18

Vibration Computations and Nomographs

18.1 18.2 18.3 18.4 18.5 18.6 Models for Numerical Simulation Numerical Integration Vibration Response by Computer Simulation Commercial Software for Simulation Computing Natural Frequencies and Mode Shapes Nomograph Fundamentals

Daniel J. Inman
Virginia Polytechnic Institute and State University

A primary concern in performing vibration analysis is just how to represent the response once the model and the various inputs of interest are known. For linear systems with a single degree of freedom, the analytical solution is closed form and can be simply plotted to illustrate the response and its important features. Historically, computation has been difcult and response vibration data have been presented in nomographs consisting of log plots of the maximum amplitudes of displacement, velocity, and acceleration versus frequency on a single two-dimensional four-axis plot. Although this approach is useful and incorporated into military and manufacturer specications, wide availability of high-speed computing and computer codes to simulate detailed responses has produced a trend to display exact responses in both the time and frequency domains. Advances in computer algorithms and commercialization of codes has also made the computing of natural frequencies and mode shapes much easier. The following section introduces an example of a commercial computer simulation package and its use in representing the response of vibrating systems. Methods of computing natural frequencies and mode shapes are also given. In addition, the basic use of nomographs is presented.

18.1 Models for Numerical Simulation

The most common model of a vibrating system is the multiple degree of freedom (MDOF) model, which can be expressed as a vector differential equation with matrix coefcients of the form (t ) + Cx (t ) + Kx(t ) = f(t ) Mx (0) = x 0 x(0) = x 0 , x (18.1)

(t ) is an n 1 vector of velocities, where x(t) is an n 1 vector of displacement coordinates, its derivative x (t ) is an n 1 vector of accelerations. The coefcients M, C, and K are n and its second derivative x n matrices of mass, damping, and stiffness elements, respectively (see Chapter 14 for examples). These coefcient matrices are often symmetric and at least positive semidenite for most common devices and

2005 by CRC Press LLC

structures. Equation (18.1) follows from simple modeling using Newtons laws, energy methods, or 0 represent the required initial conditions. The dynamic nite elements. The constant vectors x0 and x simulation problem consists of calculating x(t), satisfying Equation (18.1) as time evolves, and producing a time record of each element of x(t), or of each degree of freedom, as opposed to a single coordinate as used in nomographs. Currently many very well-written numerical integration codes are available commercially for less cost than is involved in writing the code and, more importantly, with less error. Codes used to solve Equation (18.1) are written based on the denition of a derivative and almost all require the equations of motion to appear in rst-order form, that is, with only one instead of two time derivatives. Equation (18.1) can be easily placed into the form of a rst-order vector differential equation by some simple matrix manipulations. If the inverse of the mass matrix M exists, then the second-order vibration model of Equation (18.1) can be written as an equivalent rst-order equation using new coordinates dened by the 2n 1 vector: x(t ) z(t ) = (t ) x (t ); then Equation (18.1) can be written as Let z1 = x(t) and z2 = x 1 = z2 z 2 = - M -1Kz1 - M -1Cz 2 + M -1f(t ) z x z(0) = 0 0 x which combine to form = Az + F(t ), z z(0) = z 0 (18.3) (18.2)

Here z is called a state vector and the state matrix A is dened by 0 A= -1 - M K I - M -1C

(18.4)

where I is the n n identity matrix and 0 is an n n matrix of zeros. The forcing function F(t) is mass scaled to be the 2n 1 vector 0 F(t ) = -1 M f(t )

(18.5)

Here 0 denotes an n 1 vector of zeros. Note that in solving vibration problems using this state space coordinate system, the rst n components of the solution vector z(t) correspond to the individual displacements of the n degrees of freedom and the second list of n components of z(t) are the velocities.

18.2 Numerical Integration

The numerical solution or simulation of the system described by Equation (18.1) is easiest to discuss by rst examining the scalar homogeneous (unforced) case given by
2005 by CRC Press LLC

(t ) = ax(t ) x(0) = x 0 x (t ) is written from its denition as where a is a simple constant. The derivative x x(t i + Dt ) - x(t i ) = ax(t i ) Dt where D t is a nite interval of time. Rewriting this expression yields x(t i + Dt ) = x(t i ) + ax(t i )Dt or using a simpler notation x i +1 = x i + ax i Dt (18.8) (18.7) (18.6)

where xi denotes x(ti). This formula gives a value of the response xi + 1 at the next time interval, given the equations coefcient a, the time increment Dt, and the previous value of the response xi. Thus, starting with the initial value x0, the solution is computed at each time step incrementally until the entire record over the interval of interest is calculated. This simple numerical solution is called the Euler formula or tangent line method and only involves addition and multiplication. Of course, the smaller Dt is, the more accurate the approximation becomes (recall the derivative is dened as the limit Dt 0). Unfortunately, reducing the step size Dt increases the computational time. Numerical errors (rounding and truncation) also prevent the simulations from being perfect, and users should always check their results accordingly. The rule used to perform the simulation is often called an algorithm. One way to improve the accuracy of numerical simulation is to use more sophisticated algorithms. In the late 1800s, C. Runge and M.W. Kutta developed some clever formulas to improve the simple tangent or Euler methods. Essentially, these methods examine x(t + Dt) as a Taylor series expanded in powers of Dt. The Runge-Kutta methods insert extra values between xi and xi + 1 to provide estimates of the higher-order derivative in the Taylor expansion and thus improve the accuracy of the simulation. There are several Runge-Kutta methods, thus only an example is given here. = f (x,t ) with initial One of the most widely used Runge-Kutta methods solves the scalar equation x condition x (0) = x0 where f(x, t) can be linear or nonlinear as well as time varying. This includes the (t ) = ax(t ) + g(t ) where a is constant and g(t) is an externally applied force. With xi and Dt dened case x as before, the formulas for the response are x i +1 = x i + where Dt Dt hi 2 = f x i + hi1,t i + . 2 2 hi 4 = f ( x i + Dthi3 ,t i + Dt ) Dt (h + 2hi 2 + 2hi3 + hi 4 ) 6 i1 (18.9)

hi1 = f (x i ,t i ) Dt Dt hi3 = f x i + hi 2 ,t i + 2 2

This is referred to as a four-stage formula and represents a substantial improvement over the Euler method.
2005 by CRC Press LLC

Additional improvement can be gained by adjusting the time step Dt at each interval based on how rapidly the solution x(t) is changing. If the solution is not changing very rapidly, a large value of Dti, the ith increment of time is used. On the other hand if x(t) is changing rapidly, a small Dti is chosen. In fact, the Dti can be chosen automatically as part of the algorithm.

18.3 Vibration Response by Computer Simulation

All of these methods (as well as many others not mentioned) can be applied to the simulation of the response of vibrating systems. Essentially the Runge-Kutta and Euler formulas can be applied directly to Equation (18.3) by simply enforcing a vector notation. For instance, the Euler formula applied to Equation (18.3) becomes z(t i +1 ) = z(t i ) + DtAz(t i ) + F(t i ) (18.10)

using z(0) as the initial value. The result will be a list of numerical values for z(ti) versus the successive times ti. Equation (18.10) can be programmed on a programmable calculator or computer system. However, many commercially available codes provide more than adequate numerical integration schemes for solving ordinary differential equations and systems of ordinary differential equations as described by Equation (18.10). Such codes are easy to use and allow studies of the effects of initial conditions and parameter changes while providing detailed solutions to complex problems. Next, a simple example is introduced to illustrate the formulation of a vibration problem into state space form in preparation for numerical simulation. Consider then the equations of motion of a damped two-degree-of-freedom system 9 0 1(t ) 2.7 0 x + x 1 2 (t ) -0.3 1 27 -0.3 x + x 0.3 2 -3 -3 x 1 3 = sin3t 3 x 2 0

(18.11)

subject to the initial conditions 0.1 x(0) = , 0 0 0) = x( 0

Here all units are SI so that x(t) is in meters, etc. However, any consistent set of units can be used. This is a simple example with only two degrees of freedom so chosen to t the given space limitations. The procedure is, however, not dependent on such low order. The matrix M -1 in this case is simply 1 M -1 = 9 0 so that 3 M -1K = -3 and the state matrix becomes -0.333 0.3 -1 and M C = 3 -0.3 -0.033 0.3 0 1

2005 by CRC Press LLC

0 0 A= -3 3 where the state vector and forcing vector are x1 x z = 2 1 x x 2

0 0 0.333 -3

1 0 0.3 0.3

0 1 0.033 -0.3

0 0 and F(t ) = sin3t 0

respectively. Next, Equation (18.10) can be applied with these values. In the past it was required of the vibration engineer to program Equation (18.10) or some other versions of it. However, commercial software allows the matrix A to be easily assembled and allows easy simulation of the time response and plotting the frequency response.

18.4 Commercial Software for Simulation

A variety of simple-to-use, efcient, and relatively inexpensive interactive software packages are available for simulating the response. Such programs reduce by a factor of ten the amount of computer code that actually has to be written by the user. Some examples of available software containing numerical integration packages are MALAB, Mathematica, and Mathcad. Many nite element packages also contain numerical integration routines. Here we illustrate the use of MALAB to simulate the result of the simple example above and to print the results. The MALAB code is listed in Table 18.1 and the output is plotted in Figure 18.2. The algorithm used in Table 18.1 is a modication of the formulas given in Equation (18.9), known as a Runge-Kutta-Fehlberg integration method. This method uses a fourth- and fthorder pair of formulas and an automatic step size. The use of high-level computational software such as MALAB, Mathcad, and Mathematica is commonplace. The slide rule has given way to the calculator, and the calculator to the personal computer. Combined with commercial software, the simulation of large and complex vibration problems can be performed without resorting to writing code in lower-level languages. This time-saving allows the vibration engineer more time to devote to design and analysis. It is, however, important to note that simulation through numerical integration is still an approximation and as such is subject to error both formula errors and round-off errors. These should be well understood by the user. More detailed examples of these three codes used to solve vibration problems can be found in Inman [2001]. These same computer codes also contain sophisticated routines for obtaining the solutions to the eigenvalue problems associated with vibration and hence are very useful in obtaining the mode shapes and natural frequencies. These methods are discussed next.

18.5 Computing Natural Frequencies and Mode Shapes

Natural frequencies and mode shapes form the language of modern vibration analysis and are often used among practicing engineers and researchers to describe the vibration characteristics of a structure or machine. Computing natural frequencies and mode shapes use to involve many approximate techniques. The rise of modern computing and advances in numerical linear algebra and subsequent commercialization have made the computing of natural frequencies and mode shapes relatively easy. However,
2005 by CRC Press LLC

TABLE 18.1 MATLAB SIMULATION Code

MATLAB code for computing and plotting the displacement versus time response of the two-degree-of-freedom system in the text. The % symbol denotes comments. Part A indicates how to input the given system and part B illustrates how to integrate and plot the response. Part A function zdot = system (t, z) %This m-file defines the mechanical properties of the system being studied. %The input is the current time and the previous state vector (initial conditions). %The output is obtained by solving the state equation: zdot = A * z + f %First, the mass, damping and stiffness matrices are defined. M = [9, 0; 0, 1]; C = [2, 7, -0.3; -0.3, 0.3]; K = [27, -3; -3, 3]; f = [0; 0; sin(3 * t); 0]; %The vector of external forces is defined next. %The state matrix is assembled. A = [zeros(2, 2) eye(2, 2); -inv(M) * K -inv(M) * C]; Part B %This is the main file used in the simulation. %First the initial state is defined. z0 = [0.1; 0; 0; 0]; %Then, the solution is obtained through numerical integration using the ode command calling the system input file proposed in part A. ti = 0; %Initial time of the simulation tf = 50; %Final time of the simulation [time,solution] = ode45(system, ti, tf, z0); %Perform integration %The displacement of each degree-of-freedom is plotted. subplot (2,1,1); plot (time,solution(:,1)); xlabel(t[s]); ylabel(z1[m]); title(First DOF;); subplot (2,1,2); plot(time,solution(:,2)); xlabel(t[s]); ylabel(z2[m]); title(Second DOF);

commercial codes have centered on implementing numerical linear algebra results, i.e., implementing solutions of the algebraic eigenvalue problem, rather than focusing on vibration problems directly. Thus, computations of natural frequencies and mode shapes are best done by converting the denition of these physical quantities into the algebraic eigenvalue problem. There are several ways to make this transformation and these are presented here along with their strengths and weaknesses. Undamped systems are systems modeled by Equation (18.1) with no damping present (C = 0). Natural frequencies are dened as the frequencies at which a system will vibrate given an arbitrary initial displacement and velocity disturbance. Undamped natural frequencies (rad/sec) are denoted wn , and there will be one frequency for each degree of freedom in a linear system such as those successfully modeled by Equation (18.1) with C = 0. Natural frequencies are calculated from assuming a solution to Equation (18.1), for the undamped case of C = 0, of the form x(t) = uejwt, where u is an unknown vector of constants, j = -1 , t is the time, and w is the yet-to-be-determined natural frequency. Substitution of this assumed solution into Equation (18.1) yields:

2005 by CRC Press LLC

First DOF 0.4 0.2 Z1 [m] 0 0.2 0.4

10

15

20

25 t [s] Second DOF

30

35

40

45

50

0.5

0 Z2 [m] 0.5 1

10

15

20

25 t [s]

30

35

40

45

50

FIGURE 18.1 The output of the MALAB Code of Table 18.1, illustrating the simulation of the displacement versus time response of the system given by Equation (18.11).

(-w M + K )u = 0,
2

u0

(18.12)

This last expression can only hold true if the coefcient matrix is singular, which happens if and only if the determinant of (w2M + K) vanishes. This yields an nth order polynomial in w2, the roots of which are the n values of the squares of the natural frequencies w i2 indexed i = 1, 2, 3, n. The yet unknown vectors ui corresponding to each wi are called the mode shapes, and they are found from solving the n scalar equations given by Equation (18.12) for each of the n values of w i2 . Since the coefcient is singular (because its determinant is zero) at each value of w i2 , Equation (18.12) will result in only n 1 independent equations for each element of the mode shape vector ui. This allows for the solution of all but one value of the n elements of each mode shape vector. The nth value is computed by forcing the vector to have a unit magnitude (normalized). Rewriting Equation (18.12) reveals its relationship to the generalized eigenvalue problem: Ku i = w i2 Mu i Au i = l i Bu i ui 0 (18.13)

The generalized eigenvalue problem, stated on the right side of the above, is to compute ui and li given the matrices A = K and B = M. This is exactly equivalent to computing mode shapes and natural frequencies. This is a symmetric generalized eigenvalue problem, since the matrices M and K are both symmetric. The algorithms for computing the generalized eigenvalue problem are more costly then those of the normal eigenvalue problem, which can be obtained by multiplying Equation (18.13) by the inverse of the mass matrix M (or B) to get

2005 by CRC Press LLC

Velocity (mm/s) (peak) 2000

100 80

60

40

20

Displacement (mm) (peak)

1000 800 600

10 8 6

400

4 2

200

10000 2 4 Frequency (Hz) 1000 Acceleration (mm/s2) (peak) 6 8 10 8000 2000 4000 6000

FIGURE 18.2 An example of a vibration nomograph for specifying acceptable limits of sinusoidal vibration. (Source: Inman, D. J. 2001. Engineering Vibration, 2nd Edition, Prentice Hall, Upper Saddle River, NJ. With permission.)

M -1Ku i = w i2u i Au i = l i u i u i 0

(18.14)

Here the matrix A in the statement of the eigenvalue problem is now A = M1K, where M1 denotes the matrix inverse. The mode shapes are given directly by the vector u but since A is not symmetric there is no guarantee that the eigenvectors will be real valued or orthogonal (in fact they are not), nor is it obvious that the values of l i = w i2 will be nonnegative real numbers, as we expect for the vibration problem. If the eigenvalue problem of Equation (18.13) is transformed by mass normalizing the stiffness matrix, a symmetric eigenvalue problem results. There are two strong advantages to using the symmetric eigenvalue problem. The rst is that symmetric eigenvalue problems are known to produce orthogonal eigenvectors, even in the case of repeated frequencies, and the second is that the numerical procedure for computing the solution to the symmetric eigenvalue problem is not as subject to round-off error as the asymmetric eigenvalue problem is. To this end, assume the matrix M is positive denite and -1 symmetric (it usually is), so that it has a positive denite matrix square root. Let u = M 2 v in Equation 1 (18.13) and multiply the equation by M 2. This yields: M
-12

KM

-12

v i = w i2 v i Av i = l i v i v i 0

(18.15)

2005 by CRC Press LLC

 = M - 1 2 KM - 1 2, called the mass normalized stiffness matrix, is now symmetric, where the matrix A = K and the corresponding eigenvalue problem is symmetric. From the theory of eigenvalue problems (matrix theory or linear algebra) it is well known that the solution of a symmetric eigenvalue problem results in a set of n real valued, orthogonal eigenvectors, one for each value of li (regardless of repeated eigenvalues) and that the eigenvalues are all real valued and nonnegative if the matrix A is positive semidenite (which it is in this case). This is precisely the theory needed to ensure analytical solutions such as modal expansion methods (since the vi form an orthonormal basis) and in computing the time response given initial conditions. In addition, the computational algorithms for symmetric eigenvalue problems are considerably simplied compared with those needed for Equation (18.14). In particular, all the calculations assume real rather than complex arithmetic. Note that while the eigenvalues are the squares of the natural frequencies in all three eigenvalue problems listed above, the mode shapes dened from Equation (18.14) -1 are different from the eigenvectors of Equation (18.15) and are related by u = M 2 v . In particular, the mode shapes computed by Equation (18.14) are not orthogonal as often stated in many texts, but rather are only orthogonal when weighted by the mass matrix. This symmetric method can also proceed by using the Cholesky factors of M rather than the square root of the mass matrix. A symmetric positive denite matrix M has an upper triangular matrix (all elements below the diagonal are zero) factor L such that LTL = M. If M is diagonal, then the factor L is identical to the square root of M as used in Equation (18.15). For dynamically coupled systems (i.e., those with nondiagonal mass matrices), it is more efcient to compute the mass normalized stiffness matrix in Equation (18.15) using the Cholesky factors rather than the matrix square root. The mass normalized = (LT )-1 KL-1 . This version of the mass normalized stiffness matrix is stiffness matrix then becomes K used as the matrix A in Equation (18.15) to compute the eigenvalues and eigenvectors. (Note that some codes use a lower triangular Cholesky factor rather than upper, causing the transpose to ip sides.) Proportionally damped systems form the most commonly used damped case. Computing natural frequencies and mode shapes when viscous damping is present, C 0, becomes much more complicated and bifurcates into two distinct cases. The rst case, often referred to as proportional damping, occurs if and only if the coefcient matrices (assumed to be symmetric) form a product CM1K that is in turn symmetric. Then the eigenvectors calculated from the (undamped) symmetric eigenvalue problem of = M - 1 2CM - 1 2 ): Equation (18.15) become the eigenvectors of the damped expression as well and satisfy ( C +K v = 0, (l I + lC )
2

v0

(18.16)

= M - 1 2CM - 1 2 is called the mass normalized where I denotes the n n identity matrix. The matrix C damping matrix. Here, li are the system eigenvalues which are in general complex (for underdamped systems), and these are related to the natural frequency and damping ratio by: - Re(l i ) Re(l i )2 + Im(l i )2

w i = Re(l i )2 + Im(l i )2 , zi =

(18.17)

Not many numerical algorithms exist for solving Equation (18.16) for li and vi directly from Equation (18.16). However, because the assumption of symmetry of CM1K, the eigenvectors of the mass normalized stiffness matrix are also eigenvectors of the mass normalized damping matrix. Hence the eigenvectors of Equation (18.15) serve as the system eigenvectors for Equation (18.16), and the undamped natural frequencies are also computed from Equation (18.15). Hence, the natural frequencies and eigenvectors for Equation (18.16) are best calculated using Equation (18.15). To compute the damping ratios, the system of Equation (18.16) is transformed into a diagonal system using the eigenvectors to form the orthogonal matrix P constructed by taking the columns of P to be the orthonormal vectors vi:

2005 by CRC Press LLC

P = v1

v2

vn

(18.18)

Substitution of v = Pr into Equation (18.16) and multiplying PT yields (l2 I + diag(2zi w i )l + diag(w i ))r = 0 (18.19)

Here the vector r is called the modal coordinate system and the coefcient matrix of r is diagonal. The properties of the orthonormal matrix P for the case of proportional damping are
2 w 1 = 0 P T KP M 0

0 w2 2 M 0

PTP = I

L L O L

2 wn 0 0 M

2z1w 1 = 0 P TCP M 0

0 2z 2 w 2 M 0

L L O L

(18.20) 2z nw n

0 0 M

This last expression is used to compute the modal damping ratios zi given the matrices M, C, and K. In practice, the modal damping ratios are guessed based on material measurements or measured using a vibration test. The symmetry of CM1K is a necessary and sufcient condition for the above analysis and results from the linear algebra theory that two matrices share the same eigenvectors if and only if they commute. However, other, simpler conditions are used in practice. These are C = aM + bK
n

C=

a
i =1

i -1

K i -1

(18.21)

Here the constants a, ai , and b are arbitrary scalars chosen by the analyst so that C is positive semidenite and to produce some reasonable amount of damping. If these conditions are satised, then the eigenvectors, damping ratios, and natural frequencies can be determined by Equation (18.15) and Equation (18.20). Such systems are said to possess classical normal modes (because the eigenvectors are real valued and form an orthonormal basis). They are also called normal mode systems. General viscous damping refers to the situation when viscous damping is present but not of the proportional type. If the matrix CM1K is not symmetric, then the above analysis fails and the system is said not to be proportionally damped and in general (assuming all modes to be underdamped) will have complex mode shapes. The only choice to compute the modal information is then to use the state space matrix of Equation (18.4). The associated eigenvalue problem is not symmetric and is given by Az = lz (18.22)

Here the eigenvector z is related to the eigenvector in physical coordinates of Equation (18.12) by u z= lu

(18.23)

Here both the eigenvectors z and the eigenvalues l are in general complex numbers appearing in complex conjugate pairs. The relationships of these eigenvalues computed from Equation (18.22) to the natural frequencies and damping ratios are given by Equations (18.17). The mode shapes are a little more difcult to compute but are basically the rst n values of the vector z of Equation (18.23).
2005 by CRC Press LLC

TABLE 18.2 Comparison of Eigenvalue Calculation Time Using MATLAB

Method Equation (18.13) Equation (18.14) Equation (18.15) Cholesky State matrix Flops 417 191 208 97 2626

TABLE 18.3 MATLAB EIGENVALUE/EIGENVECTOR Codes

>%Enter the mass and stiffness matrices >M=[9 0;0 1];K=[27 -3;-3 3]; >%Compute the cholesky decomposition and the mass normalized stiffness >L=chol(M);S=inv(L);Kh=S'*K*S; >%Compute the eigenvalues store in D and the P matrix >[P,D]=eig(Kh) P = -0.7071 -0.7071 0.7071 -0.7071 D = 4 0 0 2

Next, the simple example of Equation (18.11) is used to illustrate the various eigenvalue problems and their solution in MATLAB. First note that this example has proportional damping since 2.7 CM -1K = -0.3 -0.3 19 3 0 0 27 1 -3 -3 9 = 3 -1.8 -1.8 1

is obviously symmetric. Thus the eigenvectors and mode shapes of the undamped system serve as the eigenvectors and mode shapes of the damped system. When one proceeds to compute the solution of the eigenvalue problem (for C = 0) using a code, the question naturally arises: is there a numerically superior method? Table 18.2 illustrates the results of computing the eigenvalues and eigenvectors for each of the eigenvalue problems listed above applied to this simple example. The ops command (no longer available in MATLAB) is used to indicate how many oating-point operations are used in each calculation. Note that the most expensive choice of computing the eigenvalues and eigenvectors is using the state space formulation of Equation (18.22) followed by the generalized eigenvector problem of Equation (18.13). The cheapest computationally uses the Cholesky form of the symmetric eigenvalue problem of Equation (18.15). It should be noted that while using Equation (18.14) is relatively low, it does not result in orthogonal eigenvectors, and additional steps are required to solve the associated vibration problem. A sample code is given in Table 18.3. Detailed examples in MATLAB, Mathematica, and Mathcad for vibration analysis can be found in Inman [2001]. For simple small-order problems, the differences between eigenvalue problems is numerically not too signicant. However, for practical, larger-order systems or for systems with order of magnitude differences in mass and stiffness terms, the choice of eigenvalue problem formulation could become critical.

18.6 Nomograph Fundamentals

Nomographs are graphs used to represent the relationship between displacement, velocity, acceleration, and frequency for vibrating systems. These graphs are frequently used to represent vibration limits for given parts, machines, buildings, and components. The basic premise behind a vibration nomograph is
2005 by CRC Press LLC

that the response of a system is harmonic of the form x(t ) = A sin w n(t ) . Here A is the amplitude of vibration and wn is the natural frequency of vibration in rad/sec. The velocity is the derivative (t ) = w n A cos w nt, and the acceleration is the second derivative x (t ) = -w 2 x n A sin w nt . Thus, if a vibrating system has displacement amplitude A, its velocity has amplitude wnA and the acceleration amplitude is w2 n A. For a given harmonic motion, these three amplitudes can be plotted versus frequency, commonly using a log scale as illustrated in Figure 18.1. In this log scale plot, the log of the frequency (denoted f in Hz, where f = wn/ 2p) is plotted along the horizontal and the corresponding velocity amplitude is plotted along the vertical, also on a log scale. The lines slanting to the right, of slope +1, correspond to the log of the displacement amplitude versus frequency, whereas those slanting to the left, with slope 1, correspond to the log of the acceleration amplitude versus frequency. Thus a given point on the monograph corresponds to the amplitude of displacement, velocity, and acceleration at a specic frequency. Nomographs can be used to specify regions of acceptable vibration. Often it is not enough to specify just displacement; restrictions may also exist on velocity and acceleration amplitudes. By sketching a closed shape on a nomograph, ranges of acceptable levels of maximum displacement, velocity, and acceleration over a frequency range of interest can be easily specied. The bold lines in Figure 18.2 illustrate an example. The bold lines in the gure are used to illustrate vibration between 2 and 8 Hz with displacement amplitude limited by 30 mm, velocity amplitude limited by 400 mm/sec and acceleration limited to amplitude by 104 mm/sec2. Rather than maximum amplitude, root mean square values of displacement, velocity, and acceleration can be plotted as nomographs. As mentioned, such plots are often used in formal documents, vendor specications, and military specications. The International Organization for Standardization presents vibration standards for severity, which are often represented in nomograph form. As useful as nomographs are and as frequently as they appear in codes and standards, modern computational abilities allow detailed representations of vibration data for much more complicated systems. In particular, it has become very common to simulate time responses directly.

Dening Terms
Eigenvalue problem The algebraic process of computing the eigenvalues and eigenvectors of a given matrix. Euler method A simple numerical solution to a rst-order ordinary differential equation based on approximating the derivative by a slope. Formula error Error in the computed response due to the difference between the exact solution and the approximate formula. Mode shape The eigenvector of the mass normalized stiffness matrix multiplied by the inverse square root of the mass matrix, indicating the shape in which a system naturally tends to vibrate. Natural frequencies The square roots of the eigenvalues of the mass normalized stiffness matrix and the frequencies at which a freely responding system will oscillate. Nomogaph A graph of displacement, velocity, and acceleration versus frequency for a single-degreeof-freedom system. Round-off error Error in the computed response due to numerical round-off and truncation in computer arithmetic. Runge-Kutta method A numerical solution to a rst-order ordinary differential equation based on approximating the derivative by several estimates of the rst few terms of a Taylor series expansion of the solution. Simulation Numerical integration to solve an (ordinary) differential equation using time steps to produce the time history of the response (of a vibrating system).

2005 by CRC Press LLC

References
Boyce, W. E. and DePrima, P. C. 1986. Elementary Differential Equations and Boundary Value Problems, 4th ed. John Wiley & Sons, New York. Forsythe, G. E., Malcolm, M. A., and Moler, C. B. 1977. Computer Methods for Mathematical Computation, Prentice Hall, Englewood Cliffs, NJ. Golub, G. E. and Van Loan, C. R., 1996. Matrix Computations, 3rd ed., Johns Hopkins University Press, Baltimore, MD. Inman, D. J. 2001. Engineering Vibration, 2nd ed., Prentice Hall, Upper Saddle River, NJ. Macinante, J. A. 1984. Seismic Mountings for Vibration Isolation, John Wiley & Sons, New York. Moler, C. B. 1980. MATLAB Users Guide Technical Report CS81-1, Department of Computer Sciences, University of New Mexico, Albuquerque.

Further Information
Further information can be found by consulting the references. More information on MALAB can be obtained from The MathWorks, Inc. 3 Apple Hill Drive Natick, MA 01760-2098 www.mathworks.com Information on Mathcad can be obtained from Headquarters Mathsoft Engineering & Education, Inc. 101 Main Street Cambridge, MA 02142-1521 www.mathsoft.com Information on Mathematica can be obtained from Corporate Headquarters Wolfram Research, Inc. 100 Trade Center Drive Champaign, IL 61820-7237 www.wolfram.com

2005 by CRC Press LLC