modeling of Structures, Assessment and Review, Proc. 5th Intl. Modal Anal. Conf., 1651-1660 (1987) 22. W. Heylen and P. Sas, Review of Model Optimization Techniques, Proc. 5th Intl. Modal Anal. Conf., 1177-1182 (1987) 23. R.J. Guyan, Reduction of Stiffness and Mass Matrices, AIAA J., 3 (2), 380 (1965) 24. M. Paz, Dynamic Condensation Method, AIAA J., 22 (5), 724-727 (1984) 25. H.P. Gysin, Comparison of Expansion Methods of FE Modeling Error Localization, Proc. 8th Intl. Modal Anal. Conf., 195-204 (1990) 26. Kennedy, C.C. and Pancu, C.D.P., Use of Vectors in Vibration Measurements and Analysis, J. Aero. Sci., Vol. 14(11), Nov. 1947, 603-625. 27. Lewis, R.C. and Wrisley, D.L., A System for the Excitation of Pure Natural Modes of Complex Structures, J. Aero.Sci., Vol 17(11), Nov. 1950, 705-722, 735. 28. Ewins, D.J., Modal Testing: Theory and Practice, John Wiley and Sons, Inc., New York, 1984.

2020