Shock and Vibration Response Spectra Study Course - Tom Irvine

SHOCK AND VIBRATION RESPONSE SPECTRA COURSE
Unit 1A. Natural Frequency Calculation

By Tom Irvine
Email: tomirvine@aol.com
Introduction
The simplest model in vibration analysis is the single-degree-of-freedom system. Examples are
the pendulum in Figure 1 and the spring-mass system in Figure 2.
g
L
θ m = mass
L = length
g = gravity
θ = angular displacement
m
Figure 1.
m m = mass
k = spring stiffness
c = damping coefficient
c X = displacement
k
Figure 2.
Each of these systems has an equilibrium or rest position. Now assume that the mass in either
figure is given an initial displacement and then released so that it may vibrate freely. The mass
1
will oscillate at a single frequency which is called the natural frequency. This is know as
"simple harmonic motion."
Eventually, the oscillation will decay. The mass will return to its rest position. The decay is
referred to as "damping."
Some Simple Physics

Kinetic energy is the energy of motion. It is proportional to the velocity squared. The mass
achieves its greatest kinetic energy as it passes through its rest position.
Potential energy is stored energy. The pendulum's potential energy is proportional to the hieght
that it is raised above its rest position with respect to the gravity field. A spring's potential
energy is proporational to the distance squared that it is stretched or compressed from its rest
position.
The pendulum achieves its greatest potential energy when it reaches its maximum angular
displacement. The spring reaches achieves its greatest potential energy when the mass reaches
its maximum displacement in either the positive or negative direction.
The energy relationships are summarized in Table 1.
Table 1. Energy Relationships for an Oscillating Mass

Displacement Kinetic Energy Potential Energy
Zero Maximum Zero
Maximum Zero Maximum
The conservation of energy requries that maximum kinetic energy must equal the maximum
potential energy, assuming no energy dissipation.
Lord Rayleigh derived mathematical equations which showed that the conservation of energy
can only be fulfilled at a system's natural frequency.
Natural Frequency Formulas
The natural frequency ω n for a pendulum is
g
ωn = (1)
L
The natural frequency for a spring-mass system is
k
ωn = (2)
m
2
The ω n natural frequency has dimensions of radians/time. The typical unit is radians/second.
Note that
ωn = 2 π f n (3)
ωn
fn = (4)
2π
The variable f n is the natural frequency in cycles/time. The typical unit is cycles/second, which
is called Hertz. The unit Hertz is abbreviated as Hz.
Note that the period T is
1
T= (5)
fn
The period is the time required for one complete cycle of oscillation.
Benefits of Natural Frequency

All physical objects have one or more natural frequencies.
As an example, a piano wire must be tuned to a particular frequency to match a particular
musical note. The "A above middle C" note is defined as having a natural frequency of 440 Hz.
The wire thus vibrates 440 times per second. This vibration produces acoustic pressure waves at
the same frequency. Music is one example where natural frequencies are thus necessary and
desirable.
Natural frequencies are beneficial in other areas. Pendulum motion has been used to measure
time. In addition, a child enjoys the pendulum motion of swing.
Astronauts aboard a space shuttle orbiting the Earth experience "apparent weightlessness." They
simply cannot use a bathroom-type scale device to measure their weight. Instead, the astronauts
sit on chair that is mounted to a frame via springs, as shown in Figure 3. The chair oscillates at a
natural frequency that is dependent on the astronaut's mass.
3
Figure 3. Spring-loaded Chair Device for Measuring Astronaut's Mass
The astronaut can then calculate his or her mass from the natural frequency using equation (2).
The equivalent Earth weight can then be calculated from the mass.
Drawbacks of Natural Frequency

The excitation of a natural frequency may cause failure or even fatalities in certain situations.
There are four methods by which a structure's natural frequency may be excited. They are
summarized in Table 2.
Table 2. Methods of Excitation

Method Example
Applied Pressure or Force 1. Hammer strikes mass
2. Airflow or wind excites structure such as an
aircraft wing
3. Rotating mass imbalance in motor
4. Pressure oscillation in rocket motor
Base Excitation 1. Vehicle traveling down washboard road
2. Earthquake
Self-excited Instability Airfoil or Bridge Flutter
Initial Displacement or Velocity Plucking guitar string
Resonance occurs when the applied force or base excitation frequency coincides with the
system's natural frequency. During resonant vibration, the response displacement may increase
until the structure experiences buckling, yielding, fatigue, or some other failure mechanism.
4
The Tacoma Narrows Bridge failure is often cited as an example of resonant vibration. In
reality, it was a case of self-excited vibration, as explained in Reference 1.
The failure of the Cypress Viaduct in the 1989 Loma Prieta Earthquake is example of failure due
to resonant excitation. A collapsed deck is shown in Figure 4.
Resonant vibration caused 50 of the 124 spans of the Viaduct to collapse. The reinforced
concrete frames of those spans were mounted on weak soil. As a result, the natural frequency of
those spans coincided with the frequency of the earthquake ground motion. The Viaduct
structure thus amplified the ground motion. The spans suffered increasing vertical motion.
Cracks formed in the support frames. Finally, the upper roadway collapsed, slamming down on
the lower road.
The remaining spans that were mounted on firm soil withstood the earthquake.
Figure 4. Cypress Viaduct Failure
There are many additional examples whereby knowledge of structure's natural frequency is
critical.
5
Consider a rocket vehicle with a closed-loop guidance system. The autopilot has an internal
navigation system that uses accelerometers and gyroscopes to determine the vehicle's attitude
and direction. The navigation system then sends commands to actuators that rotate the exhaust
nozzle to steer the vehicle during its powered flight. Feedback sensors measure the position of
the nozzle. The data is sent back to the navigation computer. Unfortunately, the feedback
sensors, accelerometers, and gyroscopes could be affected by the vehicle's vibration.
Specifically, instability could result if the vibration frequency coincides with the control
frequency.
Multi-degree-of-freedom Systems
In reality, structures have numerous natural frequencies. The lowest natural frequency is called
the "fundamental frequency."
The higher natural frequencies may, or may not, occur at integer multiples of the fundamental
frequency.
The higher natural frequencies are called harmonics if they occur at integer multiples of the
fundamental frequency. As an example, the "A above middle C" note has a fundamental
frequency of 440 Hz, with harmonics at 880 Hz, 1320 Hz, 1760 Hz, etc. The blend of harmonics
gives the music a rich sound.
The natural frequencies of a beam or plate, however, do not occur at harmonic frequencies.
In many cases, the fundamental frequency is the dominant frequency. Thus, many systems can
be modeled as single-degree-of-freedom systems.
For example, a cantilever beam has numerous natural frequencies. Striking the free end tends to
yield a response dominated by the fundamental frequency.
Tall Building
A tall building is clearly a multi-degree-of-freedom system. It can be modeled as a single-
degree-of-freedom system, however, for preliminary calculations.
The 1994 Uniform Building Code (UBC) gives the following empirical formulas for the
fundamental frequency fn and period T of a tall building.
1
fn = (6)
Ct h 3 / 4
T = Ct h 3 / 4 (7)
6
where
h is the height.
Ct is an empirical factor given in Table 3.
Table 3. Empirical Factor Ct

Type English Metric
 sec   sec 
Ct   Ct  
 ft 3 / 4   meters 3 / 4 
Masonry and wood buildings 0.020 0.049
Reinforced concrete moment resisting frames 0.030 0.073
and eccentrically braced frame
Steel moment resisting frames 0.035 0.0853
Note that the frequency is a bending frequency. The motion is in the lateral direction, parallel to
the ground. The mode shape is similar to that of a cantilever beam.
7
Figure 5.
Earth's Natural Frequency
The Earth experiences seismic vibration.
The fundamental natural frequency of the Earth is 309.286 micro Hertz. This is equivalent to a
period of 3233.25 seconds, or approximately 54 minutes.
Reference: T. Lay and T. Wallace, Modern Global Seismology, Academic Press, New York,
1995.
8
Figure 6.
Golden Gate Bridge
The Golden Gate Bridge is a steel suspension bridge. The total length is 8980 ft.
In addition to traffic loading, the Golden Gate Bridge must withstand the following
environments:
1. Earthquakes, primarily originating on the San Andreas and Hayward faults

2. Winds of up to 70 miles per hour
3. Strong ocean currents
The Golden Gate Bridge has performed well in all earthquakes to date, including the 1989 Loma
Prieta Earthquake. Several phases of seismic retrofitting have been performed since the initial
construction.
Note that current Caltrans standards require bridges to withstand an equivalent static earthquake
force (EQ) of 2.0 G.
The natural frequencies of the Golden Gate Bridge are shown in Table 4.
9
Table 4. Golden Gate Bridge Natural Frequencies
Mode Type Period of Natural

vibration Frequency
(sec) (Hz)
Transverse 18.2 0.055

Vertical 10.9 0.092
Longitudinal 3.81 0.262
Torsional 4.43 0.226
Circuit Boards
Circuit Boards are often modeled as Single-degree-of-freedom systems. The fundamental

bending frequencies of sample circuit boards are shown in Table 5. The avionics components
are from rocket vehicles.
Table 5. Sample Circuit Boards
Avionics Component Fundamental

Frequency (Hz)
CEP 65
PSSL 210
MUX 220
PDU 225
PCM Encoder 395
TVC 580
Average = 328 Hz
Std Dev = 203 Hz
Range = 65 Hz to 600 Hz
10
Helicopter Ground Resonance
f(t)
m
m
k c
Figure 7.
A new helicopter design undergoing testing may encounter severe vibration while it is on the
ground, preparing for takeoff. As the rotor accelerates to its full operating speed, a structural
natural frequency of the helicopter may be excited. This condition is called resonant excitation.
11
Frequency Measurement
A sample acceleration time history is shown in Figure 8.
10 CYLCES / 1 SECOND = 10 Hz
1 2 3 4 5 6 7 8 9 10
1
ACCEL (G)
-1
-2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
TIME (SEC)
Figure 8.
The time history in Figure 8 was generated analytically. Assume that this data represents the
free vibration response of a system with no damping. Also assume single-degree-of-freedom
behavior, although the data justifies this assumption.
12
The natural frequency can be calculated via a variety of tools. An example is the Fourier
transform which will be covered in a later unit.
In certain cases, a manual counting of the peaks is more accurate any other method. The
example in Figure 8 shows a signal that undergoes 10 cycles in 1 second. The natural frequency
is thus 10 Hz. The period of an individual cycle is the reciprocal of the natural frequency, 0.1
seconds in this example.
Again, the example in Figure 8 is free vibration. The system could have been subjected to an
initial displacement and then left to vibrate freely.
Thus, there is no force or base excitation acting upon the system in this example. Otherwise, the
simple counting method might be unreliable. Note that a system which is excited at a frequency
of 50 Hz will vibrate at 50 Hz regardless of its natural frequency.
More sophisticated methods are required to determine the natural frequency when an external
excitation is acting upon a system.
Homework Assignment
Hints for 1 and 2: Hand calculations often produce errors due to unit conversion errors.
Use the sdof_fn program available at: http://www.vibrationdata.com/software.htm
Please remember to download the DLL files that are given at the top of the web page.
1. A particular circuit board can be modeled as a single-degree-of-freedom system. Its weight

is 0.1 pounds. Its stiffness is 400 pounds per inch. Calculate the natural frequency.
2. A 5 pound avionics box is to be mounted on a rocket vehicle bulkhead via rubber bushings.
The bulkhead will experience a 240 Hz oscillation in flight due to the motor pressure
oscillation. The design goal is to avoid resonant excitation of the avionics box natural
frequency. As a precaution, the goal is to achieve a natural frequency of 120 Hz for the
avionics box. The avionics box natural frequency would thus be one-half the forcing
frequency. What should the bushing stiffness be (pounds per inch) to achieve this natural
frequency?
3. A rocket vehicle is carried underneath an aircraft. It experiences an initial displacement

because gravity causes it to bow downward while it is attached to the aircraft. It is suddenly
released and allowed to vibrate freely as it falls. It continues falling for about 5 seconds prior
to its motor ignition, as a safety precaution.
An acceleration time history of the drop is given in file: drop.txt.
The acceleration time history shows the body-bending natural frequency. Note that the
signal is filtered such that the rigid-body acceleration is not included in the data.
13
The file has two columns: time (sec) and accel (G).
The data is in ASCII text format. Call this file into Excel or some other graphics program.
Make a hardcopy.
Estimate the natural frequency by calculating the number of peaks and then dividing by time.
The decay effect is not important for this exercise. Try to make a reasonable estimate. In
other words, do not agonize about achieving accuracy finer than 0.1 Hz.
Important note: the drop.txt file is actual flight data measured via telemetry. The amplitude
has been scaled by an undisclosed factor, however.
4. A building is 200 feet high. It is constructed of steel moment resisting frames. Calculate the
period and natural frequency. Use equations (6) and (7).
5. A flagpole is made from steel pipe.
The height is 180 inches.

The pipe O.D. is 3 inches.
The wall thickness is 0.25 inches.
The boundary conditions are fixed-free.
Determine the fundamental lateral frequency. Use program: beam.exe
6. Read References 1 and 2.

7. Optional Assignment. Procure a tuning fork from a music store. Perform the demonstrations
described in Reference 2.
References
1. T. Irvine, Tacoma Narrows Bridge Failure, Vibrationdata Publications, 1999.

2. T. Irvine, Tuning Fork Sound and Vibration, Vibrationdata Publications, 2000.
14
Unit 1B. Damping
By Tom Irvine
Introduction
Recall the homework assignment from Unit 1A. The data.txt time history represented a
rocket vehicle dropped from an aircraft. The rocket vehicle oscillated as a free-free beam
during the interval between release and motor ignition. A beam is a continuous structure
with numerous vibration modes. The fundamental frequency, however, dominated the
response. Thus, the rocket vehicle could be modeled as a single-degree-of-freedom
system with a natural frequency of 9.6 Hz.
The oscillation of the rocket vehicle was the result of initial displacement. The vehicle
bowed downward due to gravity while attached to the aircraft. Thus, there was initial
potential energy in the form of strain energy.
During free vibration, the oscillation amplitude decreases with time due to damping
effects. The damping of the rocket of vehicle is due to several effects: internal material
friction, friction at joints, and air resistance.
In addition, structures may lose energy by sound radiation and by friction at boundaries.
Modal Analysis
A multi-degree-of-freedom system is composed of numerous vibration modes. Each
mode has an independent natural frequency and damping ratio. "Modal analysis and
testing" is carried out to resolve the natural frequencies and damping ratios.
Free Vibration Equation

Return to the single-degree-of-freedom example. The acceleration time history &y&( t ) for
free vibration can be approximated by the following formula
&y&( t ) = A exp( − ξ ω n t )sin ( ω d t + φ) (1)
where
A is the acceleration amplitude
ξ is the damping ratio or fraction of critical damping
ωn is the natural frequency
ωd is the damped natural frequency
φ is the phase angle
1
Note that the damping ratio is sometimes represented by ζ in other references.
The phase angle is determined by the initial conditions, as explained in Reference 1.
The damped natural frequency is
ω d = ω n 1− ξ 2 (2)
Lightly-Damped Systems
For many systems, the damping ratio ξ < 0.25. Thus,
ω d ≈ω n (3)
Equation (1) simplifies to
&y&( t ) ≈ A exp ( − ξ ω n t )sin ( ω n t + φ) (4)
Measuring Damping
Consider the analytical time history function in Figure 1. The natural frequency is 10 Hz,
as determined by inspection. The task is to determine the damping ratio.
2
ACCELERATION TIME HISTORY
1.0
0.5
ACCEL (G)
-0.5
-1.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
TIME (SEC)
Figure 1.
Thomson gives a "log decrement method" for determining the damping ratio in Reference
2. This method uses the ratio of peak amplitudes between two consecutive peaks to
determine the damping ratio. Furthermore, the method can also be used for two peaks
separated by n cycles.
Thomson's log decrement method has merit. Nevertheless, test data often has slight non-
linearities and other effects which pose an obstacle for this method.
A better method for test data is to perform a trial-and-error curve-fit using the exponential
term in equation (4). Let E(t) be the envelope function.
E( t ) = A exp ( − ξ ω n t ) (5)
The sample time history is shown again in Figure 2 along with an envelope curve. The
envelope curve was obtained by trial-and-error inspection.
3
1.0
y = 1 exp( -3.1 t )
Accel Data
0.5
ACCEL (G)
-0.5
-1.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
TIME (SEC)
Figure 2.
The envelope equation in Figure 2 is
E(t ) = 1 exp( − 3.1 t ) (6)
Compare equation (5) with (6).
ξ ω n = 3.1 (7)
3.1
ξ= (8)
ωn
3.1
ξ= (9)
2π f n
The natural frequency is 10 Hz.
4
3.1
ξ= (10)
2π (10 )
3.1
ξ= (11)
2π (10)
ξ = 0.049 (12)
Or
ξ = 4.9 % (13)
The exact value is 5%. The envelope method thus yields a very good estimate.
Note that damping is represented in a variety of forms. For example, the amplification
factor Q is
1
Q= (14)
2ξ
The sample problem thus has an amplification factor of Q = 10.
Homework
1. Fun assignment. The Rex2launch.qt file is a public domain file of a Pegasus launch
in QuickTime format. View the video. The video covers the period from drop
through early stage 1 burn. The body-bending motion is not visible in the video, but
nevertheless is measured by the accelerometers. Again, the drop.txt file is taken from
a Pegasus launch, but the amplitude has been modified by an undisclosed scale factor
for proprietary reasons. Furthermore, the drop.txt data is taken from a flight other
than the one shown in the video. The drop transient data is fully disclosed to the
payload engineers, however. In fact, the mass of the payload affects the frequency of
the drop transient. Thus, the launch vehicle provider and the payload engineers must
perform a "coupled-loads" analysis to determine how the launch vehicle will affect
the payload and vice-versa. This is carried out using a finite element model.
2. Recall the drop.txt file from Unit 1A. Use the envelope method to estimate the
damping ratio. Program dratio.exe can be used to generate sample envelope curves.
Hint: the time history takes one or two cycle to work itself into a pure bending
oscillation. Thus ignore the first few peaks. Take the third or fourth peak coordinate
as a reference for program dratio.exe. An alternate method is to generate the curves
within the graphics program itself. Note: perfect envelope fit is impossible due to
non-linearities. Try to come up with reasonable fit.
5
3. What is the amplification factor for drop.txt time history?
4. Review tutorials damping.pdf and dcross.pdf.
References
1. T. Irvine, Free Vibration of a Single-Degree-of-Freedom System, Vibrationdata
Publications, 1999.
2. W. Thomson, Theory of Vibration with Applications, Second Edition, Prentice-Hall,
New Jersey, 1981.
6
Unit 2A. Sine Vibration Characteristics
By Tom Irvine
Introduction
Consider a single-degree-of-freedom system undergoing excitation from an applied force or base

excitation. Either the input forcing function or the output response may take the form of a
sinusoidal oscillation. This form of oscillation is very important in vibration analysis. Even
complex, random vibration can be represented by a series of sine functions.
Statistical Parameters
There are certain descriptive statistical parameters used to characterize a signal. These
parameters can be used for any type of signal whatsoever, although the focus of this unit is
sinusoidal signals.
The first parameter is the average or mean value. Consider a continuous function a(t) with
duration T. The mean value X is
1 T
X = lim T → ∞ ∫
T 0
a (t ) dt (1)
A practical signal would not have an infinite duration, however. The formula can be modified
for finite signals.
Consider a sine function with amplitude A which has a duration of one cycle.
1 T 2πt 
X= ∫
T 0
A sin  dt
T 
(2)
T
1  T  2πt 
X= − A   cos   (3)
T 2π   T  0
1 T 
X=− A {cos(2π)− 0} (4)
T 2π 

X=0 (5)
The mean value of a single cycle of a sine wave is zero. Likewise, the mean value of n cycles is
zero, where n is an integer.
1
Now assume that the number of cycles is not an integer. An example is 78.683 cycles.
The mean value equation becomes
1 cT 2πt 
X=
cT 0 ∫A sin  dt
T 
(6)
where c is the number of cycles.
The mean value approaches zero as c becomes large. Thus, the mean value of pure sine
oscillation is considered as zero regardless of the number of cycles.
Usually, the mean value of a signal is of little use in vibration analysis. Here are two exceptions:
1. A segment of measured accelerometer data has a spurious baseline offset due to
instrumentation problems. This "DC offset" must be removed to accurately portray
the data.
2. The test item is undergoing "rigid-body" acceleration. The accelerometer is capable
of measuring this acceleration, down to a frequency of zero Hz. Furthermore,
measurement of rigid-body acceleration is a requirement.
Rigid-body acceleration must be measured in a guided rocket vehicle. This is done via the
servo accelerometers in the inertial navigation system. The flight computer then "double
integrates" the acceleration in order to determine the displacement.1
Most accelerometer measurements, however, are taken on items such as machines and other
structures which experience zero net displacement. Hence, the expected mean acceleration
value is zero. Some data acquisition systems thus have "AC coupling" or high-pass filtering
to remove any spurious DC component. Otherwise, the removal of the DC component must
be performed on the digital data in post-processing.
Further descriptive statistics are needed. The next candidate is the mean square value X 2 .
1 T
X 2 = lim T → ∞ ∫[a ( t)]2 dt (7)
T 0
1
Rotational data from gyroscopes is also used in the displacement calculation.
2
The mean square value for one cycle of a sine wave is
2
1 T 2πt 
∫
2
X = A sin   dt (8)
T 0   T 
Apply a trig identity.
A2 T  4πt 
X2 = ∫1 − cos  dt
2T 0   T 
(9)
T
2 A2   T  4πt 
X = T −  sin   (10)
2T  4π   T  0
A2
X2 = (11)
2
The root-mean-square (RMS) value XRMS is simply the square root of the mean value. Thus,
the RMS value for one cycle of a sine function is
1
X RMS = A (12)
2
X RMS ≈0.707 A (13)
Note that equations (11) through (13) are also true for a sine function with numerous cycles.
Warning: equations (11) through (13) apply to pure sine vibration. Random vibration has a
completely different relationship as discussed in a later unit.
Standard Deviation
Another important parameter is the standard deviation σ . It is a measure of the dispersion about
the mean.
The standard deviation is related to the mean and mean-square values as follows:
σ2 = X 2 − [
X ]2 (14)
Note that σ2 is called the variance.
3
X2 − [
X]
2
σ= (15)
Thus, the standard deviation is equal to the RMS value if the mean is zero.
Discrete Forms
Vibration data is often represented in digital format. Consider a series of amplitude points ai
where there is a total of n points.
The mean value is
n
∑
1
X = xi (16)
n
i =0
The mean-square value is
n
∑ 
x i 2 
1
X2 =  (17)
n  
i=0
The standard deviation equation is still
X2 − [
X]
2
σ= (18)
An equivalent formula for the standard deviation is
n
∑
1
σ = ( x i − X) 2 (19)
n
i =0
Crest Factor
The relationship of the peak amplitude to the standard deviation value is another parameter of
interest. This ratio is called the crest factor. The mean value should be removed form the peak
amplitude to make this calculation.
The crest factor of a pure sine wave is √2. The standard deviation of a sine wave is thus
relatively close to its peak value.
Random vibration does not have a well-defined crest factor. A value of 3 is often assumed,
however.
4
Kurtosis
Kurtosis is a measurement of the “peakedness" of the data. A sine wave, for example, has a
kurtosis value of 1.5. A pure Gaussian random signal has a kurtosis value of 3.0. A higher
kurtosis value may indicate the presence of higher sigma peaks on the time history than would be
expected from a Gaussian distribution. These peaks could, for example, represent a transient
event. On the other hand, they may simply represent random vibration which has a probability
density function which departs from the Gaussian ideal.
The kurtosis equation is:
kurtosis =
∑ (x − X) 4
(20)
nσ4
Note that the kurtosis value appears dimensionless.
Histogram
Consider the sine function in Figure 1. The sine function is a digital signal with 4000 points.
The amplitude is 10 units. Now divide the amplitude into bands, each representing a range of 1.
Count how many points fall into each band. The result is a histogram, as shown in Figure 2.
Note that the histogram has a "bathtub" shape. The amplitude tends to remain either at the
positive or negative peak. Specifically, the probability is 28% that the signal amplitude has an
absolute value greater than 9 units.
A related function is the probability density function. This is obtained by dividing each
histogram amplitude by the number of total points.
Amplitude Specification
There are several methods for specifying the amplitude of a sine oscillation. Assume that the
oscillation begins at zero and reaches a maximum value of A.
Table 1.
Amplitude Parameter for Sine Vibration
Amplitude Parameter Formula
Zero-to-Peak A
Peak-to-Peak 2A
RMS 0.707 A
Care should be taken to make sure the proper format is understood. For example, household
electricity in the U.S. is 110 V AC. This is actually the RMS value.
5
SINE FUNCTION EXAMPLE
15
10
5
AMPLITUDE (units)
-5
-10
-15
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
TIME (SEC)
Figure 1.
6
HISTOGRAM OF SINE FUNCTION, 4000 TOTAL POINTS
800
700
600
500
COUNTS
400
300
200
100
0
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
AMPLITUDE (units)
Figure 2.
7
Homework
1. Calculate the RMS value for a constant amplitude signal x(t) = A.
2. Calculate the standard deviation for the signal in problem 1.
3. A signal consists of two components:
2πt 
x (t ) = A + B sin  
T 
what is the standard deviation?
4. Plot the data file sine.txt. This is measured data taken on a floor adjacent to a wafer
polishing machine in a semiconductor facility. It is a velocity time history with amplitude in
units of (in/sec). It is not a pure sine signal, but measured data never really is. What is the
dominant frequency?
5. The program maxfind.exe generates descriptive statistics for a time history with two
columns: time(sec) and amplitude(units). Copy this file to same directory as the sine.txt
file. Then type:
maxfind sine.txt
What is the average, standard deviation, RMS, and kurtosis values of this signal?
Note that the maxfind.exe program assumes amplitude units of G, but this is irrelevant.
Substitute in/sec for G in your notes.
6. Pogo is a type of sinusoidal vibration that occurs in certain rocket vehicles due to combustion
instability. Read tutorial pogo.pdf to learn more about this effect. This tutorial is not
concerned with descriptive statistics per se, but it gives a real-world example of sinusoidal
vibration.
8
Unit 2B. Sine Vibration Displacement, Velocity, and Acceleration
By Tom Irvine
Introduction
Again, consider a single-degree-of-freedom system undergoing sinusoidal excitation. The
displacement amplitude x(t) is
x ( t ) = X sin ( ω t ) (1)
where
X is the displacement
ω is the frequency (radians/time)
Note that each amplitude in this unit is taken as zero-to-peak.
The velocity x&( t ) is obtained by taking the derivative.
x&(t ) = ω X cos( ω t ) (2)
The acceleration &x&( t ) is obtained by taking the derivative of the velocity.
&x&(t ) = − ω 2 X sin ( ω t ) (3)
Disregard phase. Take the peak absolute values of each parameter. The peak values are
summarized in Table 1.
Table 1.
Peak Values Referenced to Peak Displacement.
Parameter Equation
Displacement x peak = X
Velocity x&peak = ω X
Acceleration &x&peak = ω 2 X
Note that
&x&peak = ω 2 x peak (4)
Now let A be the peak acceleration. The relationships in Table 2 can be derived via algebra.
1
Table 2.
Peak Values Referenced to Peak Acceleration
Parameter Equation
Displacement x peak = A / ω 2
Velocity x&peak = A / ω
Acceleration &x&peak = A
The equations in Table 2 have enormous consequence in vibration testing, as explained in the
next section.
Shaker Table
A satellite is shown mounted to an electromagnetic shaker table in Figure 1. The shaker supplies
a base excitation to the test item. These types of shakers are used to test a variety of components
including small RF couplers, flight computers, and spacecraft.
Figure 1.
2
This type of testing is used extensively in the aerospace industry. It is also used in the
automotive and other industries.
The testing is typically performed in a closed-loop manner. A test specification is entered into
the control computer. The computer outputs a signal to the shaker via a power amplifier. The
shaker vibrates in response, applying a base excitation to the test item. A control accelerometer
is placed on the shaker. This accelerometer is part of the feedback loop. The control computer
monitors the accelerometer. It computes a system gain factor, thereby adjusting the output to
achieve the test specification.
Specification Format
The test specification may come from a reference document such as MIL-STD-810E, MIL-STD-
1540C, or NAVMAT P9492. Or a custom test specification may have been derived for the
particular test item.
The test specification might be a random test, sine sweep, or some other format.
A test could be specified in terms of displacement, velocity, acceleration. The most common
specification parameter is acceleration, however. The reason is that acceleration is easiest of the
three parameters to measure.
Acceleration can be measured with a piezoelectric accelerometer or with a piezoresistive
accelerometer.
Velocity measurements require a Doppler laser or a geophone. The laser is expensive. The
geophone is bulky and is intended for seismology measurements.
Dynamic displacement can be measured by a linear variable displacement transducer (LVDT).
Unfortunately, the frequency response is only suited for low-frequency measurements.
If velocity is required, the typical method is to integrate the acceleration signal. If displacement
is required, the typical method is to double-integrate the acceleration signal.
Thus, acceleration is the amplitude parameter of choice for most vibration test specifications.
Recall the displacement equation from Table 2.
x peak = A / ω 2 (5)
Now consider a sine sweep specification where a test item is to be subjected to a base excitation
of 10 G amplitude. The frequency will be varied, starting with a low frequency and sweeping
upward to a higher frequency. The amplitude will be held constant, however. The displacement
at selected frequencies is shown in Table 3. The displacement is calculated per equation (5).
3
Table 2.
Displacement for 10 G Sine Excitation
Frequency (Hz) Displacement
(inches zero-to-peak)
0.1 9780
1 97.8
10 0.978
15 0.435
20 0.245
50 0.0391
100 0.00978
1000 9.78e-005
The shaker, however, has some displacement limit. A typical limit might be 1 inch peak-to-
peak, or 0.5 inch zero-to-peak. Thus, 15 Hz would be the practical lower frequency limit for the
10 G sine test. The test could be performed at lower frequencies, but the acceleration amplitude
would need to be reduced to the limit the displacement.
Most vibration test specifications thus begin at a frequency of 10 Hz or 20 Hz. Examples will be
given in upcoming units.
Homework
1. File trailer.txt is measured acceleration data from a big-rig trailer tested at an automotive
proving ground oval. The data is not purely sinusoidal, but is reasonably so. First, determine
the frequency in Hertz. Then use equation (5) to estimate the peak displacement. Remember
2
to convert G to in/sec , as part of the calculation. You may use a hand calculation or Excel.
2. Check your calculation from problem 1 using program sine.exe. Use the pure sine option in
this program.
3. Use sine.exe for this problem. A shaker has a displacement limit of 1 inch peak-to-peak, or
0.5 inch zero-to-peak. What is the maximum velocity at a frequency of 10 Hz?
4
Unit 3. Sine Sweep Frequency and Octave Calculations
By Tom Irvine
Introduction
A common specification for a base excitation test is a sine sweep test. An example is
shown in Figure 1.
SINE SWEEP EXAMPLE

ACCEL (G)
TIME (SEC)
Figure 1.
The purpose of a sine sweep test might be to identify natural frequencies of the test item.
Another purpose might be to verify the design or workmanship of the item with respect to
vibration.1 For example, consider an avionics component subjected to a sine sweep test.
The test is considered successful if the avionics component operates properly before,
1
An argument can be made that random vibration is more effective for uncovering design or
workmanship flaws. Nevertheless, some specifications require sine sweep vibration.
1
during, and after the vibration test. Note that the component is electrically powered and
monitored during the vibration test.
The essence of a sine sweep test is that the base excitation input consists of a single
frequency at any given time. The frequency itself, however, is varied with time.
The sine sweep test may begin at a low frequency and then sweep to a high frequency, or
vice-versa. Some specifications require several cycles, where one cycle is defined as
from low to high frequency and then from high back to low frequency.
The specification might require either a linear or a logarithmic sweep rate. The sweep
will spend greater time at the lower frequency end if the sweep is logarithmic. The
example in Figure 1 has a logarithmic sweep rate.
The amplitude in Figure 1 is constant. Nevertheless, the specification might require that
the amplitude vary with frequency.
Specification Example
A vendor has a product that must withstand sinusoidal vibration with an amplitude of 12
G. The desired frequency domain is 10 Hz to 500 Hz. The shaker table has a
displacement limit of 1.0 inch peak-to-peak, or 0.5 inch zero-to-peak.
Recall from Unit 2B that the displacement limit is a constraint at low frequencies. How
should the test be specified?
The answer is to use a specification with two amplitude segments. The first segment is a
ramp. The second segment is a plateau.
Furthermore, the first segment will be given in terms of displacement. The second
segment will be given in terms of acceleration. The result is shown in Figure 2.
In this case, the ramp is 1 inch peak-to-peak. The plateau is 12 G. Note that the G value
is understood by convention to be in terms of zero-to-peak. The compromise is that the
acceleration amplitude at 10 Hz is 5 G. The compromise is needed for the sake of
"testability."
2
SINE SWEEP SPECIFICATION EXAMPLE
100
Ramp is 1.0 inch peak-peak. Plateau is 12 G.
ACCEL (G)
10
1
10 15.3 20 50 100 200 500
FREQUENCY (Hz)
Figure 2.
The "crossover" frequency is 15.3 Hz. This is the frequency at which a 12 G acceleration
has a corresponding displacement of 1.0 inch peak-to-peak.
The crossover frequency f cross is calculated via equation (1).
1 &x&peak
f cross = (1)
2π x peak
where
&x&peak is the peak acceleration (zero-to-peak)
x peak is the peak displacement (zero-to-peak)
2 2
Furthermore, the acceleration should be converted from G to in/sec or G to m/sec , as
appropriate.
3
Octaves
One octave is defined as a frequency band where the upper frequency limit is equal to
twice the lower frequency limit. Thus a band from 10 Hz to 20 Hz is one octave.
Likewise, the band from 20 Hz to 40 Hz is an octave.
The band center frequency fc is taken as
fc = fl fu (2)
where
f l is the lower frequency
f u is the upper frequency
The band center frequency calculation is not particularly important for sine sweep
calculations. It will be important in later Units, however.
A greater concern to sine sweep testing is the total number of octaves. As an example
consider the following frequency sequence in Hertz.
10 - 20 - 40 - 80 -160 - 320 - 640 - 1280 - 2560
The sequence has a total of eight octaves.

Now consider a sine sweep test from 10 Hz to 2000 Hz. How many octaves are in this
example? A rough estimate is 7.5. Nevertheless, the exact number is needed.
The number of octaves n can be calculated in terms of natural logarithms as
f 
ln  2 
 f1 
n= (3)
ln []
2
where
f l is the lower frequency of the specification
f 2 is the upper frequency of the specification
Thus, the frequency domain from 10 Hz to 2000 Hz has 7.64 octaves, per equation (3).
4
The number of octaves is then used to set the sweep rate, assuming a logarithmic rate.
For example, the rate might be specified as 1 octave/minute. The excitation frequency at
any time can then be calculated from this rate.
Or perhaps the total sweep time from 10 Hz to 2000 Hz is specified as 8 minutes. Thus,
the sweep rate is 0.955 octaves/min.
Homework
1. A shaker table has a displacement limit of 1.5 inch peak-to-peak, or 0.75 inch zero-to-
peak. An amplitude of 28 G is desired from 10 Hz to 2000 Hz. The specification will
consist of a displacement ramp and an acceleration plateau. What should the
crossover frequency be? What is the maximum acceleration at 10 Hz?
2. How many octaves are in the frequency domain from 10 Hz to 500 Hz?
3. Use program sine.exe to check your work in problems 1 and 2.
4. Program generate.exe can be used to synthesize a variety of signals. Generate and

plot a few sample signals. Note: the sample rate should be at least ten times higher
than the maximum frequency.
5. Optional. Read tutorial sweep.pdf.
5
Unit 4. Random Vibration Characteristics
By Tom Irvine
Introduction
Random Forcing Function and Response
Consider a turbulent airflow passing over an aircraft wing. The turbulent airflow is a forcing
function. Furthermore, the turbulent pressure at a particular location on the wing varies in a
random manner with time.
For simplicity, consider the aircraft wing to be a single-degree-of-freedom system. The wing
would vibrate in a sinusoidal manner if it were disturbed from its rest position and then allowed
to vibrate freely. The turbulent airflow, however, forces the wing to undergo a random vibration
response.
Random Base Excitation

As another example, consider earthquake motion. The ground vibrates in random manner during
the transient duration.
Common Characteristics
One common characteristic of these examples is that the motion varies randomly with time.
Thus, the amplitude cannot be expressed in terms of a "deterministic" mathematical function.
Dave Steinberg wrote in Reference 1:

The most obvious characteristic of random vibration is that it is nonperiodic. A knowledge
of the past history of random motion is adequate to predict the probability of occurrence of
various acceleration and displacement magnitudes, but it is not sufficient to predict the
precise magnitude at a specific instant.
Frequency Content
Pure sinusoidal vibration is composed of a single frequency. On the other hand, random
vibration is composed of a multitude of frequencies. In fact, random vibration is composed of a
continuous spectrum of frequencies.
Random vibration is somewhat analogous to white light.. White light can be passed through a
prism to reveal a continuous spectrum of colors. Likewise, random vibration can be passed
through a spectrum analyzer to reveal a continuous spectrum of frequencies.
On the other hand, sinusoidal vibration is analogous to a laser beam, where the light wave is
composed of a single frequency.
1
Statistics of a Random Vibration Sample
A sample random vibration time history is shown in Figure 1. This time history was
"synthesized," or generated analytically. It has the descriptive statistics shown in Table 1.
Table 1.
Random Vibration Descriptive Statistics
Parameter Value
Duration 4 sec
Samples 4000
Mean 0.00
Std dev 0.59 G
RMS 0.59 GRMS
Kurtosis 3.04
Maximum 2.30 G
Minimum -1.96 G
The statistical parameters were calculated using the formulas in Unit 2A, as implemented in the
maxfind.exe program.
Recall that pure sine vibration has a peak value that is √2 times its RMS value.
On the other hand, random vibration has no fixed ratio between its peak and RMS values. The
ratio between the absolute peak and RMS values in Table 1 is
2.30 G peak
= 3.90 (1)
0.59 G RMS
Also recall, that the RMS value is equal to the standard deviation value if the mean is zero. The
standard deviation is often represented by sigma, σ.
Thus, the sample in Figure 1 has a peak value of 3.90 σ. A different random sample could have
a higher or lower peak value in terms of its σ, however.
A typical assumption is that random vibration has a peak value of 3.0 σ for design purposes.
Again, the example in Figure 1 deviates from this assumption with its peak value of 3.90 σ.
2
SAMPLE RANDOM VIBRATION
1
ACCEL (G)
-1
-2
-3
-4
-5
0 1 2 3 4
TIME (SEC)
Figure 1.
The time history of Figure 1 is shown again in Figure 2. The amplitude in Figure 2 is scaled in
terms of the σ value. According to theory, the amplitude should be within the ±1σ limits 68.26%
of the time.
3
SAMPLE RANDOM VIBRATION
4σ
3σ
2σ
1σ
ACCEL
0
-1σ
-2σ
-3σ
-4σ
0 1 2 3 4
TIME (SEC)
Figure 2.
Histogram
The histogram of the time history in Figure 1 is shown in Figure 3.
Note that the histogram of the random vibration sample has a "bell-shaped" curve. The
histogram is an approximate example of a Gaussian or normal distribution.
The histogram shows that the random vibration signal has a tendency to remain near its mean
value, which in this case is zero.
In contrast, recall the histogram of the sinusoidal time history in Unit 2A. The histogram of a
sinusoidal signal has the shape of a bathtub. Sine vibration thus tends to remain at its positive
and negative peak values.
For this and other reasons, sine and random are two very different forms of vibration. There
really is no "equivalency" between the two forms, although many engineers have tried to derive a
relation.
4
HISTOGRAM OF SAMPLE RANDOM VIBRATION
1000
4000 samples total.
900
800
700
600
COUNTS
500
400
300
200
100
0
-2.5 -2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 2.5
ACCELERATION (G)
Figure 3.
Probability Density Function

The histogram in Figure 2 can be converted to a "probability density function." This would be
done by dividing each bar by the total number of samples, 4000 in this case. The resulting
function would be a probability density function. Furthermore, the amplitude along the X-axis
could be represented in terms of σ. The resulting probability density function would then
approximate a "normal probability density function."
The equation which characterizes the normal probability function is well-known. It is available
in References 1 and 2. This equation can be integrated to determine the probability that an
amplitude will occur inside or outside certain limits.
5
Normal Probability Values
Consider a random vibration time history x(t). Again, the amplitude x(t) cannot be calculated for
a given time.
Nevertheless, the probability that x(t) is inside or outside of certain limits can be expressed in
terms of statistical theory.
The probability values for the amplitude are given in Tables 2a and 2b for selected levels in
terms of the standard deviation or σ value.
Table 2a. Probability for a Random Signal with Normal Distribution and Zero Mean
Statement Probability Ratio Percent Probability
-σ < x < +σ 0.6827 68.27%
-2σ < x < +2σ 0.9545 95.45%
-3σ < x < +3σ 0.9973 99.73%
Table 2b. Probability for a Random Signal with Normal Distribution and Zero Mean
Statement Probability Ratio Percent Probability
|x|>σ 0.3173 31.73%
| x | > 2σ 0.0455 4.55%
| x | > 3σ 0.0027 0.27%
The probability tables can be used as follows. Suppose that a random vibration time history has
a total duration of 60 seconds. For what amount of time will the amplitude exceed 1σ in terms
of absolute value?
( 60 sec )( 0.3173 ) = 18.22 sec (2)
Suppose the same time history is digitized such that it consists of 200,000 points. How many
points will exceed 2σ in terms of absolute value?
( 200,000 )( 0.0455 ) = 9100 samples (3)
Types of Random Vibration

Random vibration can be broadband or narrow band. It can be stationary or non-stationary. In
addition, white noise and pink noise are two special cases of random vibration. These
distinctions are covered in upcoming units.
Sometimes, measured data has a "sine-on-random" characteristic. Recall, the rocket vehicle drop
transient data from Units 1A and 1B. The raw data contained some random noise. Nevertheless,
the decaying sinusoidal signal dominated the response. The data as given to the student,
however, was "bandpass filtered" to clarify the decaying sinusoid. Filtering will be covered in a
future unit.
6
Computer hard disk drives are an example of device where both sine and random vibration
environments are a concern, as discussed in Appendix A, located after the homework section.
Furthermore, note that a great deal of analysis effort is spent searching vibration data for
particular sinusoids which may be hidden inside a random signal.
References
1. D. Steinberg, “Vibration Analysis for Electronic Equipment,” Wiley-Interscience, New York,

1988.
2. T. Irvine, Integration of the Normal Distribution Curve, Vibrationdata Publications, 1999.
Homework
1. Use program generate.exe to synthesize a white noise random vibration time history with a
standard deviation value of 1, duration of 5 seconds, and sample rate of 1000 samples/sec.
Plot the resulting time history. Use maxfind.exe to evaluate its statistical parameters. What
is the ratio between the absolute peak value and the standard deviation value?
2. a. Use the program stats.exe.

b. Select normal distribution.
c. Select "Calculate probability for a given Z value." Note that Z represents σ assuming a
zero mean.
d. Click on -Z to +Z.
e. Input Z = 3.2, which is equivalent to 3.2 σ.
f. What is the probability that a random vibration time history will have peaks in excess of
3.2 σ in terms of absolute value?
g. Given a time history with 4000 points. How may will exceed 3.2 σ in terms of absolute
value?
3. Read NAVMAT P-9492. It may be downloaded from:

http://www.vibrationdata.com/tutorials.htm
4. Consider an avionics component. It is powered and monitored during a bench test. It passes
this "functional test." Nevertheless, it may have some latent defects such as bad solder joints
or bad parts. A decision is made to subject the component to a base excitation test on a
shaker table to check for these defects. Which would be a more effective test: sine sweep or
random vibration? Why?
5. Review question. What is the kurtosis value of pure sine vibration? What is the kurtosis
value of broadband random vibration?
7
APPENDIX A
Hard Disk Drive Vibration
A hard disk is made from metal or some other rigid material. The disk is called a platter. It is
coated with a magnetic material that is used to store data as transitions of magnetic polarity.
Each polarity corresponds to a “1” or a “0.”
One or more platters are mounted on a single spindle shaft. The drive platters are divided into
cylinders.
Each drive may have a spin rate somewhere between 3600 rpm and 10,000 rpm. This
corresponds to the frequency domain between 60 Hz and 167 Hz. Future designs may have even
higher spin rates.
A read/write head is mounted on the actuator arm. There is typically one head on each side of
every platter.
The heads move in unison back and forth across the platter according to a control algorithm. The
algorithm compensates for the flexibility of the actuator arm, platter vibration, and other
disturbances.
Source Energy
Rotating imbalance, misalignment, and other defects may produce sinusoidal vibration at the
spin rate or at integer multiples thereof.
Movement of the head assembly produces additional vibration. For example, random seek
movement may cause rotational oscillations with a broadband random frequency content. This
oscillation causes in-plane rotation of the drive and wrapper assembly.
Error and Failure Modes

Hard drive units have a number of error and failure modes related to vibration.
Impact
A particular concern is that the head or arm might impact against the surface of one or more
disks, thereby creating voids in the recording film. This damage could cause data errors.
There are several means by which this impact could occur. For example, the vibration could
come from an external source that propagates into the arm, possibly exciting a head/arm natural
frequency. Another scenario is that excessive disk vibration causes the disk to impact against the
head or arm.
Stack Shift
Stack shift is also a concern. This occurs when individual disks shift from their initial center of
rotation. This shift could occur if the shock and vibration forces overcome the initial clamping
forces. Stack shift may produce a sinusoidal position error signal with a frequency equal to two
times (2X) the spin frequency.
8
Servo Control Algorithm
Vibration may interfere with the control algorithm. This is a particular concern during write
operations. The algorithm is designed to prevent existing data from being inadvertently
overwritten. Excessive vibration may cause a delay in the writing process. This condition is
called latency.
Control algorithms vary from one hard drive model to the next. Most algorithms, however tend
to be sensitive in the frequency domain from 500 Hz to 600 Hz.
9
Unit 5. Nonstationary Random Vibration
By Tom Irvine
Automobile Example
Consider that a vendor is developing a GPS navigation system for rental cars. The GPS system
interfaces with a computer mounted underneath the dashboard. The computer has a small
display screen with superimposed road maps so that the driver can navigate to his or her
destination.
The navigation computer must be designed to withstand vibration. The engine, transmission, and
aerodynamic effects are each sources of vibration. Nevertheless, the dominant source is
vibration transmitted from the road, through the tires and suspension, and into the vehicle
chassis.
Appropriate vibration test levels must thus be derived. The vendor begins by searching for
reference data in MIL-STD-810E, SAE standards, and other sources. Although some useful data
is available, the vendor reaches the inevitable conclusion that he must take field measurements to
establish the maximum expected vibration level for his specific product.
The vendor thus mounts accelerometers inside a test car adjacent to the navigation system
mounting location. A data acquisition system is used to monitor the data. A test driver then
drives the car through a variety of roads, including city streets and highways.
What type of vibration is expected? The data will probably show that vibration is random. An
exception might occur if a particular natural frequency is excited. This would be resonant
excitation. The natural frequency would have a sinusoidal time history response. Even so, the
amplitude of the sinusoid might vary in a somewhat random manner. For simplicity, consider
that the response will be random vibration.
The next question is whether the random vibration "characteristics" vary with speed, road
conditions, total vehicle weight, or other variables. The amount of variability will affect the
amount of data which must be collected and analyzed.
A possible outcome is that the character of the random vibration might be reasonably consistent
as the vehicle travels at 60 mph down a particular interstate highway. On the other hand, the
character could vary considerably as the vehicle experiences stop-and-go driving over a city
street filled with potholes, railroad tracks, and construction detours.
Definition
For simplicity, consider that random vibration can be "characterized" in terms of its statistical
properties, such as mean value, standard deviation, and kurtosis.
The random vibration is "stationary" if these statistical properties remain constant with time.
Otherwise, it is "nonstationary."
1
Stationary vibration is an idealized concept. All vibration is ultimately nonstationary.
Nevertheless, certain types of random vibration may be regarded as reasonably stationary.
For example, the vehicle in the previous example experiences stationary vibration as it travels at
a constant speed of 60 mph down a flat interstate highway. Or at least the vibration is stationary
as long as the vehicle remains under those conditions.
The vibration during the stop-and-go city driving is clearly non-stationary, however. The vehicle
experiences a series of transient vibration events as it crosses railroad tracks, encounters potholes
and other obstacles. Transient vibration is nonstationary.
Rocket Vehicle Example

A rocket vehicle clearly experiences nonstationary vibration during its powered flight. A sample
acceleration time history is shown in Figure 1.
ACCELERATION TIME HISTORY ROCKET VEHICLE

Top Curve is Instantaneous Time History
Bottom Curve is Standard Deviation Time History
30 20
20 18
(Left Scale)
10 16
0 14
STD DEV ACCEL (G)

-10 12
ACCEL (G)
-20 10
-30 8
-40 6
(Right Scale)
-50 4
-60 2
-70 0
0 10 20 30 40 50 60 70
TIME (SEC)
Figure 1.
2
Note that Figure 1 is a double-Y plot which contains two time history curves. The upper curve is
an instantaneous time history, as measured directly by the accelerometer. The bottom curve is a
standard deviation time history. For this example, a standard deviation value was calculated for
every 0.2 second interval.
There are several reasons for computing a standard deviation time history.
1. An instantaneous time history may have well over 100,000 data points. Plotting this
complete set may be impractical depending on the available computer hardware and graphing
software. On the other hand, the standard deviation calculation can reduce this set to a 100
or so points. Thus, the standard deviation time history is a "data reduction tool."
2. The standard deviation time history can be used to check whether a signal is stationary. The
example in Figure 1 is clearly nonstationary judging from the instantaneous time history.
Nevertheless, there may be other cases where the distinction is subtle.
3. The standard deviation time history can be used to clarify sinusoidal signals which are
masked by random vibration. The standard deviation of a sine function is 0.707 times the
peak value. On the other hand, the standard deviation of random vibration may be 0.30 times
the peak value. Thus, the standard deviation calculation favors sine vibration.
Qualitative Description of the Rocket Example

The acceleration time history in Figure 1 is measured data from a ground-launched rocket
vehicle. The rocket experience motor ignition vibration and launch acoustics effects during the
first few two seconds.
Thereafter, the vibration level decays to a relatively benign level for several seconds. During
this phase, the rocket is traveling at a subsonic speed. The main vibration source at this time is
the motor burn.
At 14 seconds, the vibration level abruptly increases. The vehicle is accelerating through the
transonic velocity. Shock waves form around the vehicle. Furthermore, the vehicle passes
through its "maximum dynamic pressure" condition. Thus, aerodynamic buffeting effects
become the dominant vibration source. The aerodynamic effects continue until about the 40
second mark.
The motor burn ends near the 60 second mark. Note that the vehicle has attitude control
thrusters. The thrusters use bursts of nitrogen gas to correct the orientation of the vehicle. The
thrusters were the main source of vibration from 60 to 65 seconds.
Homework
1. The file: rocket.txt is included. This is the data from Figure 1. Use program maxfind.exe to
check the descriptive statistics of the file. Note the number of samples. Judge whether your
graphics program is capable of plotting the instantaneous time history. If so, plot the data.
Otherwise, proceed to the next problem.
3
2. Program sstt.exe can be used to generate a standard deviation time history. The convention
is:
sstt rocket.txt 0.2
Note that the filename is specified in the command line. The number 0.2 directs the program to
calculate a standard deviation time history for every 0.2 second interval. This number can be
varied.
Actually, program sstt.exe generates three output files:
sa.dat - mean value vs. time

ss.dat - standard deviation vs. time
s.dat - four columns: time, mean, std dev, RMS.
The homework problem is to practice plotting these files. Also, experiment with various time
interval values. For example, change 0.2 to 0.1 and repeat the calculation.
3. Is kurtosis a useful parameter for evaluating nonstationary random vibration? Recall the
kurtosis value from problem 1.
4. Use program generate.exe to synthesize a white noise time history.
Set the duration to 12 seconds.

Use a sample rate of 500 samples per second.
Set the standard deviation value to 1.
Call the output file: white.out
Now use program sstt to calculate the standard deviation time history of the white noise time
history. Use an interval of 1.
sstt white.out 1.
Does the standard deviation time history tend to shown that the white noise time history is
stationary or nonstationary? For the purpose of this problem, assume that the signal is stationary
if the standard deviation values are within, say 20% of one another.
4
Unit 6A. The Fourier Transform
By Tom Irvine
Introduction
Stationary vibration signals can be placed along a continuum in terms of the their qualitative
characteristics.
A pure sine oscillation is at one end of the continuum. A form of broadband random vibration
called white noise is at the other end.
Reasonable examples of each extreme occur in the physical world. Most signals, however, are
somewhere in the middle of the continuum. An example is shown in Figure 1.
TIME HISTORY EXAMPLE
5
ACCEL (G)
-5
0 0.5 1.0 1.5 2.0
TIME (SEC)
Figure 1.
1
The time history in Figure 1 appears to be the sum of several sine functions. What are the
frequencies and amplitudes of the components? Resolving this question is the goal of this Unit.
At the risk of short-circuiting the process, the equation of the signal in Figure 1 is
y ( t ) = 1.0 sin [2π (10 )t ]+ 1.5 sin [2π (16 )t ]+ 1.2 sin [2π ( 22 )t ] (1)
The signal thus consists of three components with frequencies of 10, 16, and 22 Hz, respectively.
The respective amplitudes are 1.0, 1.5, and 1.2 G.
In addition, each component could have had a phase angle. In this example, the phase angle was
zero for each component.
Thus, we seek some sort of "spectral function" to display the frequency and amplitude data.
Ideally, the spectral function would have the form shown in Figure 2.
"SPECTRAL FUNCTION" OF TIME HISTORY EXAMPLE
2.0
1.5
ACCEL (G)
1.0
0.5
0
0 5 10 15 20 25
FREQUENCY (Hz)
Figure 2.
2
Some engineers would claim that Figure 2 is the Fourier transform of the signal in equation (1).
In some sense, this is true.
The Fourier transform, however, is… .well… a methodology. Furthermore, it is a methodology
which can be applied via many different forms and formulas. Here, the author is tempted to
draw some analogy to philosophy or religion.
Mathematicians and engineers make such different use of Fourier transforms that a
mathematician would likely be unable to comprehend an engineer's application and vice versa.
This Unit will attempt to bridge the gap. A method will eventually be derived to transform the
time history in Figure 1 to the frequency domain "spectral function" in Figure 2. The desired
"spectral function" will be shown to be based on the Fourier transform. Nevertheless, the phrase
which most aptly describes this process is "some assembly required."
Continuous Fourier Transform

The Fourier transform is a method for representing a time history signal in terms of a frequency
domain function.
The Fourier transform is a complex exponential transform which is related to the Laplace
transform.
The Fourier transform is also referred to as a trigonometric transformation since the complex
exponential function can be represented in terms of trigonometric functions. Specifically,
exp[ jω t ] = cos(ω t ) + j sin(ω t ) (2a)
exp[ − j ω t ] = cos(ω t ) − j sin(ω t ) (2b)
where j = − 1
The Fourier transform X(f) for a continuous time series x(t) is defined as
∞
X(f) = ∫ x(t)exp[-j2 π f t ]dt (3)
-∞
where -∞ < f < ∞
Thus, the Fourier transform is continuous over an infinite frequency range.
3
The inverse transform is
∞
x(t) = ∫ X(f)exp[+ j2 π f t ]df (4)
-∞
Equations (3) and (4) are taken from Reference 1. Note that X(f) has dimensions of
[amplitude-time].
Also note that X(f) is a complex function. It may be represented in terms of real and imaginary
components, or in terms of magnitude and phase.
The conversion to magnitude and phase is made as follows for a complex variable V.
V = a + jb (5)
Magnitude V = a2 + b2 (6)
Phase V = arctan( b / a) (7)
Note that the inverse Fourier transform in equation (4) calculates the original time history in a
complex form. The inverse Fourier transform will be entirely real if the original time history
was real, however.
Continuous Example
Consider a sine function
x( t ) = A sin 2π f$ t[ ] (8)
where
-∞ < t < ∞
The Fourier transform of the sine function is
 jA 
 2 
{ ( ) (
X( f ) =   − δ f − f$ + δ − f − f$ )} (9)
where δis the Dirac delta function.

Note that
( )
δf − fˆ = 0 for f ≠ fˆ (10)
And
∫− ∞ δ(f − fˆ)dt = 1
∞
(11)
4
The derivation is given in Appendix A. The Fourier transform is plotted in Figure 3.
Imaginary X(f)
A
2
(
δ − f − f$ )
− f$ f$ − A
δ( f − f$ )
2
Figure 3. Fourier Transform of a Sine Function
The transform of a sine function is purely imaginary. The real component,

which is zero, is not plotted.
On the other hand, the Fourier transform of a cosine function is
A
2
{( ) (
X( f ) =   δ f − f$ + δ − f − f$ )} (12)
The Fourier transform is plotted in Figure 4.
Real X(f)
A
2
(
δ − f − f$ ) A
(
δ f − f$ )
2
− f$ f$ f
Figure 4. Fourier Transform of a Cosine Function

The transform of a cosine function is purely real. The imaginary component,
which is zero, is not plotted.
5
Characteristics of the Continuous Fourier Transform
The plots in Figures 1 and 2 demonstrate two characteristics of the Fourier transforms of real
time history functions:
1. The real Fourier transform is symmetric about the f = 0 line.

2. The imaginary Fourier transform is antisymmetric about the f = 0 line.
As an aside, the Dirac delta function is purely delightful from a mathematics point of view.
Some mathematicians even promote it from a lowly function to a "distribution."
The Dirac delta distribution is of little or no use to the engineer in the test lab, however. A
different approach is needed for engineers.
Discrete Fourier Transform

An accelerometer returns an analog signal. The analog signal could be displayed in a continuous
form on a traditional oscilloscope.
Current practice, however, is to digitize the signal, which allows for post-processing on a digital
computer. Thus, the Fourier transform equation must be modified to accommodate digital data.
This is essentially the dividing line between mathematicians and engineers in regard Fourier
transformation methodology. Nevertheless, further assembly is required to meet the engineering
goal, which is still the "spectral function" in Figure 2.
The discrete Fourier Transform F̂k for a digital time series x n is
N− 1
  2π 
F̂k = ∆ t ∑  x n exp − j nk  ,
 N 
for k = 0, 1, ... , N − 1 (13)
n = 0
where
N is the number of time domain samples,
n is the time domain sample index,
k is the frequency domain index,
∆ t is the time step between adjacent points.
Note that F̂k has dimensions of [amplitude-time].
The corresponding inverse transform is
N− 1
  2π 
xn = ∆ f ∑  F̂k exp + j nk  , for n = 0, 1, ..., N − 1
 
(14)
n = 0
N
Note that the frequency increment ∆ f is equal to the time domain period T as follows
6
1
∆f = (15)
T
The frequency is obtained from the index parameter k as follows
frequency (k) = k ∆ f (16)
The discrete Fourier transform in equation (13) requires further modification to meet the
engineering goal set forth in Figure 2.
The following equation set is taken from Reference 2. As an alternate form, the Fourier
transform Fk for a discrete time series x n can be expressed as
N− 1
  2 π 
∑
1
Fk =  x n exp − j nk  , for k = 0, 1, ..., N − 1 (17)
N   N 
n= 0
N− 1
  2π 
xn = ∑  F k exp + j nk  , for n = 0, 1, ..., N − 1
  N 
(18)
k= 0
Note that Fk has dimensions of [amplitude]. Thus, an important milestone is reached.
Discrete Example
The discrete Fourier transform of a sine wave is given in Figure 5.
A characteristic of the discrete Fourier transform is that the frequency domain is taken from 0 to
(N-1)∆ f. The line of symmetry is at a frequency of
 N − 1
 2 ∆ f (19)
Nyquist Frequency
Note that the line of symmetry in Figure 5 marks the Nyquist frequency. The Nyquist frequency
is equal to one-half of the sampling rate. Shannon’s sampling theorem states that a sampled time
signal must not contain components at frequencies above half the Nyquist frequency, from
Reference 3.
7
HALF-AMPLITUDE DISCRETE FOURIER TRANSFORM OF
y(t) = 1 sin [ 2π (1 Hz) t ] G
1.0
N = 16384 Dashed Line is Line of Symmetry
∆ t = ( 1 / 32 ) sec
T = 512 sec
∆ f = ( 1 / 512 ) Hz
0.5
ACCELERATION (G)
-0.5
-1.0
0 1 2 3 4 5 6 7 8 9 10 1112 13 14 15 1617 18 19 2021 22 23 24 2526 27 28 2930 31 32
FREQUENCY (Hz)
Figure 5. Fourier Transform of a Sine Wave
Note that the sine wave has a frequency of 1 Hz. The total number of cycles is 512,
with a resulting period of 512 seconds. Again, the Fourier transform of a sine wave is
imaginary and antisymmetric. The real component, which is zero, is not plotted.
Spectrum Analyzer Approach

Spectrum analyzer devices typically represent the Fourier transform in terms of magnitude and
phase rather than real and imaginary components. Furthermore, spectrum analyzers typically
only show one-half the total frequency band due to the symmetry relationship.
The spectrum analyzer amplitude may either represent the half-amplitude or the full-amplitude
of the spectral components. Care must be taken to understand the particular convention of the
spectrum analyzer. Note that the half-amplitude convention has been represented in the
equations thus far, particularly equations (14) and (17).
8
The full-amplitude Fourier transform magnitude G k would be calculated as
  N− 1 

 magnitude 
1
 N  ∑ {x n } for k = 0
  
 n= 0
Gk = 
  2 N − 1 
    2π  
 2 magnitude
∑
N
     x n exp − j nk   for k = 1, ..., − 1
   N  n = 0   N  

2
with N as an even integer.

(20)
Note that k = 0 is a special case. The Fourier transform at this frequency is already at full-
amplitude.
For example, a sine wave with an amplitude of 1 G and a frequency of 1 Hz would simply have a
full-amplitude Fourier magnitude of 1 G at 1 Hz, as shown in Figure 6.
FULL-AMPLITUDE, ONE-SIDED DISCRETE FOURIER TRANSFORM OF

y(t) = 1 sin [ 2π (1 Hz) t ] G
1.5
N = 16384
∆ t = ( 1 / 32 ) sec
T = 512 sec
∆ f = ( 1 / 512 ) Hz
1.0
ACCELERATION (G)
0.5
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
FREQUENCY (Hz)
Figure 6.
9
Goal
The sine function considered in Figures 5 and 6 had a long duration of 512 seconds. The time
history in Figure 1 has a duration of only 2 seconds, however. Note that the Fourier transform
frequency resolution is the inverse of the duration, as given in equation (15). The frequency
resolution is thus 0.5 Hz for a duration of 2 seconds.
The full-amplitude Fourier transform of the time history in Figure 1 is given in Figure 7. The
"spectral function" goal is thus reasonably met, at least for this example. The course frequency
resolution, however, gives the spectral lines a peak shape.
FULL-AMPLITUDE, ONE-SIDED DISCRETE FOURIER TRANSFORM OF

TIME HISTORY IN FIGURE 1
2.0
N = 400
∆ t = 0.0025 sec
T = 2 sec
∆ f = 0.5 Hz
1.5
ACCELERATION (G)
1.0
0.5
0
5 10 15 20 25
FREQUENCY (Hz)
Figure 7.
The 10, 16, and 22 Hz sinusoidal frequencies are thus clearly apparent in Figure 7. The
corresponding amplitudes are also correct per equation (1).
10
Note that this example is somewhat idealistic. The Fourier transform data in Figure 7 is defined
at each 0.5 Hz frequency increment, beginning at 0. Thus, three of the spectral lines occur
exactly at 10, 16, and 22 Hz.
What if the 10 Hz component in equation (1) were shifted to 9.75 Hz? The answer is that some
of the energy would be shifted to 9.5 Hz and some to 10.0 Hz in the Fourier transform. This
effect is one of several error sources in the Fourier transform. This error can be avoided by
taking a longer duration.
Other error sources will be discussed in upcoming units. At length, the Fourier transform will be
shown to be a marginal "spectral function" approach even using the full-amplitude equation (20).
Nevertheless, more suitable tools can be built from the Fourier transform, as shown in upcoming
units. Thus, "further assembly required."
Preview of Unit 6B
The discrete Fourier transform requires a tremendous amount of calculations. A Fast Fourier
transform should be used if the number of time history samples is greater than 5000. The Fast
Fourier transform is covered in Unit 6B.
Homework
1. Convert the following complex number into magnitude and phase: x = 5 + j 9
2. A time history has a duration of 20 seconds. What is the frequency resolution of the Fourier
transform?
3. Recall file white.out from Unit 5. Take the Fourier transform using program fourier.exe.
Plot both the half.out and full.out files. How does the amplitude vary with frequency?
4. Plot the half-sine time history hs.txt. Then take the Fourier transform. How does the
amplitude vary with frequency?
References
1. W. Thomson, Theory of Vibration with Applications, 2nd Ed, Prentice-Hall, 1981.

2. GenRad TSL25 Time Series Language for 2500-Series Systems, Santa Clara, California,
1981.
3. R. Randall, Frequency Analysis 3rd edition, Bruel & Kjaer, 1987.
4. F. Harris, Trigonometric Transforms, Scientific-Atlanta, Technical Publication DSP-005, San
Diego, CA.
5. T. Irvine, Statistical Degrees of Freedom, Vibrationdata Publications, 1998.
11
APPENDIX A
Consider a sine wave
[
x( t ) = A sin 2π f$ t ] (A-1)
where
-∞ < t < ∞
The Fourier transform is calculated indirectly, by considering the inverse transform. Note that
the sine wave is a special case in this regard.
Recall
∞
[
x(t) = ∫ X(f)exp + j2 π f t df
-∞
] (A-2)
Thus
[ ] ∞
A sin 2 π f$ t = ∫ X( f ) exp + j 2 π f t df
− ∞
[ ] (A-3)
[ ] ∞
{ [ ]
A sin 2 π f$ t = ∫ X( f ) cos 2 π f t + j sin 2 π f t df
− ∞
[ ]} (A-4)
Let
X(f) = P(f) + j Q(f) (A-5)
where
P(f) and Q(f) are both real coefficients
and
-∞ < f < ∞ .
[ ]
A sin 2 π f$ t = ∫
∞
− ∞
{ P( f ) + { [ ] [ ]}
j Q( f )} cos 2 π f t + j sin 2 π f t df (A-6)
12
[ ]
A sin 2 π f$ t = ∫
∞
− ∞
{ P( f ) cos[2π f t ]− Q(f ) sin[2π f t ]}df
(A-7)
j ∫ { P ( f ) sin[2 π f t ]+ Q( f ) cos[2 π f t ]}df
∞
+
− ∞
Equation (A-7) can be broken into two parts
[
A sin 2 π f$ t = ∫]
∞
− ∞
{ P(f ) cos[2π f t ]− [ ]}
Q( f ) sin 2 π f t df (A-8)
0 = j∫
∞
− ∞
{ P( f ) sin[2π f t]+ [ ]}
Q( f ) cos 2 π f t df (A-9)
Consider equation (A-8)
[ ]
A sin 2 π f$ t = ∫
∞
− ∞
{ P(f ) cos[2π f t ]− [ ]}
Q( f ) sin 2 π f t df (A-10)
Now assume
P(f)=0 (A-11)
With this assumption,
[ ∞
]
A sin 2 π f$ t = − ∫ Q( f ) sin 2 π f t df
− ∞
[ ] (A-12)
Now let
Q(f)= q1 (f) + q2 (f) (A-13)
[ ] ∫−∞∞ [q1 (f ) +
A sin 2 π f$ t = − ] [ ]
q 2 ( f ) sin 2 π f t df (A-14)
[ ]
A sin 2 π f$ t = − ∫
∞
− ∞
[q 1 ( f )]sin[2 π f t ]dt − ∫ [q 2 ( f )]sin[2 π f t ]df
∞
− ∞
(A-15)
[ ]
A sin 2 π f$ t = − ∫
∞
− ∞
[q 1 ( f )]sin[2π f t ]dt + ∫ [q 2 ( f )]sin[−
∞
− ∞
]
2 π f t df (A-16)
13
Equation (A-14) is satisfied by the pair of equations
q1 ( f ) = −
A
2
(
δ f − f$ ) (A-17)
q 2 (f ) =
A
2
(
δ − f − f$ ) (A-18)
where δis the Dirac delta function.
By substitution,
− A
Q( f ) =
2
(
δ f − f$ +
A
2
)
δ − f − f$ ( ) (A-19)
Verification must be made that equation (A-9) is satisfied. Recall
0 = j∫
∞
− ∞
{ P( f ) sin[2π f t]+ [ ]}
Q( f ) cos 2 π f t df (A-20)
∞  − A 
( 
) ( )
?
0 sin[2 π f t ]+  δ − f − f$  cos[2 π f t ] df
A
0 = j∫  δ f − f$ + (A-21)
− ∞   2 2  
− A 
[ ] [ ]
? A
0 = j cos 2 πf$ t + cos − 2 πf$ t  (A-22)
 2 2 
− A 
[ ] [ ]
? A
0 = j cos 2π f$ t + cos 2 π f$ t  (A-23)
 2 2 
0=0 (A-24)
Recall the time domain function
[
x( t ) = A sin 2π f$ t ] (A-25)
where
-∞ < t < ∞
14
The Fourier transform is thus
− jA
X( f ) =
2
( )
δ f − f$ +
jA
2
(
δ − f − f$ ) (A-26)
 jA 
 2 
{ ( ) (
X( f ) =   − δ f − f$ + δ − f − f$ )} (A-27)
15
Unit 6B. Notes on the Fourier Transform Magnitude
By Tom Irvine
Introduction
Fourier transforms, which were introduced in Unit 6A, have a number of potential error
sources and other peculiar characteristics. The purpose of this unit is to discuss the
Fourier transform magnitude, which must be interpreted with great care.
Sine Example
Consider a sine function with a 1 Hz frequency and 1 G amplitude. Let the period be 20
seconds, which is equivalent to 20 cycles. Thus, ∆f = 0.05 Hz. The corresponding
Fourier transform magnitude is shown in Figure 1.
Now define the same sine function over a period of 40 seconds. Thus, ∆f = 0.025 Hz.
The Fourier transform magnitude is shown in Figure 2.
The Fourier transform magnitude at 1 Hz is 1 G in each case, independent of the duration
difference. Thus, the Fourier transform magnitude is shown to be a good tool for
resolving sinusoidal amplitudes.
In each Fourier transform, there is a spectral line exactly at a frequency of 1 Hz.
Otherwise, the acceleration amplitude would be smeared between frequencies adjacent to
1 Hz. This smearing effect is not a concern if the duration is sufficiently long and hence
the frequency resolution is sufficiently narrow.
1
ONE-SIDED, FULL-AMPLITUDE FOURIER TRANSFORM OF
Y(t) = 1.0 sin [ 2π ( 1 Hz ) t ] G, 0 ≤ t ≤ 20 sec
2.0
1.5
MAGNITUDE (G)
1.0
0.5
0
0 1 2 3 4 5 6 7 8 9 10
FREQUENCY (Hz)
Figure 1.

Y(t) = 1.0 sin [ 2π ( 1 Hz ) t ] G, 0 ≤ t ≤ 40 sec
2.0
1.5
MAGNITUDE (G)
1.0
0.5
0
0 1 2 3 4 5 6 7 8 9 10
FREQUENCY (Hz)
Figure 2.
2
White Noise Example
Consider the two white noise time histories in Figures 3 and 4. Each has a sample rate of
200 samples per second. Each has a standard deviation of 1 G. The overall level is 1
GRMS since the mean is zero, in each case.
The parameters for the Fourier transform calculation are given in Table 1.
Table 1. White Noise Parameters for Fourier Transform

Parameter Figure 3 Figure 4
Overall Level 1 GRMS 1 GRMS
Duration 5 sec 10 sec
∆f 0.2 Hz 0.1 Hz
Sample Rate 200 sps 200 sps
Frequency 0 to 100 Hz 0 to 100 Hz
Domain (Hz)
Number of 500 1000
Spectral Lines
sps = samples per second.
Recall that the frequency resolution ∆f is the inverse of the duration T.
∆f = 1 / T (1)
The frequency domain is taken from zero to one-half the sample rate.
The number of spectral lines N is equal to the maximum frequency divided by the
frequency resolution.
F max
N= (2)
∆f
The Fourier transforms of the respective white noise time histories are shown in Figures 5
and 6.
3
WHITE NOISE TIME HISTORY
10
5
ACCEL (G)
-5
-10
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 3.
WHITE NOISE TIME HISTORY
10
5
ACCEL (G)
-5
-10
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 4.
4
5 SECOND DURATION WHITE NOISE TIME HISTORY
0.30
Mean value of Fourier Transform magnitude = 0.056 G
0.25
MAGNITUDE (G)
0.20
0.15
0.10
0.05
0
0 10 20 30 40 50 60 70 80 90 100
FREQUENCY (Hz)
Figure 5.

10 SECOND DURATION WHITE NOISE TIME HISTORY
0.30
Mean value of Fourier Transform magnitude = 0.020 G
0.25
MAGNITUDE (G)
0.20
0.15
0.10
0.05
0
0 10 20 30 40 50 60 70 80 90 100
FREQUENCY (Hz)
Figure 6.
5
Ideally, the "white noise" would have a constant Fourier transform magnitude with
respect to frequency. The fact that there is some variation within Figures 5 and within
Figure 6 is unimportant for this example. The pertinent point is that the mean magnitude
decreases by about one-half, comparing the transform in Figure 6 to the transform in
Figure 5.
The reason for the decrease is that the transform in Figure 6 has 1000 spectral lines
compared to the 500 spectral lines in the Figure 5 transform. Thus, the "energy" is
divided into a greater number of spectral lines in the Figure 6 transform.
Each transform, however, yields the same overall value of 1 GRMS. This is found as
follows:
1. Divide each spectral magnitude by √2 to convert from peak to RMS.

2. Square each spectral RMS value to convert to mean square.
3. Sum the mean square values.
4. Take the square root of the sum.
Conclusion
For a random signal, the Fourier transform magnitude depends on the number of spectral
lines.
This drawback is overcome by the power spectral density function, which is covered in a
later unit.
Homework
1. Use the generate.exe program to synthesize white noise time histories similar to
those in Figures 3 and 4. Then calculate and plot the respective Fourier magnitudes
using the fourier.exe program.
2. Optional problem intended for avid Excel users. Call each Fourier transform
magnitude into Excel. Calculate the overall RMS value using the four steps shown
previously in the text. Verify that this agrees with the RMS value of the time history.
Program maxfind.exe can be used to calculate the RMS value of the time history.
For reference, a sample Excel spreadsheet for problem 2 is given in file: Fourier.xls.
Students who prefer to use Matlab or some other software tool are welcome to do so.
6
Unit 6C. Leakage Error in Fourier Transforms
By Tom Irvine
Introduction
There are a number of error sources associated with the Fourier transform. One error source is
called "leakage."
Leakage is a smearing of energy throughout the frequency domain.
Leakage results when both of the following conditions are present:
1. The signal is taken over a finite duration.
2. The signal is "non-periodic" in the time record.
Both these conditions are usually present in engineering data. Thus, leakage usually occurs.
For example, leakage occurs if a Fourier transform is calculated for a non-integral number of
sine function cycles.
Sine Function Example 1
Consider that a data acquisition system is used to monitor a continuous sine function. The sine
function has an amplitude of 1 G and a frequency of 1 Hz, as shown in Figure 1. The sample
rate is 32 samples per second.
Y(t) = 1.0 sin [ 2π t ] G
1
ACCEL (G)
-1
-2
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 1.
1
Now consider that the data acquisition system measures three cycles as shown in Figure 2. Note
that the time history amplitude is zero at the start and end of the record.
In essence, the Fourier transform will correctly assume that the original signal is a series of
three-cycle segments as shown in the time history in Figure 3.
Y(t) = 1.0 sin [ 2π t ] G
2
3 cycles
1
ACCEL (G)
-1
-2
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 2.
Y(t) = 1.0 sin [ 2π t ] G
2
Assumed input based on series of 3 cycles
1
ACCEL (G)
-1
-2
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 3.
The three-cycle sine function in Figure 2 is converted to a Fourier transform in Figure 4. As

expected, a spectral line of 1 G appears at 1 Hz. Note that ∆f = 0.333 Hz.
2
ONE-SIDED, FULL MAGNITUDE FOURIER TRANSFORM
Y(t) = 1.0 sin [ 2π t ] G, 3 < t < 6
2.0
( from 3 cycle time history ) Shown as Bar Graph
MAGNITUDE (G)
1.5
1.0
0.5
0
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 4.
Sine Function Example 2
Now assume that the data acquisition system has a limited memory buffer and is only able to
1
capture 2 cycles of the sine function, as shown in Figure 5.
2
1
In essence, the Fourier transform will assume that the original signal is a series of 2 cycle
2
segments as shown in Figure 6. Distortion is clearly visible in the time history in Figure 6.
Specifically, the input signal is not periodic in the time record.
1
The 2 cycle sine function in Figure 5 is converted to a Fourier transform in Figure 7. Note
2
that leakage occurs as shown by the smearing of energy across the frequency band.
A related problem is that ∆f = 0.4 Hz. Thus, there are spectral lines at the following frequencies
in Hz: 0, 0.4, 0.8, 1.2, …. There is no spectral line at 1 Hz, however, which is the frequency of
the sine function.
3
Y(t) = 1.0 sin [ 2π t ] G
2
2 1/2 cycles
1
ACCEL (G)
-1
-2
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 5.
Y(t) = 1.0 sin [ 2π t ] G
2
Assumed input based on series of 2 1/2 cycles.
1
ACCEL (G)
-1
-2
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 6.
4
ONE-SIDED, FULL MAGNITUDE FOURIER TRANSFORM
Y(t) = 1.0 sin [ 2π t ] G, 3.0 < t < 5.5
1.5
( from 2 1/2 cycle time history ) Bar graph.
MAGNITUDE (G)
Leakage error.
1.0
0.5
0
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
FREQUENCY (Hz)
Figure 7.
Homework
There are no homework problems associated with this unit. As an option, the student may try
experimenting with the generate.exe program and the Fourier.exe program to create examples of
leakage.
5
Unit 6D. Hanning Window
By Tom Irvine
Introduction
A Fourier transform may have a leakage error, as discussed in Unit 6C. The leakage
error can be reduced by subjecting the time history to a window, as discussed in
References 1 through 6.
Two common types of windows are the rectangular window and the Hanning window.
Rectangular Window
The rectangular, or flat, window leaves the time history data unmodified. Thus, a
rectangular window is equivalent to no window at all. A rectangular window is
appropriate for transient data or nonstationary data. Ideally, the time history includes
some data during the "quiet" periods both before and after the event. An example of a
transient event is shown in Figure 1.
A RECTANGULAR WINDOW WOULD BE

APPROPRIATE FOR THIS TRANSIENT EVENT.
30
20
10
ACCEL (G)
-10
-20
-30
0 1 2 3 4
TIME (SEC)
Figure 1.
1
Hanning Window
One of the most common windows is the Hanning window, or the cosine squared
window. It is appropriate for stationary vibration.
This window tapers the time history data so that the amplitude envelope decreases to zero
at both the beginning and end of the time segment. The Hanning window w(t) can be
defined as
 2 t 
1 − cos πT , 0 ≤ t ≤T
  
w (t) =  (1)
 0, elsewhere


Equation (1) is plotted in Figure 2.
2
HANNING WINDOW Y(t) = 1 - [ cos (π t/T) ]
1.1
1.0
0.9
0.8
0.7
0.6
Y(t)
0.5
0.4
0.3
0.2
0.1
0
0 T/2 T
TIME
Figure 2.
Furthermore, a normalization factor of 8 / 3 may be applied to the Hanned data to

compensate for the lost energy, from Reference 6.
2
Example
A 1 Hz sine function is shown in Figure 3. The same function is shown after a
normalized Hanning window is applied in Figure 4.
SINE FUNCTION: Y(t) = 1.0 sin (2π t) G, 9.5 cycles
1.0
0.5
ACCEL (G)
-0.5
-1.0
0 1 2 3 4 5 6 7 8 9
TIME (SEC)
Figure 3.
NORMALIZED HANNING WINDOW OF SINE FUNCTION:

Y(t) = 1.0 sin (2π t) G
1
ACCEL (G)
-1
-2
0 1 2 3 4 5 6 7 8 9
TIME (SEC)
Figure 4.
3
The Fourier transforms of two time histories are shown together in Figure 5.
ONE-SIDED, FULL AMPLITUDE, FOURIER TRANSFORM

OF SINE FUNCTION: Y(t) = 1.0 sin (2π t) G, ∆f=0.105 Hz
1.0
Rectangular Window
Hanning Window
0.8
MAGNITUDE (G)
0.6
0.4
0.2
0
0 0.5 1.0 1.5 2.0
FREQUENCY (Hz)
Figure 5.
Ideally, the Fourier transform would have a single, discrete line at 1 Hz with an
amplitude of 1 G.
Both the rectangular and Hanning Fourier transforms have some leakage error, however.
The rectangular window produces more leakage error than the Hanning window. Thus,
the Hanning window is recommended for stationary data.
Homework
1. This unit includes the fourier.exe program, version 1.3. This version has options for
both the rectangular and Hanning windows. Use this program to recreate the
examples shown in this unit.
References
1. C. Harris, editor; Shock and Vibration Handbook, 3rd edition; R. Randall, "Chapter 13
Vibration Measurement Equipment and Signal Analyzers," McGraw-Hill, New York, 1988.
2. MAC/RAN Applications Manual Revision 2, University Software Systems, Los Angeles,
CA, 1991.
4
3. Vibration Testing, Introduction to Vibration Testing, Section 9 (I), Scientific-Atlanta,
Spectral Dynamics Division, San Diego, CA, Rev 5/72.
4. R. Randall, Frequency Analysis, Bruel & Kjaer, Denmark, 1987.
5. TSL25, Time Series Language for 2500-Series Systems, GenRad, Santa Clara, CA, 1981.
6. F. Harris, Trigonometric Transforms, Scientific-Atlanta, Spectral Dynamics Division,
Technical Publication DSP-005 (8-81), San Diego, CA.
5
Unit 6E. Fast Fourier Transform (FFT)
By Tom Irvine
Introduction
The discrete Fourier transform requires a tremendous amount of calculations. A time
2
history with M coordinates would require M complex multiplication steps.
The discrete Fourier transform can be carried out by a Fast Fourier transform method,
N
however. The method is based on a time series with a number of points equal to 2 ,
where N is an integer.
N
The FFT requires M log 2 M complex multiplication steps, where M = 2 .
The details of the FFT algorithm are given in Reference 1.
Example
Now consider a time history with 1,000,000 points. A regular Fourier transform would
12
require 10 complex multiplication steps. On the other hand, an FFT would only require
7
approximately 2(10 ) steps. Thus, the FFT achieves the calculation in 1/50,000th of the
time.
Limitation of the FFT

N
The above example is not quite correct. Again, the FFT is based on a time series with 2
coordinates. Note that
19
2 = 524,288
and
20
2 = 1,048,576
Unfortunately, a time history with 1,000,000 points falls between these two cases.
There are two options for dealing with a time history that is not an integer power of 2.
One option is to truncate the time history. This should be acceptable if the data is
stationary. In the above example, the time history would thus be truncated to 524,288
points.
1
The second option is to pad the time history with trailing zeroes to bring its length to an
integer power of 2. A problem with this option is that it artificially reduces the amplitude
of the Fourier transform spectral lines.
Truncation, rather than zero-padding, is the preferred method in this course.
Homework
1. Plot the accelerometer time history in file panel.txt. The file has two columns:
time(sec) and accel(G). The data was measured on the front panel of a big-rig trailer,
as it was driven over a test course. The data has 8192 points, which is conveniently
an integer power of 2. In many cases, data acquisition systems are set-up to measure
data segments which are an integer power of 2.
2. Calculate the Fourier transform of panel.txt using program fourier.exe. Use the
Hanning window. Plot the full.out magnitude file.
3. Calculate the FFT of panel.txt using program FFT.exe. Use the mean removal and
Hanning window options. Plot the fft_full.out magnitude file.
4. Compare the results of steps 2 and 3. Are the results similar or different?
5. Aside from the interactive data entry, what was the difference in processing time
between the fourier.exe program and the FFT.exe program? You may give a
qualitative answer.
6. What were the two dominant frequencies in the Fourier transform magnitude?
Reference
1. T. Irvine, The Fast Fourier Transform (FFT), Vibrationdata Publications, 1998.
2
Unit 6F. Inverse Fourier Fast Transform
By Tom Irvine
Introduction
Recall that the Fourier transform Fk for a discrete time series x n can be expressed as
N− 1
 2π 
∑ x n exp−
1
Fk = j nk , for k = 0, 1, ..., N − 1 (1)
N N 
n =0
N− 1
 2π 
xn = ∑ F k exp+ j
N
nk , for n = 0, 1, ..., N − 1

(2)
k =0
Note that Fk has dimensions of [amplitude].
Points
Here are some important points about the Fourier transform and its inverse:
1. The Fourier transform converts a time history to the frequency domain. The inverse
Fourier transform converts the frequency domain function back to a time history.
2. In some cases, an intermediate calculation may be performed on the Fourier
transform prior to taking its inverse. This calculation might involve a transfer
function. This will be covered in future units.
3. The Fourier transform and its inverse must be a matched pair. This is an absolute
requirement since there are many different types of Fourier transforms.
4. The main difference between the two transforms is the polarity of the argument in the
exponential function. In addition, the Fourier transform has a scale factor of 1/N.
5. A measured time history consists only of a real amplitude. The Fourier transform
converts this to a complex function. The inverse Fourier transform converts this
complex function to a complex time history, but the resulting imaginary component
should be zero.
6. An exception to point 5 could occur if some intermediate calculation were performed
prior to taking the inverse Fourier transform.
1
7. Theoretically, the input time history could be complex with a non-zero imaginary
component. For practical purposes, this never occurs.
8. A Hanning window may be applied in the Fourier transform, but it is never applied to
the inverse calculation.
9. An inverse Fourier transform can be performed as an "inverse Fast Fourier transform"
if the number of points is an integer power of 2.
Homework
1. Plot the time history file panel.txt from Unit6E.
2. Take the FFT of panel.txt. Use program fft.exe with a rectangular window.
3. Take the inverse FFT using program invfft.exe. The input file is fft.out from step 2.
4. Program invfft.exe will generate two files: real.out and complex.out. Compare
real.out with panel.txt.
5. Verify that file complex.out has an approximately zero imaginary component.
2
Unit 7A. Power Spectral Density Function
By Tom Irvine
Introduction
A Fourier transform by itself is a poor format for representing random vibration because
the Fourier magnitude depends on the number of spectral lines, as shown in previous
units.
The power spectral density function, which can be calculated from a Fourier transform,
overcomes this limitation. Again, some assembly is required.
Note that the power spectral density function represents the magnitude, but it discards the
2
phase angle. The magnitude is typically represented as G /Hz. The G is actually GRMS.
Calculation Method
Power spectral density functions may be calculated via three methods:
1. Measuring the RMS value of the amplitude in successive frequency bands, where the
signal in each band has been bandpass filtered.
2. Taking the Fourier transform of the autocorrelation function. This is the Wierner-
Khintchine approach.
3. Taking the limit of the Fourier transform X(f) times its complex conjugate divided by
its period T as the period approaches infinity. Symbolically, the power spectral
density function XPSD(f) is
lim X (f )X * (f )
X PSD (f ) = (1)
T→ ∞ T
These methods are summarized in Reference 1. Only the third method is considered in
this unit.
Fourier Transform Method

Equation (1) assumes that the Fourier transform has a dimension of [amplitude-time].
The following equations are taken from Reference 2.
1
The discrete Fourier transform [amplitude-time] is
N− 1
2π
X(k ) = ∆ t ∑ x(n) exp(− j N nk ) for k = 0, 1, ..., N − 1 (2a)
n=0
Note that the index k can be related to the frequency
frequency (k) = k ∆f (2b)
The inverse transform is
N− 1
2π
x ( n) = ∆ f ∑ X(k ) exp(+ j N nk) for n = 0, 1, ..., N − 1 (3)
k =0
These equations give the Fourier transform values X(k) at the N discrete frequencies
k∆f and give the time series x(n) at the N discrete time points n ∆t. The total period of
the signal is thus
T = N∆t (4)
where
N is number of samples in the time function and in the Fourier transform
T is the record length of the time function
∆t is the time sample separation
Consider a sine wave with a frequency such that one period is equal to the record length.
This frequency is thus the smallest sine wave frequency which can be resolved. This
frequency ∆f is the inverse of the record length.
∆f = 1/T (5)
This frequency is also the frequency increment for the Fourier transform.
Alternate Fourier Transform Method

The Fourier transform with dimension of [amplitude-time] is rather awkward.
Fortunately, the power spectral density can be calculated from a Fourier transform with
dimension of [amplitude]. The corresponding formula is
lim F(f )F * (f )
X PSD (f ) = (6)
∆f → 0 ∆f
2
The Fourier transform F(k ) for the discrete time series x (n) is
N− 1
 2π 
∑ x (n) exp−
1
F(k ) = j nk , for k = 0, 1,..., N − 1 (7a)
N N 
n =0
Note that the index k can be related to the frequency
frequency (k) = k ∆f (7b)
N− 1
 2π 
x ( n) = ∑ F(k) exp+ j nk , for n = 0, 1, ..., N − 1
N 
(8)
k =0
One-sided Fourier Transform Approach

The power spectral density functions in equations (1) and (6) were both double-sided.
The power spectral density amplitude would be symmetric about the Nyquist frequency.
A one-sided, or single-sided, power spectral density function is desired.
Let X̂ PSD (f ) be the one-sided power spectral density function.
lim G (f )G * (f )
X̂ PSD (f ) = (9)
∆f → 0 ∆f
3
The one-sided Fourier transform G(k) is
  N− 1 
 1  


magnitude 

N 
 
∑ {x(n)}

for k = 0
  n =0 
G(k) = 
  N− 1 
 1    2π  
 2 magnitude
∑
N
  N  x (n) exp− j nk   for k = 1, ..., − 1
     N   2
  n =0 
with
N as an even integer
Frequency (k) = k∆f
(10)
Implementation
Calculation of a power spectral density requires that the user select the ∆f value from a
list of options. The ∆f value is linked to the number of degrees of freedom.
Statistical degrees of freedom

The reliability of the power spectral density data is proportional to the degrees of
freedom.
The statistical degree of freedom parameter is defined from References 3 and 4 as
follows:
dof = 2BT (11)
where dof is the number of statistical degrees of freedom and B is the bandwidth of an
ideal rectangular filter. This filter is equivalent to taking the time signal “as is,” with no
tapering applied. Note that the bandwidth B equals ∆f, again assuming an ideal
rectangular filter.
The 2 coefficient in equation (11) results from the fact that a single-sided power spectral
density is calculated from a double-sided Fourier transform. The symmetries of the
Fourier transform allow this double-sided to single-sided conversion.
For a single time history record, the period is T and the bandwidth B is the reciprocal so
that the BT product is unity, which is equal to 2 statistical degrees of freedom from the
definition in equation (11).
A given time history is thus worth 2 degrees of freedoms, which is poor accuracy per
Chi-Square theory, as well as per experimental data per Reference 3. Note that the Chi-
Square theory is discussed in Reference 5.
4
Breakthrough
The breakthrough is that a given time history record can be subdivided into small records,
each yielding 2 degrees of freedom, as discussed in Reference 4 for example. The total
degrees of freedom value is then equal to twice the number of individual records. The
penalty, however, is that the frequency resolution widens as the record is subdivided.
Narrow peaks could thus become smeared as the resolution is widened.
An example of this subdivision process is shown in Table 1. The process is summarized
in equations (12) through (16).
Table 1. Example: 4096 samples taken over 16 seconds, rectangular filter.

Number of Number of Period of Frequency dof Total dof
Records Time Each Resolution per
Samples per Record Ti Bi=1/Ti Record
NR Record (sec) (Hz) =2Bi TI
1 4096 16. 0.0625 2 2
2 2048 8. 0.125 2 4
4 1024 4. 0.25 2 6
8 512 2. 0.5 2 16
16 256 1. 1. 2 32
32 128 .5 2. 2 64
64 64 .25 4. 2 128
Notes:
1. The subscript “i” is used to denote “individual” in Table 1.
2. The rows in the table could be continued until a single sample per record remained.
Also note that:
Total dof = 2 NR (12)

NR = T / Ti (13)
Bi = 1 / Ti (14)
NR = Bi T (15)
Total dof = 2 Bi T (16)
Window
A window is typically applied to each time segment during the power spectral density
calculation, as discussed in References 3, 4, and 6. The purpose of the window is to
5
reduce a type of error called leakage. One of the most common windows is the Hanning
window, or the cosine squared window. This window tapers the data so that the
amplitude envelope decreases to zero at both the beginning and end of the time segment.
The Hanning window w(t) can be defined as
 2 t 
 1 − cos πT  , 0 ≤ t ≤T
  
w (t) =  (21)
 0, elsewhere


The window operation reduces the leakage error but also has the effect of reducing the
statistical degrees-of-freedom.
Also, a normalization factor of 8 / 3 is applied to the Hanned data to compensate for

the lost energy, from Reference 7.
Overlap
The lost degrees-of-freedom can be recovered by overlapping the time segments, each of
which is subjected to a Hanning window. Nearly 90% of the degrees-of-freedom are
recovered with a 50% overlap, according to Reference 3.
The concept of windows and overlapping is represented in Figure 1.
6
Figure 1.
7
Fast Fourier Transform
Three variations of the discrete Fourier transform have been given in this report. The
solution to any of these transforms requires a great deal of processing steps for a given
time history. Fast Fourier transform methods have been developed, however, to greatly
reduce the required steps. These methods typically require that the number of time
N
history data points be equal to 2 , where N is some integer. The derivation method is
via a butterfly algorithm, as shown, for example, in Reference 8.
Records with sample numbers which are not equal to an integer power of 2 can still be
processed via the fast Fourier transform method. Such a record must either be truncated
or padded with zeroes so that its length becomes an integer power of 2.
Summary
Time history data is subdivided into segments to increase the statistical-degrees-of-
freedom by broadening the frequency bandwidth. Next, a window is applied to each
segment to taper the ends of the data. Finally, overlapping is used to recover degrees-of-
freedom lost during the window operations. The effect of these steps is to increase the
accuracy of the power spectral density data. Nevertheless, there are some tradeoffs as
shown in the following examples.
Homework
1. Use program generate.exe to synthesize a white noise time history with 1 G standard
deviation, 10 second duration, and 1000 samples per second.
2. Use program poweri.exe to calculate the power spectral density. Choose 256 samples
per second, which corresponds to 78 dof and ∆f = 3.9 Hz. Select the mean removal
and Hanning window options. Plot the output file a.out, preferably in log-log format.
3. Repeat step 3 for 128 samples per second, which corresponds to 156 dof and ∆f = 7.8
Hz.
4. Compare the power spectral density curves from steps 2 and 3. Do the curves have a
similar or different amplitude?
References
1. W. Thomson, Theory of Vibration with Applications, Second Edition, Prentice-Hall,
New Jersey, 1981.
2. C. Harris, editor; Shock and Vibration Handbook, 3rd edition; R. Randall, "Chapter
13 Vibration Measurement Equipment and Signal Analyzers," McGraw-Hill, New
York, 1988.
8
3. MAC/RAN Applications Manual Revision 2, University Software Systems, Los
Angeles, CA, 1991.
4. Vibration Testing, Introduction to Vibration Testing, Section 9 (I), Scientific-Atlanta,
Spectral Dynamics Division, San Diego, CA, Rev 5/72.
5. Walpole and Myers, Probability and Statistics for Engineers and Scientists,
Macmillan, New York, 1978.
6. R. Randall, Frequency Analysis, Bruel & Kjaer, Denmark, 1987.
7. TSL25, Time Series Language for 2500-Series Systems, GenRad, Santa Clara, CA,
1981.
8. F. Harris, Trigonometric Transforms, Scientific-Atlanta, Spectral Dynamics Division,
Technical Publication DSP-005 (8-81), San Diego, CA.
9
Unit 7B. Overall GRMS Value of Power Spectral Density Function
By Tom Irvine
_____________________________________________________________________
Introduction
Random vibration may be represented in the frequency domain by a power spectral density
function. The power spectral density amplitude is represented in units of (G 2/Hz). This is
an abbreviated notation. The actual unit is (GRMS 2/Hz).
GRMS from Power Spectral Density

The overall GRMS value is equal to the square root of the area under the power spectral
density curve. The power spectral density function can be integrated using the trapezoidal
rule. This approach is suitable if the power spectral density has a constant ∆f frequency
resolution.
GRMS from Time History

Recall that the GRMS value may also be calculated directly from the time domain using
equation (1).
N
∑ xi2
1
G RMS = (1)
N
i =1
Sample Time History

A flight accelerometer time history is shown in Figure 1. The data is from a suborbital
rocket vehicle. The source energy was aerodynamic buffeting during powered flight. The
overall level is 2.95 GRMS, as calculated using the maxfind.exe program.
Corresponding Power Spectral Density

The power spectral density is shown in Figure 2, as calculated using the poweri.exe
program. The vibration is clearly random, although it is not white noise. The square root
of the area under the curve is 2.94 GRMS, as calculated using the psdtrap.exe program.
1
FLIGHT ACCELEROMETER DATA
50
Overall Level = 2.95 GRMS
40
30
20
ACCEL (G)
10
0
-10
-20
-30
-40
-50
52 53 54 55 56
TIME (SEC)
Figure 1.
POWER SPECTRAL DENSITY FLIGHT ACCELEROMETER DATA

-1
10
-2
10
-3
ACCEL (G /Hz)
10
2
-4
10
-5
10
-6
10
-7
10
10 100 1000 2000
FREQUENCY (Hz)
Figure 2.
2
Comparison
Thus, the GRMS value from the time history equals the GRMS value from the power
spectral density, ignoring the small numerical error.
Homework
Repeat the examples in this unit using data file: flight2.txt. This is actual flight
accelerometer data.
Use a frequency resolution of ∆f = 10 Hz. The data is somewhat nonstationary, but this is
not a concern for this exercise.
What are the GRMS values of the time history and power spectral density?
3
Unit 7C. Overall GRMS Value of a PSD Specification
By Tom Irvine
_____________________________________________________________________
Introduction
Certain components must be designed and tested to withstand random vibration
environments. Typically, the test specification is in the form of a power spectral density
function.
A power spectral density specification is usually represented as follows:
1. The specification is represented as a series of piecewise continuous
segments.
2. Each segment is a straight line on a log-log plot.
An example is shown in Figure 1.
POWER SPECTRAL DENSITY
0.1
ACCELERATION (G /Hz)
2
0.01
0.001
10 100 1000 2000
FREQUENCY (Hz)
Figure 1.
1
The specifications are almost always given in the form of an acceleration power spectral
density. The main reason is that acceleration is easier to measure than velocity or
displacement, in a vibration sense.
2
Note that the power spectral density amplitude is represented in units of (G /Hz). This is
2
an abbreviated notation. The actual unit is (GRMS /Hz).
Derivation
The goal is to calculate the overall GRMS value. Use of the trapezoidal rule would
require tedious interpolation. Instead, a direct integration approach is available. Special
rules must be followed due to the log-log format.
The equation for each segment is
y 
y( f ) =  1  f
n (1)
 f1n 
The starting coordinate is ( f 1, y 1 ) .
The exponent n is a real number which represents the slope. The slope between two
coordinates ( f 1, y 1 ) and ( f 2 , y 2 ) is
y 
log 2 
 y1 
n= (2)
f 
log 2 
 f1 
The area a 1 under segment 1 is
f 2  y1  n
a 1= ∫
f1
 n  f df (3)
 f 1 
There are two cases depending on the exponent n.
The first case is
 y1   1  n+ 1 f 2
a1 =   f , for n ≠ − 1 (4)
n 
 f 1   n + 1 f1
2
y 
a1 =  1 
n
 f 1 
 1  n+ 1
[
 n + 1 f 2 ]
− f 1n + 1 , for n ≠ − 1 (5)
The second case is
f 2  y1  − 1
a1 = ∫ f1
 − 1  f df , for n = − 1 (6)
 f 1 
∫f 1 [y 1f1 ] f
f2 df
a1 = , for n = − 1 (7)
[ ] ( )f
f2
a 1 = y 1f 1 ln f , for n = − 1 (8)
1
[ ]
a 1 = [y 1 f1 ] ln( f 2 ) − ln( f 1 ) , for n = − 1 (9)
  f 
a 1 = [y 1 f1 ] ln 2   , for n = − 1 (10)
  f 1 
In summary, the area under segment i is

[ ]
 yi  1 
  n   f i + 1n + 1 − f i n + 1 , for n ≠ − 1
  f i   n + 1

a i=  (11)
  f 
 [y i f i ] lni + 1   , for n = − 1

   f i  
The overall level L is
m
L = ∑ ai (12)
i= 1
where m is the total number of segments.
3
Example
Consider the power spectral density function in Figure 1. The breakpoints are given in
Table 1.
Table 1.
Power Spectral Density
Freq Level
(Hz) (G 2/Hz)
10 0.002
100 0.04
1000 0.04
2000 0.02
Consider the first pair of coordinates:

2
f 1 = 10 Hz y 1 = 0.002 G /Hz
2
f 2 = 100 Hz y 2 = 0.04 G /Hz
Calculate the slope.
 0.04 
log 
 0.002 
n= (13)
 100 
log 
 10 
n = 1.3 (14)
Substitute into equation (11).
 0.002   1 
a1 =   13
 10   . + 1
1.3  [
. + 1 − 1013
10013 . +1
] (15)
 0.002 
a1 =  
 101.3 
1 
[ 2.3
 2.3  100 − 10
2.3
] (16)
a 1 = 1726
. G2 (17)
4
Consider the second pair:
2
f 2 = 100 Hz y 2 = 0.04 G /Hz
2
f 3 = 1000 Hz y 3 = 0.04 G /Hz
 0.04 
log 
 0.04 
n= (18)
 1000 
log 
 100 
n = 0. (19)
 0.04 
a2 =  
100 0 
 1 
[
 0 + 1 1000
0 + 1 − 100 0 + 1
] (20)
 0.04  1  1 1
a2 =  1 1000 − 100  (21)
 1 
a 2 = 36.000 G 2 (22)
Consider the third pair:

2
f 3 = 1000 Hz y 3 = 0.04 G /Hz
2
f 4 = 2000 Hz y 4 = 0.02 G /Hz
 0.02 
log 
 0.04 
n= (23)
 2000 
log 
 1000 
n = -1. (24)
5
  2000  
a 3 = [(0.04)(1000) ] ln  (25)
 1000  
a 3 = 27.726 (26)
Now substitute the individual area values into equation (12).
L= ( 1.726 + 36.000 + 27.726) G 2 (27)
The overall level is
L = 8.09 G RMS (28)
Homework
1. Determine the overall level of the power spectral density in Table 2. Use hand
calculations.
Table 2.
Freq Level
(Hz) (G 2/Hz)
10 0.001
200 0.08
500 0.08
2000 0.02
2. Verify the overall GRMS value for Table 2 using program psdint.exe.
3. Why do random vibration specifications typically begin at a frequency > 10 Hz?
6
Unit 7D. Level Difference in Decibels
By Tom Irvine
_____________________________________________________________________
Introduction
Power density levels are often scaled upward or downward in terms of decibels (dB).
G or GRMS
Assume that A and B each has an amplitude either in G or GRMS. The difference in dB
between A and B is
A 
∆dB = 20 log   (1)
B 
Note that the log function is base ten.
[ ]
Furthermore,
A = B 10 ∆dB / 20 (2)
2 2
G or G /Hz
2 2
Assume that C and D each has an amplitude either in G or G /Hz. The difference in dB
between C and D is
C 
∆dB = 10 log   (3)
D 
[ ]
Furthermore,
C = D 10 ∆dB / 10 (4)
Example
The level in Table 1 is to be raised by 6 dB. Calculate the new level both in terms of
2
GRMS and G /Hz.
1
Table 1.
Original Power
Spectral Density,
8.09 GRMS Overall
Freq Level
(Hz) (G 2/Hz)
10 0.002
100 0.04
1000 0.04
2000 0.02
A 6 dB increase causes the overall GRMS level to increase by a factor of 2.0, per
2
equation (2). This same increase causes the G /Hz amplitudes to increase by a factor of
4.0, per equation (4). The resulting level is shown in Table 2.
Table 2.
New Power
Spectral Density,
16.2 GRMS Overall
Freq Level
(Hz) (G 2/Hz)
10 0.008
100 0.16
1000 0.16
2000 0.08
Homework
1. Calculate a new level for Table 3 based on a 3 dB increase.
Table 3.
Power Spectral Density,
9.3 GRMS Overall
Freq Level
(Hz) (G 2/Hz)
10 0.001
200 0.08
500 0.08
2000 0.02
2
2. What is the dB difference between 6 GRMS and 10.5 GRMS? Use hand calculations.
Then verify your answer using program dboct.exe. (Use the coordinates at same
frequency option).
3
Unit 7E. dB/octave Slopes
By Tom Irvine
_____________________________________________________________________
Introduction
NAVMAT P-9492 gives the power spectral density specification shown in Figure 1.
0.1
Overall Level = 6.0 grms
2
0.04 g / Hz
+3 dB / octave -3 dB / octave
PSD ( g / Hz )
2
0.01
0.001
20 80 350 2000
FREQUENCY (Hz)
Figure 1.
The task is to determine the coordinates of the endpoints.
Derivation
2
Assume that a1 and a 2 each has an amplitude in G /Hz. The difference in dB between
a1 and a 2 is
a 
∆dB = 10 log  2  (1)
a 1 
1
Furthermore,
a 2 = a1  10 ∆dB / 10  (2)

 

Additional equations are needed.
The slope N between two coordinates (f1, a1 )and (f 2 , a 2 )in a log-log plot is
a 
log  2 
N=  a1  (3)
f 
log  2 
f1 
Solve for a 2 .
f  a 
N log  2  = log  2  (4)
f1   a1 
 N
f 2   a 
log    = log  2  (5)
f1    a1 
 
Take the anti-log.

N
f 2  a 2 
  =  (6)
f1   a1 
N
a 2  f 2 
 =   (7)
a1  f1 
Thus,
N
f 2 
a 2 = a1   (8)
f1 
2
Now consider a one-octave frequency separation.
f 2 = 2 f1 (9)
Substitute equation (9) into (3).
a 
log  2 
N=  a1  (10)
log[]2
Substitute equation (1) into (10).
∆dB / 10
N= (11)
log[]
2
Note that ∆dB represents the dB/octave slope in equation (11). Again, equations (10) and
(11) assume a one-octave frequency separation.
Now substitute equation (11) into (8).
∆dB / 10 
f  
a 2 = a1  2  log[]
2  (12)
f1 
Example
Calculate the amplitude at 2000 Hz for the power spectral density in Figure 1.
Note
f1 = 350 Hz
f 2 = 2000 Hz
a1 = 0.04 G 2 / Hz
− 3 dB / 10 
 

a 2 = 0.04 G 2 / Hz 
2000 Hz  log[]2 
 (13)
 350 Hz 
3
a 2 = 0.007 G 2 / Hz (14)
Homework
1. Calculate the amplitude at 20 Hz for the power spectral density curve in Figure 1.
2. Calculate the overall GRMS value for the power spectral density curve in Figure 1.
Use program psdint.exe. Give the answer to two decimal places.
4
Unit 8. Transmissibility Function for Acceleration
By Tom Irvine
_____________________________________________________________________
Introduction
Certain systems are subjected to a base excitation vibration. Examples include:
1. A building during an earthquake
2. An automobile traveling down a washboard road
3. An avionics component on a rocket vehicle bulkhead during powered flight
The purpose of this unit is to determine the steady state response of a single-degree-of-
freedom system to sinusoidal base excitation. The transmissibility function is the ratio of
the response to the input.
Model
Consider the single-degree-of-freedom system subjected to base excitation shown in
Figure 1. The free-body diagram is shown in Figure 2.
&x&
m
k c &y&
Figure 1. Single-degree-of-freedom System
The variables are

m = mass,
c = viscous damping coefficient,
k = stiffness,
x = absolute displacement of the mass,
y = base input displacement.
The double-dot notation indicates acceleration
&x&
m
k(y-x) c (y& − x&)
Figure 2. Free-body Diagram
1
Summation of forces in the vertical direction
∑ F= &&
mx (1)
&& = c ( y& − x&) + k ( y − x)

mx (2)
Let z = x - y. The variable z is thus the relative displacement.
Substituting the relative displacement into equation (2) yields
m(&z&+ &y&) = − cz& − kz (3)
m&z&+ cz& + kz = − m&y& (4)
Dividing through by mass yields
&z&+ (c / m)z& + (k / m)z = − &y& (5)
By convention,
(c / m) = 2ξω n (6)
( k / m) = ω n 2 (7)
where ω n is the natural frequency in (radians/sec), and ξ is the damping ratio.
Substituting the convention terms into equation (5),
&z&+ 2ξω n z& + ω n 2 z = − &y& (8)
Either Laplace or Fourier transforms may be used to derive the steady state
transmissibility function for the absolute response acceleration, as shown in Reference 1.
After many steps, the resulting magnitude function is
&x& 1 + (2ξρ)2
=
( )
,
&y& 2
1 − ρ2 + (2ξρ)2
(9)
ρ = f / fn
where f is the base excitation frequency and fn is the natural frequency.
2
Recall that the damping is often represented in terms of the quality factor Q.
1
Q= (10)
2ξ
The transmissibility function is plotted for several Q values in Figure 3.
TRANSMISSIBILTY MAGNITUDE
SDOF SYSTEM SUBJECTED TO BASE EXCITATION
20
Q=1
Q=2
10 Q=10
TRANSMISSIBILITY ( G out / G in )
0.1
0.1 0.5 1 2 10
FREQUENCY RATIO ( f / fn )
Figure 3.
Note that f is the base excitation frequency in Figure 3.

The transmissibility curves in Figure 3 have several important features:
1. The response amplitude is independent of Q for f << fn.
2. The response is approximately equal to the input for f << fn.
3. Resonance occurs when f ≅ fn.
4. The peak transmissibility is approximately equal to Q for f = fn and Q > 2.
5. The transmissibility ratio is 1.0 for f = 2 fn regardless of Q.
6. Isolation is achieved for f >> fn.
3
The curves in Figure 3 are particularly useful for designing isolation systems.
Example 1
As a review, the natural frequency of a single-degree-of-freedom system is
1 k
fn = (11)
2π m
Now consider an avionics component which has a natural frequency fn = 200 Hz and an
amplification factor Q = 10.
The component is hard-mounted to a bulkhead in a rocket vehicle. Assume that the
component will be subjected to a 200 Hz sinusoidal oscillation with a base input
amplitude of 10 G during powered flight. The Q = 10 curve in Figure 3 shows that the
response will be 100 G, which is severe. How can the response be reduced?
The Q = 10 curve in Figure 3 shows that the response can be reduced to approximately
3.2 G if f = 2 fn. The excitation frequency f is fixed, however. Thus, fn must be reduced.
Recall that fn = 200 Hz. Thus, change the mounting design so that fn=100 Hz. This can
be achieved by mounting the avionics component with the appropriate isolator grommets.
The grommets will act as a spring in series with the component. The grommets thus
reduce the natural frequency by reducing the overall stiffness.
In reality, the grommets will decrease the Q value, thus changing the calculation
somewhat. Nevertheless, the main effect is the natural frequency reduction.
Example 2
Grommets are typically made from some rubber or plastic material. They are effective
when they break "metal-to-metal contact" between the component and the mounting
surface.
Suppose isolation grommets cannot be used in the previous example. There could be
several reasons. One might be that the component must be hard-mounted because the
mounting surface serves as a thermal ground plane. How can the response be reduced in
this case?
The answer is to use the opposite approach as was used in Example 1. In this case, the
goal should be f = 0.5 fn. The response would thus be reduced to about 1.3 G. The
natural frequency goal would thus be 400 Hz, since the base excitation frequency is fixed
at 200 Hz. The frequency increase could be achieved by increasing the stiffness or by
decreasing the mass. In most cases, increasing the stiffness would be the practical
choice.
4
Further Notes
An alternative for both examples would be to decrease the Q value while leaving the
natural frequency fixed. Thus, the damping ratio would increase per equation (10).
Nevertheless, the stiffness is usually the easiest parameter to modify for practical design
purposes.
Homework
1. Consider a system with a natural frequency fn = 100 Hz and amplification factor
Q=10. The system is subjected to a sinusoidal base input with a variable frequency.
The input amplitude is 1 G. Use program steady.exe to complete the following table.
Excitation Response
Frequency (G)
(Hz)
50
60
70
80
90
95
99
100
101
105
110
120
130
140
150
160
180
200
2. Plot the Response (G) versus Excitation Frequency (Hz).
3. How does damping effect the response of a system subjected to base excitation?
4. Optional. Procure a slinky. Hold the top of the slinky as shown in Figure 4.
5
Figure 4.
Very, very slowly move your hand up-and-down in a sinusoidal manner. How does the
free end of the slinky respond?
Now find the natural frequency of the slinky by trial-and-error experiment. Do this in a
qualitative sense. Excite the natural frequency by moving your hand at the natural
frequency but with a small amplitude. How does the free end of the slinky respond?
Now excite the slinky at a frequency much higher than its natural frequency. Well, the
slinky may undergo some chaotic motion, but how would it respond if it were an ideal
single-degree-of-freedom system?
References
1. T. Irvine, An Introduction to the Vibration Response Spectrum, Vibrationdata.com
Publications, 1999.
6
Unit 10. Vibration response spectrum - Miles rule
By Tom Irvine
_____________________________________________________________________
Introduction
Again, certain systems are subjected to a base excitation vibration. Examples include:
1. A building during an earthquake
2. An automobile traveling down a washboard road
3. An avionics component on a rocket vehicle bulkhead during powered flight
The purpose of this unit is to determine the response of a single-degree-of-freedom system to
random base excitation.
Model
Consider the single-degree-of-freedom system subjected to base excitation shown in Figure 1.
The free-body diagram is shown in Figure 2.
&x&
m
k c
&y&
The variables are

m = mass,
k = stiffness,
&x&
m
k(y-x) c ( y& − x&)
1
The following equation of motion for the relative displacement z was derived in Unit 8.
&z&+ 2ξω n z& + ω n 2 z = − &y& (1)
Either Laplace or Fourier transforms may be used to derive the steady state transmissibility
function for the absolute response acceleration.
Base Input
A sample base input power spectral density is shown in Figure 3.
0.1
ACCELERATION (G /Hz)
2
0.01
0.001
10 100 1000 2000
FREQUENCY (Hz)
Figure 3
2
The breakpoints for the level in Figure are given in Table 1.
Table 1.
Power Spectral Density,
8.1 GRMS Overall Level
Freq Level
(Hz) (G 2/Hz)
10 0.002
100 0.04
1000 0.04
2000 0.02
The level in Figure 3 could represent an envelope of flight accelerometer data. A safety margin
might be included in this level.
Assume that this level is the random vibration test level for an avionics component. The
component will be tested on a shaker table. The component will be powered and monitored
during the vibration test.
The next task is to determine the response of the component to the base input level.
Response
The Miles equation is a simplified method of calculating the response of a single-degree-of-
freedom system to a random vibration base input, where the input is in the form of a power
spectral density.
The Miles equation is
π 1
&x&GRMS = P fn   (2)
2 2 ξ 
where
&x&GRMS = the overall response,

P = the power spectral density level at the natural frequency,
f n = the natural frequency.
Note that the damping is often represented in terms of the quality factor Q.
1
Q= (3)
2ξ
3
Miles equation can thus be expressed as
π
&x&GRMS = P fn Q (4)
2
This equation is given in Reference 1.

Miles equation must be calculated for each natural frequency system of interest. The damping
term is typically held as a constant.
Furthermore, equation (2) is an approximate formula which assumes a flat power spectral density
from zero to infinity Hz. As a rule-of-thumb, it may be used if the power spectral density is flat
over at least two octaves centered at the natural frequency.
Response for Known Natural Frequency

A certain avionics component has a circuit board which can be idealized as a single-degree-of-
freedom system. The natural frequency is fn = 200 Hz. The damping ratio is 0.05, equivalent to
Q=10.
What is the response of the component to the base input shown in Figure 3?
Use Miles equation. Note that the base input level is 0.04 G^2/Hz at 200 Hz
π G2 
&x&GRMS = 0.04 [ ][ ]
 200 Hz 10 (5)

2 Hz 

&x&GRMS = 11.2 GRMS (6)
Note that the RMS value is equal to the 1σ value assuming a zero mean. Recall that the 1σ value
is the standard deviation.
What is the peak response value?
The precise answer is unknown. It depends on the duration and on other factors. Often, the peak
response is assumed to be 3σ.
&x&peak = 33.6 G (3σ) (7)
The response time history has the following probability characteristics assuming a normal
distribution.
Probability inside ±3σ Limits = 99.73%
Probability outside ±3σ Limits = 0.27%
4
Vibration Response Spectrum
Now consider that the natural frequency of the avionics component is unknown, but assume an
amplification factor of Q=10. The base input is the level in Figure 3.
How can Miles equation be used for an unknown or variable natural frequency?
First, choose a sample of frequencies.
Next, determine the base input power spectral density at each frequency. This requires
interpolation at certain frequencies. The interpolation can be done by hand calculations. A more
efficient method is to use the dboct.exe program.
Finally, calculate the response at each frequency.
The result of these steps is shown in Table 2. The response level was calculated via an Excel
spreadsheet.
Table 2.
Vibration Response Spectrum, Q=10
Natural Base Input Response

Frequency Level Level
(Hz) (G 2/Hz) (GRMS)
10 0.002 0.56
20 0.005 1.25
50 0.016 3.54
100 0.04 7.93
200 0.04 11.21
400 0.04 15.85
600 0.04 19.42
800 0.04 22.42
1000 0.04 25.07
1200 0.033 24.94
1600 0.025 25.07
2000 0.02 25.07
The vibration response spectrum is the response level versus natural frequency, as shown in
Figure 4.
Note that the analysis in Table 2 violates the assumption that the input level is flat within one
octave on either side of the natural frequency. Some error may occur as a result. An improved
method which overcomes this assumption will be discussed in Unit 11.
5
VIBRATION RESPONSE SPECTRUM, Q=10
100
ACCELERATION (GRMS)
10
0.1
10 100 1000 2000
NATURAL FREQUENCY (Hz)
Figure 4.
The vibration response spectrum shows that reducing the natural frequency reduces the response
level, with respect to the given base input.
The natural frequency reduction could be achieved by mounting the avionics component with
isolator grommets.
For example, the response is 11.2 GRMS at 200 Hz. The response is 7.9 GRMS at 100 Hz,
which is 3.0 dB lower than the response at 100 Hz.
References
1. Dave Steinberg, Vibration Analysis for Electronic Equipment, Wiley-Interscience, New
York, 1988.
6
Homework
1. NAVMAT P-9492 gives the power spectral density specification shown in Figure A-1.
0.1
2
0.04 g / Hz
PSD ( g / Hz )
2
0.01
0.001
20 80 350 2000
FREQUENCY (Hz)
Figure A-1.
Calculate the vibration response spectrum using Miles equation for the level in Figure A-1.
Assume an amplification factor of Q = 10. Use program dBoct.exe to calculate the input level at
the ramp frequencies.
2. Plot the vibration response spectrum from problem 1. Superimpose the vibration response
spectrum from Table 2 in the main text.
3. A component has a natural frequency of 90 Hz and an amplification factor of Q = 10. It has

been successfully tested to the 6.0 GRMS level in Figure A-1. Next, the design requirements
are changed. The component must now be tested to the 8.1 GRMS level in Figure 3. A
program management goal is to avoid additional testing despite the new requirement. How
could you justify that the original test is sufficient to meet the revised test specification? This
is a very, very typical problem in the "real-world " sense.
7
Unit 11. Vibration Response Spectrum - General Method
By Tom Irvine
_____________________________________________________________________
Introduction
The previous unit presented the Miles equation for determining the response of a single-
degree-of-freedom system subjected to base excitation, where the excitation is in the
form of a power spectral density.
Again, the Miles equation assumes that the base input is white noise, with a frequency
content from 0 to infinity Hertz.
Measured power spectral density functions, however, often contain distinct spectral peaks
superimposed on broadband random noise. The Miles equation can produce erroneous
results for these functions.
This obstacle is overcome by the "general method." The general method allows the base
input power spectral density to vary with frequency. It then calculates the response at
each frequency. The overall response is then calculated from the responses at the
individual frequencies.
The general method thus gives a more accurate response value than the Miles equation.
A drawback of the general method is that it is too cumbersome for hand calculation.
Fortunately, it can be performed via a software calculation.
Text
The main text for this unit is given in the following tutorial papers:
"An Introduction to the Vibration Response Spectrum."
"Enveloping Data via the Vibration Response Spectrum."
Please read these papers before performing the homework problems.
Homework
No hand calculations are needed for these problems. Use the software programs as noted.
1. An actual power spectral density measured on a suborbital rocket vehicle during
flight is given in file mflight.psd. This level is regarded as the base input level to
certain avionics components. Plot this function.
2. Calculate the overall GRMS value using program psdint.exe.
1
3.
a) Calculate the vibration response spectrum of mflight.psd using program vrs.exe.
Assume Q = 10. The output file is: vrs.grp. The dimensions of vrs.grp are: response
acceleration (GRMS) versus natural frequency (Hz).
b) Rename vrs.grp as flight.grp.
c) Plot flight.grp
4.
Assume that a simplified power spectral density envelope of mflight.psd is required. The
enveloping is to be performed in terms of vibration response spectra.
The power spectral envelope is permitted to clip the peaks of mflight.psd. Clipping is
desired to minimize the overall level of the envelope. Assume that the envelope will be
defined by four breakpoints. Furthermore, the middle segment will be flat. The
breakpoints are partially defined as shown in Table 1.
Table 1.
Envelope
Frequency Accel
(Hz) (G^2/Hz)
10 0.0005
1600 A
1800 A
2000 0.008
The task is to perform a trial-and-error calculation to choose the value of the "A"
amplitude. Again, there are two goals. The primary goal is that the vibration response
spectrum of the envelope power spectral density must envelop that of the flight power
spectral density. The secondary goal is to minimize the A value in Table 1.
The frequency range of interest is 10 Hz to 2000 Hz.
The task is carried out using program vrs.exe.
a) Choose an A trial value.

b) Run program vrs.exe using the breakpoints in Table 1 and an amplification factor of
Q=10.
c) Plot the resulting vrs.grp file against the flight.grp file from problem 3c.
d) Perform steps a through c until the trial value A yields a vrs.grp function which just
barely envelops the flight.grp curve.
e) The value A in Table 1 is now chosen. Calculate the overall level of Table 1.
f) Plot the Table 1 power spectral density against the mflight.psd level.
2
g) The mflight.psd curve has a distinct peak at 1710 Hz. The Table 1 envelope curve
should clip this peak by a substantial margin. What is the margin in dB? Use
program dBoct.exe.
h) The Table 1 envelope curve should have a higher overall GRMS value than the
mflight.psd curve. What is the difference in dB? Use program dBoct.exe with the
SRS option.
3
SHOCK AND VIBRATION RESPONSE SPECTRA COURSE Unit 12.
Synthesizing a Time History to Satisfy a Power Spectral Density using Sinusoids
By Tom Irvine
Introduction
Consider the power spectral density specification Table 1 in Figure 1. This level is taken
from MIL-STD-1540C.
POWER SPECTRAL DENSITY MIL-STD-1540C ACCEPTANCE LEVEL 6.1 GRMS
0.1
ACCEL (G /Hz)
2
0.01
0.001
10 100 1000 2000
FREQUENCY (Hz)
Figure 1.
Table 1.
MIL-STD-1540C Acceptance
Level,
6.1 GRMS Overall
Frequency PSD
(Hz) 2
(G /Hz)
20 0.0053
150 0.04
600 0.04
2000 0.0036
1
Previous Units gave methods for determining the response of a single-degree-of-freedom
system to a base input power spectral density, such as Figure 1. The specific tools were
the Miles equation and the general method.
Indeed, vibration analyses are often performed in the frequency domain. Nevertheless,
there are certain situations that require analysis in the time domain. In these situations,
the analyst may need to convert a power spectral density specification into an equivalent
time history.
Furthermore, consider a closed-loop random vibration test where a computer controls a
shaker to a specified power spectral density, such as Figure 1. The control software must
synthesize a corresponding time history. Control computers thus have built-in synthesis
algorithms
The purpose of this Unit is to give a method for synthesizing a time history to satisfy a
power spectral density specification. The time history will be composed of a series of
sinusoids.
Phase Angles
As an important note, a given time history has a unique power spectral density.
On the other hand, a given power spectral density does not have a unique time history.
The reason is that the phase angles are discarded in the power spectral density
calculation.
Synthesis Steps
The synthesis steps are outlined in Table 2.
This synthesis method yields a time history that is, in reality, periodic or deterministic.
Nevertheless, the goal is to obtain a time history that reasonably resembles a random
signal. Obviously, the time history must also satisfy the specified power spectral density.
2
Table 2. Synthesis Steps
Step Description
1 Choose the frequencies f i. These frequencies can be chosen arbitrarily. For
simplicity, constant frequency spacing is recommended.
2 Choose the phase angles φi. Typically, the phase angles are generated using a
random number routine.
Calculate the amplitudes A i from the specified power spectral density. Let P i
3
represent the power spectral density amplitude at frequency index i. Note that
2
the units of Pi are G /Hz. Furthermore, the G is actually GRMS.
A i = 2 Pi ∆ f i
∆f i is the frequency bandwidth associated with each frequency f i. Again, a

simple approach is to make the bandwidth a constant term.
Sum the individual spectral components for each time t of interest. Choose a
constant sampling rate. The sampling rate should be ten times the highest
spectral frequency. The acceleration is Y(t) as given by the following formula.
4
n
Y( t ) = ∑ A i sin(2πf i t + φi )
i =1
5 Judge the quality of the time history. Take a histogram. The histogram should
resemble a Gaussian bell-shaped curve. Also, take the kurtosis value, as
explained in Appendix A. The kurtosis should be approximately 3.0.
6 Calculate the power spectral density of Y(t) to verify compliance with the
specification.
Synthesis with a Coarse Frequency Resolution

Recall the power spectral density specification from Figure 1. A proposed synthesized
time history is shown in Figure 2a. The duration is 5 seconds. A close-up view of the
time history is given in Figure 2b.
The time history was made using a frequency resolution of 10 Hz. Thus, there is a
sinusoidal function defined at each 10 Hz increment.
Obviously, the signal has a periodic appearance.
3
SYNTHESIZED TIME HISTORY WITH ∆f = 10 Hz
200
150
100
50
ACCEL (G)
-50
-100
-150
-200
0 1 2 3 4 5
TIME (SEC)
Figure 2a.
4
SYNTHESIZED TIME HISTORY WITH ∆f = 10 Hz
100
50
ACCEL (G)
-50
-100
0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 2b.
5
HISTOGRAM OF SYNTHESIZED TIME HISTORY ∆f=10 Hz
16000
12000
COUNTS
8000
4000
0
-20 -15 -10 -5 0 5 10 15 20
AMPLITUDE (G)
Figure 3.
The histogram of the synthesized time history is given in Figure 3. The function is
somewhat skewed to the negative side. It thus departs somewhat from the bell-curve
ideal.
6
0.1
Synthesis, ∆f=10 Hz, 6.1 GRMS
Specification, 6.1 GRMS
ACCEL (G /Hz)
2
0.01
0.001
10 100 1000 2000
FREQUENCY (Hz)
Figure 4.
The power spectral density function of the synthesized time history is given in Figure 4.
It agrees well with the specification.
Note that the power spectral density function has a bandwidth of 9.77 Hz. On the other
hand, the time history has sinusoids at each 10 Hz increment.
The descriptive statistics of the time history are given in Table 3.
Table 3. Synthesized Time History with ∆f = 10 Hz

Parameter Value
Overall GRMS 6.1
Kurtosis 3.4
Again, the ideal kurtosis value for random vibration is 3.0.
7
Synthesis with a Fine Frequency Resolution
Another proposed time history is given in Figure 5a. A close-up view is given in Figure
5b. The frequency resolution is 0.1 Hz, which is 100 times finer than the previous
example. A sinusoidal function is thus defined at each 0.1 Hz increment.
The time history has a random appearance.
SYNTHESIZED TIME HISTORY WITH ∆f = 0.1 Hz
200
150
100
50
ACCEL (G)
-50
-100
-150
-200
0 1 2 3 4 5
TIME (SEC)
Figure 5a.
8
SYNTHESIZED TIME HISTORY WITH ∆f = 0.1 Hz
100
50
ACCEL (G)
-50
-100
0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 5b.
9
HISTOGRAM OF SYNTHESIZED TIME HISTORY ∆f=0.1 Hz
16000
12000
COUNTS
8000
4000
0
-25 -20 -15 -10 -5 0 5 10 15 20 25
AMPLITUDE (G)
Figure 6.
The histogram of the 0.1 Hz synthesized time history is shown in Figure 6. The
histogram has a bell-curve appearance. The histogram in Figure 6 is much closer to the
bell-shape curve ideal than the histogram of the 10 Hz synthesis, as was shown in Figure
4.
10
0.1
Synthesis, 0.1 Hz, 6.2 GRMS
Specification, 6.1 GRMS
ACCEL (G /Hz)
2
0.01
0.001
10 100 1000 2000
FREQUENCY (Hz)
Figure 7.
The power spectral density is given in Figure 7. The 0.1 Hz synthesis has a good match
with the specification, although the 10 Hz synthesis gave a closer match as shown in
Figure 4.
The descriptive statistics of the 0.1 Hz synthesis are given in Table 4.
Table 4. Synthesized Time History with ∆f = 0.1 Hz

Parameter Value
Overall GRMS 6.2
Kurtosis 2.998
The measured kurtosis value nearly meets the 3.0 ideal.

Overall, the 0.1 Hz synthesis is a very good approximation of a random signal, which
meets the specified power spectral density.
11
Conclusion
This Unit has given a method for synthesizing a time history to meet a specified power
spectral density function. The synthesized time history is periodic rather than random.
Nevertheless, a periodic time history can be made to appear random by using a very fine
frequency resolution.
The drawback to this method is excessive computational time. For the second example,
nearly 2 billion data points were required in the computation.
 samples  ( 2000 − 20) Hz 



20,000
sec  [ ]
5 sec   ≈2 billion
 0.1 Hz 
A more efficient algorithm will be considered in Unit 13.
12
Homework
1. Calculate the overall GRMS level of the power spectral density Table 5 using
program psdint.exe. Plot the power spectral density.
Table 5.
Frequency Level
(Hz) (G^2/Hz)
10 0.001
40 0.2
60 0.2
100 0.01
200 0.01
2. Use program psd_sine.exe to synthesize a time history to satisfy the power spectral
density in Table 5. Use the parameters in Table 6. Plot the time history.
Table 6. Synthesis Parameters

Duration 4 seconds
Sampling Rate 2000 samples/sec
Frequency 0.1 Hz
Resolution
3. Use program maxfind.exe to calculate the GRMS and kurtosis of the time history.
Compare the GRMS value of the time history with the GRMS value of the power
spectral density.
4. Calculate and plot the power spectral density of the synthesized time history. Use
program poweri.exe. Superimpose this time history against the level in Table 1.
5. Calculate and plot the histogram of the synthesized time history using program
histogra.exe. Use a delta x value of 2.
13
APPENDIX A
Kurtosis Theory
Kurtosis is a measurement of the “peakedness" of the data. A sine wave, for example,
has a kurtosis value of 1.5. A pure Gaussian random signal has a kurtosis value of 3.0. A
higher kurtosis value may indicate the presence of higher sigma peaks on the time history
than would be expected from a Gaussian distribution. These peaks could, for example,
represent a transient event. On the other hand, they may simply represent random
vibration that has a probability density function that departs from the Gaussian ideal.
The equation is:
kurtosis =
∑ ( x − X )4 ,
nσ4
where n = number of samples,

σ = s tan dard deviation,
X = mean value.
Note that the kurtosis value appears dimensionless.
14
Synthesizing a Time History to Satisfy a Power Spectral Density using Random Vibration
By Tom Irvine
Introduction
Recall that white noise is a random time history that has constant power spectral density
amplitude.
White noise can be manipulated to yield a time history that satisfies a given power spectral
density specification. The purpose of thus Unit is to present a method for this process. Note that
the method alters the time history so that it is no longer white, unless the specification amplitude
is constant. Nevertheless, the resulting time history is random.
The white noise approach has several advantages over a series of sinusoids. One advantage is
that the derived signal has a continuous frequency spectrum. Another is that it is random.
The main text for this Unit is given in tutorial: psdsyn.pdf. Please read this tutorial before
performing the homework.
Homework
This homework assignment is essentially a repeat of the Unit 12 homework except that the
synthesis for the Unit 13 homework is based on white noise.
1. Calculate the overall GRMS level of the power spectral density Table 1 using program
psdint.exe. Plot the power spectral density.
Table 1.
Frequency Level
(Hz) (G^2/Hz)
10 0.001
40 0.2
60 0.2
100 0.01
200 0.01
2. Use program psd_syn.exe to synthesize a time history to satisfy the power spectral density in
Table 1. Use the parameters in Table 2. Plot the time history.
Table 2. Synthesis Parameters

Duration 4 seconds
Sampling Rate Selected by psd_syn
program to be 10x highest
input frequency.
Frequency Continuous
Resolution
3. Use program maxfind.exe to calculate the GRMS and kurtosis of the time history. Compare
the GRMS value of the time history with the GRMS value of the power spectral density.
4. Calculate and plot the power spectral density of the synthesized time history. Use program
poweri.exe. Superimpose this time history against the level in Table 1.
5. Calculate and plot the histogram of the synthesized time history using program histogra.exe.
Use a delta x value of 2.
Integration of an Acceleration Time History to Determine Displacement and Velocity
By Tom Irvine
Introduction
Mechanical vibration is usually characterized in terms of acceleration. The main reason is that
acceleration is easier to measure than velocity or displacement.
Acceleration can be measured with either a piezoelectric or piezoresistive accelerometer.
Velocity measurements require a Doppler laser or a geophone. The laser is expensive, and it
requires a direct line of sight. The geophone is bulky and is intended for seismology
measurements.
Dynamic displacement can be measured by a linear variable displacement transducer (LVDT).
Unfortunately, the frequency response is only suited for low-frequency measurements.
If velocity is required, the typical method is to integrate the acceleration signal. If displacement
is required, the typical method is to double-integrate the acceleration signal. The integration can
be performed either in the time domain or frequency domain.
This tutorial focuses on time domain integration of a digital signal.
Motivation
There are at least two situations in which velocity time history measurement would be required.
The first occurs in pyrotechnic shock measurement. The second regards design criteria.
Pyrotechnic Shock
Pyrotechnic shock is almost always measured in terms of acceleration by piezoelectric
accelerometers. The shock pulse can cause the piezoelectric crystal to saturate, however, thereby
degrading or corrupting the accelerometer signal. Specifically, the acceleration time history may
have a spurious offset due to the saturation effect. This symptom is usually apparent in the
acceleration time history but becomes exaggerated in the velocity time history. Thus, velocity
time histories are used to judge the validity of acceleration time histories measured during
pyrotechnic events. Reference 1 gives further information on this technique.
In addition, some experts consider that the damage potential of pyrotechnic shock has a stronger
correlation with velocity than with acceleration.
Design Criteria
Furthermore, certain specifications and criteria are given in terms of velocity rather than
acceleration. Velocity seems to be the “metric” of choice for building vibration limits. For
1
example, Colin Gordon has established a generic vibration criteria for floor vibration in terms of
velocity. His paper is included in this Unit as homework assignment 1.
Gordon actually gives the floor limits in terms of one-third octave band velocity spectra. Thus,
the floor vibration could be measured by an accelerometer. The acceleration signal could then be
integrated to determine the velocity time history. The next step would be to calculate the
velocity Fourier transform magnitude for each band.
Integration Method
The integration of a digital signal is typically performed using the trapezoidal rule.
The trapezoidal rule for a continuous function f(x) over a < x < b is
∫a f ( x)dx = 2 [f ( x o ) + + 2f ( x n − 1 ) + f ( x n ) ]
b h
2f ( x 1 ) + 2f ( x 2 ) + L (1)
where
h = (b-a)/n
n = the number of intervals.
Note that the integration of a time history is carried out on a “running sum” basis. Let the
acceleration time history be represented by a1, a2, a3, . . . , an. The velocity time history is
calculated as follows.
∆t
v1 = [a1 ] (2)
2
v 2 = v1 + ∆ t [a 2 ] (3)
v 3 = v 2 + ∆ t [a 2 ] (4)
In general, the velocity at points inside the limits is given by
[ ]
v j = v j − 1 + ∆ t a j , for 1 < j < n (5)
Finally,
∆t
vn = vn− 1 + [a n ] (6)
2
Note that ∆ t is the time step.
More accurate methods, such as Simpson’s rule, would yield better accuracy. Nevertheless, the
challenge in vibration analysis is dealing with spurious offsets rather than achieving high degrees
of numerical precision.
2
Preprocessing and Postprocessing
Typically, the acceleration signal must be highpass filtered prior to integration in order to yield a
stable velocity signal. Filtering will be covered in an upcoming Unit.
Another method is to perform a trend removal prior to integration. This method will be
demonstrated in the following example.
Furthermore, postprocessing of the signal may also be required, as shown in the following
example. The reason is that the integration procedure effectively calculates the change in
velocity. Thus, the user must supply the initial velocity. This may be done via postprocessing.
Example
Consider the following acceleration time history, consisting of a sine function with an offset.
a (t ) = 1 + 1 sin [( 2π )(1 Hz ) t ] G (7)
The function is shown in Figure 1 for duration of 10 seconds.

Assume that the 1 G offset is due to a spurious effect within the instrumentation.
Now integrate the acceleration time history to obtain the velocity time history. The resulting
velocity signal is shown in Figure 2. The velocity signal is unstable because the integration
procedure greatly exaggerates the offset by transforming it into a ramp.
Now remove the offset. Then repeat the integration. The resulting velocity signal is shown in
Figure 3. The signal is now stable, but note that it has a mean value of 61.4 in/sec. This offset
implies rigid-body motion. This offset must now be removed in order to obtain a velocity signal
with zero mean.
The resulting velocity signal with zero mean is shown in Figure 4.
3
ACCELERATION SINE FUNCTION WITH OFFSET
5
4
3
2
ACCEL (G)
1
0
-1
-2
-3
-4
-5
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 1.
VELOCITY INTEGRATED FROM ACCELERATION
4000
VELOCITY (IN/SEC)
3000
2000
1000
0
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 2.
4
VELOCITY INTEGRATED FROM ACCELERATION,
OFFSET REMOVED PRIOR TO INTEGRATION
200
150
VELOCITY (IN/SEC)
100
50
-50
-100
-150
-200
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 3.
VELOCITY INTEGRATED FROM ACCELERATION,

OFFSET REMOVED BOTH PRIOR TO AND AFTER INTEGRATION
200
150
VELOCITY (IN/SEC)
100
50
-50
-100
-150
-200
0 1 2 3 4 5 6 7 8 9 10
TIME (SEC)
Figure 4.
5
Units Note
Acceleration is typically represented in units of G. The integrated signal has units of (G sec).
The integrated signal must be multiplied by 386 to convert the units to in/sec.
Displacement
Double integrating a digital acceleration signal to obtain displacement is difficult because
excessive highpass filtering and/or trend removal is often required.
A better method is to perform the double integration on the analog signal, prior to digitization. A
signal conditioner with integration circuitry is required for this operation.
Reference
1. Himelblau, Piersol, et al., IES Recommended Practice 012.1: Handbook for Dynamic Data
Acquisition and Analysis, Institute of Environmental Sciences and Technology, Mount
Prospect, Illinois.
Homework
1. Read Colin Gordon’s paper, “Generic Vibration Criteria for Vibration-Sensitive Equipment.”
2. Plot the acceleration time history pyro.txt. The file has two columns: time(sec) and accel
(G). This is actual data measured during a pyrotechnic shock test on a rocket vehicle.
3. Use program integ.exe to integrate pyro.txt
4. Use program mult2.exe to multiply the integrated time history by 386. The resulting file
should have two columns: time(sec) and velocity (in/sec). Plot the velocity file.
5. The resulting velocity signal in step 4 has an unstable velocity. Use program zeromean.exe
to remove the offset from pyro.txt. Call the output file: pyrozero.dat.
6. Repeat steps 3 and 4 using file pyrozero.dat.
Note that the resulting velocity signal in step 6 is an improvement. Nevertheless, it seems to still
contain some questionable low-frequency energy. Actually, highpass filtering would be a better
choice than simple mean removal for this case. Again, filtering will be taught in a future Unit.
File pyro.txt will be revisited in that Unit.
6
Integration of a Power Spectral Density Function
By Tom Irvine
_____________________________________________________________________________________________________________________
Introduction
As a review, mechanical vibration is usually characterized in terms of acceleration. The main
reason is that acceleration is easier to measure than velocity or displacement, in the context of
vibration.
Acceleration time histories may be converted to power spectral density functions for the purpose
of deriving test specifications. A typical example is the MIL-STD-1540C acceptance level is
shown in Figure 1 and in Table 1.
ACCELERATION POWER SPECTRAL DENSITY

MIL-STD-1540C ACCEPTANCE LEVEL 6.1 GRMS
0.1
ACCEL (G /Hz)
2
0.01
0.001
10 100 1000 2000
FREQUENCY (Hz)
Figure 1.
1
Table 1.
MIL-STD-1540C
Acceptance Level,
6.1 GRMS Overall
Frequency PSD
(Hz) 2
(G /Hz)
20 0.0053
150 0.04
600 0.04
2000 0.0036
The curve in Figure 1 is an acceleration power spectral density function. The curve can be
integrated to determine the overall GRMS value, as explain in Unit 7b.
The same curve may also be integrated, through a separate method, to determine the velocity
power spectral density and the displacement power spectral density. The purpose of this Unit is
to describe this method.
Integration Method
Recall the sine vibration relationships from Unit 2b.
Again, consider a single-degree-of-freedom system undergoing sinusoidal excitation. The
displacement amplitude x(t) is
x ( t ) = X sin (ω t ) (1)
where
X is the displacement
ω is the frequency (radians/time)
The velocity x& ( t ) is obtained by taking the derivative.
x& ( t ) = ωX cos(ω t ) (2)
The acceleration &x&( t ) is obtained by taking the derivative of the velocity.
&x&( t ) = −ω 2 X sin (ω t ) (3)
The relationships are summarized in Tables 2 and 3.
2
Table 2.
Peak Values Referenced to Peak Displacement.
Parameter Equation
Displacement x peak = X
Velocity x& peak = ωX
Acceleration &x& peak = ω 2 X
Note that
&x& peak = ω 2 x peak (4)
Now let A be the peak acceleration. The relationships in Table 3 can be derived via algebra.
Table 3.
Peak Values Referenced to Peak Acceleration
Parameter Equation
Displacement
x peak = A / ω 2
Velocity x& peak = A / ω
Acceleration &x& peak = A
The relationships in Tables 2 and 3 can be applied to power spectral density functions. Recall
that a power spectral density functions have dimension of [ (amplitude^2 ) / Hz ]. Thus, the
appropriate ω scale factor must be squared.
Let
DPSD = displacement power spectral density
VPSD = velocity power spectral density
APSD = acceleration power spectral density
Note that each PSD function is a function of the frequency f. Furthermore, the angular frequency
is ω = 2π f .
The resulting relationships for the power spectral density functions are shown in Tables 4 and 5.
Table 4.
PSD Functions Referenced to Displacement PSD
Parameter Equation
VPSD VPSD = ω2 DPSD
APSD APSD = ω4 DPSD
3
Table 5.
PSD Functions Reference to Acceleration PSD
Parameter Equation
DPSD DPSD = APSD / ω4
VPSD
VPSD = APSD / ω2
Again, each PSD function is a function of frequency.
The steps in Table 4 are actually differentiation steps. Those in Table 5 are integration steps.
Example
The integration method can easily be performed using a computer program or Excel spreadsheet.
Calculate the velocity power spectral density from the acceleration power spectral density in
Figure 1. The results are shown in Table 6 and in Figure 2.
Table 6. Integrate APSD to Determine VPSD

Frequency APSD ω ω^2 VPSD VPSD
(Hz) (G^2/Hz) (rad/sec) (rad/sec)^2 [(G sec)^2]/Hz [(in/sec)^2]/Hz
20 0.0053 1.26E+02 1.58E+04 3.36E-07 5.00E-02
150 0.04 9.42E+02 8.88E+05 4.50E-08 6.71E-03
600 0.04 3.77E+03 1.42E+07 2.81E-09 4.19E-04
2000 0.0036 1.26E+04 1.58E+08 2.28E-11 3.40E-06
Note that a unit conversion is necessary to obtain the values in the last column.
Furthermore, the velocity power spectral density can be integrated to determine the overall
velocity level from the area under the curve. This is done using the previous integration method
as explained in Unit 7b.
4
VELOCITY POWER SPECTRAL DENSITY MIL-STD-1540C ACCEPTANCE LEVEL
OVERALL LEVEL = 1.7 inch RMS /sec
-1
10
-2
10
VELOCITY [ (in/sec) / Hz ]
2
-3
10
-4
10
-5
10
-6
10
10 100 1000 2000
FREQUENCY (Hz)
Figure 2.
5
Homework
NAVMAT P-9492 gives the acceleration power spectral density specification shown in Figure 3.
Use this function for problems 1 and 2.
0.1
2
0.04 g / Hz
PSD ( g / Hz )
2
0.01
0.001
20 80 350 2000
FREQUENCY (Hz)
Figure 3.
1. Calculate corresponding velocity power spectral density and the overall velocity RMS level.
Use hand calculations or a spreadsheet.
2. Calculate the corresponding displacement power spectral density and the overall
displacement RMS level.
3. Check your results in problems 1 and 2 using program psdint.exe. This program was
included in Unit 7c.
4. Most random vibration test specifications are given as acceleration power spectral density
functions. The starting frequency is usually 10 Hz or 20 Hz. What is the reason against
specifying a starting frequency below 10 Hz?
6
Sample Rate Criteria and the Nyquist Rule
By Tom Irvine
_________________________________________________________________
Introduction
Engineers collect accelerometer data in a variety of settings. Examples include:
1. Aerospace vehicle flight data
2. Automotive proving grounds
3. Machinery condition monitoring
4. Building response to seismic excitation
5. Modal testing of structures
The accelerometers measure the data in analog form. The analog signal is sent through a
signal conditioner. The signal conditioner may have an analog lowpass filter. Filtering
will be covered more extensively in Unit 18.
An understanding of sample rate criteria requires some preliminary consideration of
filtering, however. Lowpass filtering of the analog signal is necessary to prevent an error
source called aliasing. Aliasing is covered in Unit 17.
Eventually, the accelerometer data is passed through an analog-to-digital converter. The
proper sampling rate must be selected to ensure that the digitized data is accurate.
This Unit gives guidelines for choosing the sampling rate. It also briefly covers amplitude
resolution.
Sampling Rate, First Requirement

The first requirement is that the sampling rate must be greater than the maximum analysis
frequency. Industry has established guidelines for this requirement, as discussed in
Reference 1. These guidelines are summarized in Table 1.
Table 1. Sampling Rate First Requirement
(minimum sampling rate) > ( N )( maximum analysis frequency )
Analysis Type N
Frequency Domain 2
Time Domain 10
1
Fourier transforms and power spectral density functions are used in frequency domain
analysis.
The shock response spectrum (SRS) is an example of a time domain analysis. The shock
response spectrum will be covered in Units 19 through 22.
Frequency Domain
The frequency domain requirement in Table 1 is based on the fact that at least two time-
domain coordinates per cycle are required to resolve a sine wave for analytical purposes.
The Nyquist frequency is equal to one-half the sampling rate. The frequency domain
analysis thus extends up to the Nyquist frequency.
Note that some conservative sources specify a value of N = 2.5 for frequency domain
analysis.
Time Domain
Reference 1 gives the following guideline:
Unlike other spectral quantities evolving from the discrete Fourier transform
computations, the SRS is essentially a time domain quantity. Hence, the digital
sampling rate given by Rs=1/(delta t), introduces errors beyond those associated
with aliasing about the Nyquist frequency. Thus, Rs must be high enough to
accurately describe the response of the SRS oscillators. To minimize potential
error, it is recommended that the SRS computations be performed with a
sampling rate of Rs > 10 fh, where fh is the highest natural frequency of the SRS
computation.
A sampling rate of 100,000 samples per second is thus required for a shock response
spectrum analysis extending to 10,000 Hz per this guideline. Again, the shock response
spectrum is calculated in the time domain.
Sampling Rate, Second Requirement

The second requirement is that the sampling rate must be greater than the maximum
frequency present in the source energy at the measurement location. This requirement is
necessary to prevent aliasing.
2
The guidelines for the second requirement are summarized in Table 2
Table 2. Sample Rate Second Requirement
(minimum sampling rate) > ( M )( maximum frequency in source energy )
Analysis Type M
Frequency Domain 2
Time Domain 10
Note the similarity between Tables 1 and 2.
Shannon’s sampling theorem states that a sampled time signal must not contain
components at frequencies above the Nyquist frequency, from Reference 2. Again, the
Nyquist frequency is equal to one-half the sampling rate. Shannon's theorem applies to
frequency domain analysis.
Lowpass Filtering
In many cases, the maximum expected frequency is unknown. Thus, lowpass filtering can
be used as a precaution to ensure compliance with the requirement in Table 2. Filtering
will be covered in Units 17 and 18.
Summary
Note that the maximum source energy frequency may be independent of the maximum
analysis frequency. Thus, the first and second requirements may be independent.
A common example of this independence occurs in rocket vehicle vibration testing.
Avionics components are typically subjected to power spectral density specifications
which are defined up to 2000 Hz. The test specifications assume that the components are
immune to vibration above 2000 Hz. The same specifications, however, assume that the
components must be tested up to 2000 Hz to verify their integrity, even if the expected
flight levels occur at a lower frequency domain.
The component test specifications are derived, in part, from measured or predicted flight
levels. Note that a rocket vehicle is excited by aerodynamic turbulence and motor
pressure oscillations during its powered flight. The content of this energy may be well
below, or perhaps above, 2000 Hz.
3
An engineer designing a telemetry system must thus considered the expected flight
vibration environments as well as post-flight analytical needs.
Amplitude Resolution
Amplitude resolution is considered in this Unit as a supplementary topic.
Analog-to-digital conversion systems have amplitude resolution, which is measured in bits.
The amplitude resolution is one part in 2^(number of bits). Thus, a 12-bit system has a
resolution of one part in 4096.
Consider a 12-bit system set up to measure a full scale amplitude of 10 V peak-to-peak.
The resolution is ( 10 V / 4096) = 2.4 mV. This example is shown in Table 3, along with
two other cases.
Table 3. Example for 10 V peak-to-peak

Bit Resolution Voltage Resolution
(mV)
8 39.1
12 2.4
16 0.15
Note that telemetry data is sometimes given in terms of bits, where the bits are in integer
format. The user must apply a scale factor to the bit values. The scale factor might
convert the bit values to volts, or to some engineering unit.
The voltage resolution is proportional to the G level for an accelerometer. For example,
consider the following configuration:
1. An accelerometer has a 10 V peak-to-peak range.

2. The accelerometer sensitivity is 10 G/volt ( 0.010 G /mV).
3. The accelerometer signal is applied to a 12 bit acquisition system
In this case, the accelerometer data would have an amplitude bit resolution of 0.024 G.
This would cause a measured sine wave to have a "stair-step" appearance if the peak
amplitude were below, say, 0.2 G.
The bit resolution for a data acquisition system is usually fixed. The user can manipulate
the resolution by choosing an accelerometer with a particular sensitivity. The user may
also have control over the full-scale voltage value.
4
References
1. IES Handbook for Dynamic Data Acquisition and Analysis, Institute of Environmental
Sciences, Illinois. Particularly, paragraphs 3.7.2 and 5.5.3.5.
2. R. Randall, Frequency Analysis Third edition, Bruel & Kjaer, 1987.
Homework
1. A flight accelerometer must be mounted on a rocket motor for a static fire test. The
motor will be mounted to fixed frame for this test. The resulting vibration data will be
transformed into power spectral density curves with an upper frequency of 2000 Hz.
The highest energy component in the data is expected to be 5000 Hz, however. What
is the minimum sampling rate per the guidelines given in this Unit?
2. The same data in problem 1 will be converted to a shock response spectrum. A time
domain calculated is required to form this spectrum. The spectrum will have an upper
frequency of 10,000 Hz. What is the recommended minimum sampling rate?
3. Some actual flight telemetry data is given in file: t_flight.txt. The data has two
columns: time (sec) and amplitude (bits). The amplitude resolution is: 0.2 G/bit.
Convert the amplitude column to G units and plot the data. You may use a
spreadsheet to perform the conversion. Or you may use program A_scale.exe.
4. What are the dominant frequencies of the accelerometer signal in problem 3? Note
that it represents a transient event which occurred during the burnout of a solid,
upper-stage motor.
5
Aliasing
By Tom Irvine
_________________________________________________________________
Introduction
Again, engineers collect accelerometer data in a variety of settings. Examples include:
1. Aerospace vehicle flight data
2. Automotive proving grounds
3. Machinery condition monitoring
4. Building response to earthquake and seismic excitation
5. Modal testing
The accelerometers measure the data in analog form. The accelerometer may have an
integral mechanical lowpass filter. Furthermore, the signal conditioning unit may have an
analog lowpass filter.
Eventually, the accelerometer data is passed through an analog-to-digital converter. The
proper sampling rate must be selected to ensure that the digitized data is accurate.
Sampling rate guidelines were given in Unit 16.
Lowpass filtering of the analog signal is necessary to prevent an error called aliasing.
The purpose of this Unit is to discuss aliasing.
Filtering is briefly mentioned in this Unit. The details of filtering will be covered in Unit
18.
Aliasing Examples
The following examples show the consequences of failure to comply with the sampling
rate guidelines in Unit 16. An aliasing error results.
Consider a sine wave sampled at 2000 samples per second. The Nyquist frequency is thus
1000 Hz. The Nyquist frequency is also the upper limit for a frequency domain
calculation, per the Unit 16 guidelines.
The power spectral density function of a 200 Hz sine wave sampled at this rate is given in
Figure 1. As expected, a spectral line appears at 200 Hz.
1
The power spectral density of an 1800 Hz sine wave is given in Figure 2. Note that
aliasing occurs. The 1800 Hz signal is folded about the Nyquist frequency such that a
spectral line appears at 200 Hz. The Nyquist frequency thus forms a line of symmetry.
The power spectral density of a 200 Hz sine wave appears to equal that of a 1800 Hz sine
wave. Again, this error occurs due to inadequate sampling rate.
The time histories for each of these sine waves is given in Figure 3. Note that the 1800
Hz sine wave appears to equal a 200 Hz sine wave with a 180 degree phase difference.
The alias frequency is summarized in equation (1).
1
Alias frequency = Sf − E f , if Sf < E f < Sf
2
where (1)
Sf is the sample rate
Ef is the energy frequency
In addition, aliasing will occur if the energy frequency is above the sample rate. A
separate formula is required, however.
Lowpass Filtering
Aliasing can be prevented by lowpass filtering the analog data.
Consider a stage separation test or a launch vehicle flight. The maximum expected
frequency in the source energy is essentially unknown. Thus, there is no proper means to
set the sampling rate, other than setting it at some exceedingly high value.
The simple solution is to pass the analog data through a lowpass filter as shown in the
flowchart.
Accelerometer Signal Analog-to-Digital

Conditioner with Converter
Lowpass Filter
The lowpass filter removes the high-frequency energy from the signal. This filter is often
called an "anti-aliasing" filter.
2
The filter can be part of the signal conditioning system. Typically, a Butterworth filter is
used. The Butterworth filter has a roll-off which attenuates the signal by 3 dB at the cut-
off frequency.
The cut-off frequency is typically set at, or slightly above, the maximum analysis
frequency.
Recommended Filtering Parameters
Let fc be the cutoff frequency.

Let fN be the Nyquist frequency.
Reference 1 gives the following guidelines:

(1) A lowpass anti-aliasing filter with a cutoff rate of at least 60 dB/octave
should be used for the analog-to-digital conversion of all dynamic data.
(2) With a 60 dB/octave cutoff rate, the half-power point cutoff frequency of the
filter should be set at fc < 0.6 fN.
If the anti-aliasing filter has a more rapid cutoff rate, a higher cutoff frequency can be
used, but the bound fc < 0.8 fN should never be exceeded.
Telemetry Design Example

Ideally, the sampling rate could be chosen after the maximum excitation and analysis
frequencies were identified. Practical considerations often require a reverse approach.
Consider a telemetry system for a launch vehicle. Several accelerometers will be mounted
in the vehicle. The data will be digitized on-board the vehicle. The digitized signal will be
sent via a radio link to a ground station.
The flight dynamic environments are unknown. The maximum sampling rate, however, is
4000 samples per second for each accelerometer channel. This sampling rate is
constrained by the available radio link bandwidth and other considerations.
Given this constraint, choose an analog lowpass filter with a cut-off frequency at 2000 Hz.
This frequency does not meet the strict guidelines in Reference 1, which would set the cut-
off frequency at 1200 Hz. Some compromise is often required in telemetry system design,
however. In this case, the cut-off frequency is set higher than the guidelines in order to
capture additional data beyond 1200 Hz.
3
The lowpass filter is placed between the accelerometer and the vehicle’s analog-to-digital
converter.
Now consider that the vehicle has flown and the digital data has been received at the
ground station.
Power spectral density functions of the flight data can be calculated up to 2000 Hz, per
Table 1. Some roll-off may appear starting at about 1600 Hz depending on the filter
characteristics, but this is a practical trade-off.
Table 1. Sampling Rate First Requirement
(minimum sampling rate) > ( N )( maximum analysis frequency )
Analysis Type N
Frequency Domain 2
Time Domain 10
Recall that Fourier transforms and the power spectral density functions are used in
frequency domain analysis.
On the other hand, the shock response spectrum is a time domain function.
Shock response spectra of the flight data can be calculated accurately up to 200 Hz, per
Table 1. This frequency can be extended somewhat if greater error margins are allowed.
This telemetry system will thus yield usable vibration data.
On the other hand, the telemetry system will yield marginal shock data. The resulting
shock data may be adequate to characterize motor ignition and launch shock, which are
typically dominated by energy below 2000 Hz. Unfortunately, the telemetry data will be
inadequate to characterize high-frequency pyrotechnic shock from stage separation events.
Stage separation shock must thus be measured during ground development tests prior to
flight. Data acquisition systems with high sampling rates can be used during ground tests.
Reference
1. IES Handbook for Dynamic Data Acquisition and Analysis, Institute of Environmental
Sciences, Illinois.
4
POWER SPECTRAL DENSITY OF SINE WAVE
0.5
Time Domain Amplitude = 1 G
Frequency = 200 Hz
Sample Rate = 2000 sample/sec
0.4 Duration = 1.025 sec
Overall PSD Level = 0.71 GRMS
ACCEL (G**2/Hz)
0.3
0.2
0.1
0
0 100 200 300 400 500 600 700 800 900 1000
FREQUENCY (Hz)
Figure 1.
POWER SPECTRAL DENSITY OF SINE WAVE (ALIASING ERROR)
0.5
Time Domain Amplitude = 1 G
Frequency = 1800 Hz
Sample Rate = 2000 sample/sec
0.4 Duration = 1.025 sec
Overall PSD Level = 0.71 GRMS
ACCEL (G**2/Hz)
0.3
0.2
0.1
0
0 100 200 300 400 500 600 700 800 900 1000
FREQUENCY (Hz)
Figure 2.
5
TIME HISTORIES OF TWO SINE WAVES, EACH SAMPLED AT 2000 SAMPLES/SEC
2
1800 Hz sine wave
200 Hz sine wave
1
ACCEL (G)
-1
-2
0 0.005 0.010 0.015 0.020
TIME (SEC)
Figure 3.
Homework
1. A telemetry system has a sampling rate of 600 samples per second. It has no lowpass
filtering. The system measures a sinusoidal time history with an apparent frequency of 150
Hz.
Assume that the signal may have an aliasing error. Assume that the true frequency would
be greater than the Nyquist frequency but less than the sampling rate. What is the possible
frequency of the true energy signal?
Note: discerning the true frequency in the above example may be impossible. Additional,
data or analysis would be required to address the problem.
6
2. The student is highly encouraged to obtain a copy of Reference 1, available from:
http://www.iest.org
3. Watch an old western movie. Notice whether the stagecoach wheels appear to rotate
backwards. If so, this is a visual form of aliasing.
4. Set a ceiling fan at is lowest speed. Blink your eyes. Experiment with different blink
rates. If the fan appears to rotate backwards, this is another form of aliasing.
7
Filtering
By Tom Irvine
_________________________________________________________________
Introduction
Filtering is a tool for resolving signals. Filtering can be performed on either analog or
digital signals.
Furthermore, filtering can be used for a number of purposes.
For example, analog signals are typically routed through a lowpass filter prior to analog-
to-digital conversion. The lowpass filter in this case is designed to prevent an aliasing
error. This is an error whereby high frequency spectral components are added to lower
frequencies.
Another purpose of filtering is to clarify resonant behavior by attenuating the energy at
frequencies away from the resonance.
Filtering theory is discussed in Reference 1. The student may read this reference at his or
her own leisure.
This Unit is concerned with practical application and examples. It covers filtering in the
time domain using a digital Butterworth filter. This filter is implemented using a digital
recursive equation in the time domain.
Highpass and Lowpass Filters

A highpass filter is a filter which allows the high-frequency energy to pass through. It is
thus used to remove low-frequency energy from a signal.
A lowpass filter is a filter which allows the low-frequency energy to pass through. It is
thus used to remove high-frequency energy from a signal.
A bandpass filter may be constructed by using a highpass filter and lowpass filter in
series.
Butterworth Filter Characteristics

A Butterworth filter is one of several common infinite impulse response (IIR) filters.
Other filters in this group include Bessel and Chebyshev filters. In addition, these filters
are classified as feedback filters.
The Butterworth filter can be used either for highpass, lowpass, or bandpass filtering.
1
A Butterworth filter is characterized by its cut-off frequency. The cut-off frequency is
the frequency at which the corresponding transfer function magnitude is –3 dB,
equivalent to 0.707.
A Butterworth filter is also characterized by its order. A sixth-order Butterworth filter is
the filter of choice for this Unit. Further details on the significance of order are given in
Reference 1.
A property of Butterworth filters is that the transfer magnitude is –3 dB at the cut-off
frequency regardless of the order. Other filter types, such as Bessel, do not share this
characteristic, however.
Consider a lowpass, sixth-order Butterworth filter with a cut-off frequency of 100 Hz.
The corresponding transfer function magnitude is given in Figure 1.
TRANSFER MAGNTUDE SIXTH ORDER BUTTERWORTH FILTER

100 Hz LOWPASS
1
TRANSFER MAGNITUDE
(100 Hz, 0.707)
0.5
0.2
0.1
10 20 50 100 200
FREQUENCY (Hz)
Figure 1.
2
Note that the curve in Figure 1 has a gradual roll-off beginning at about 70 Hz. Ideally,
the transfer function would have a rectangular shape, with a corner at (100 Hz, 1.00 ).
This ideal is never realized in practice, however. Thus, a compromise is usually required
to select the cut-off frequency.
The transfer function in Figure 1 also has a corresponding phase relationship, but this is
not shown. The transfer function could also be represented in terms of a complex
function, with real and imaginary components.
A transfer function magnitude plot for a sixth-order Butterworth filter with a cut-off
frequency of 100 Hz as shown in Figure 2.
TRANSFER MAGNTUDE SIXTH ORDER BUTTERWORTH FILTER

100 Hz HIGHPASS
1
TRANSFER MAGNITUDE
(100 Hz, 0.707)
0.5
0.2
0.1
10 100 1000
FREQUENCY (Hz)
Figure 2.
Frequency Domain Implementation

The curves in Figures 1 and 2 suggests that filtering could be achieved as follows:
1. Take the Fourier transform of the input time history.
2. Multiply the Fourier transform by the filter transfer function, in complex form.
3. Take the inverse Fourier transform of the product.
3
The above frequency domain method is valid. Nevertheless, the filtering algorithm is
usually implemented in the time domain for computational efficiency.
Time Domain Implementation
The transfer function can be represented by H ( ω ) .
Digital filters are based on this transfer function, as shown in the block diagram in Figure
3. Note that xk and yk are the time domain input and output, respectively.
xk yk
Time domain
equivalent of H ( ω )
Figure 3. Filter Block Diagram
Phase Correction
Ideally, a filter should provide linear phase response. This is particularly desirable if
shock response spectra calculations are required. Butterworth filters, however, do not
have a linear phase response, for reasons discussed in Reference 2. Other IIR filters
share this problem.
A number of methods are available, however, to correct the phase response. One method
is based on time reversals and multiple filtering as shown in Figure 4.
Time Time Time

xk Time
domain Reversal domain y′
k
Reversal
equivalent equivalent
of H ( ω ) of H ( ω )
Figure 4. Phase Correction Method
Further information about refiltering is given in References 1 and 2.

An important note about refiltering is that it reduces the transfer function magnitude at
the cut-off frequency to –6 dB.
4
Example 1
Consider the synthesized time history in Figure 5. The time history appears to be
random, perhaps even white noise. The corresponding power spectral density function is
shown in Figure 6.
ACCELERATION TIME HISTORY EXAMPLE 1
15
10
5
ACCEL (G)
-5
-10
-15
0 0.2 0.4 0.6 0.8 1.0
TIME (SEC)
Figure 5.
5
POWER SPECTRAL DENSITY OF EXAMPLE 1 TIME HISTORY
0.01
0.001
ACCEL (G /Hz)
2
0.0001
0.00001
10 100 1000 5000
FREQUENCY (Hz)
Figure 6.
The power spectral density displays some characteristics of white noise. Nevertheless, a
distinct spectral peak occurs at 100 Hz. The signal is perhaps best described as “sine-on-
random.”
The behavior of the 100 Hz signal can be clarified by bandpass filtering the time history
in Figure 5. The time history is bandpass filtered from 50 Hz to 150 Hz in Figure 7. A
close-up view of a 200 millisecond segment is shown in Figure 8.
6
50 Hz to 150 Hz BP FILTERED
4
3
1
ACCEL (G)
-1
-2
-3
-4
0 0.2 0.4 0.6 0.8 1.0
TIME (SEC)
Figure 7.
7
4
3
1
ACCEL (G)
-1
-2
-3
-4
0.40 0.45 0.50 0.55 0.60
TIME (SEC)
Figure 8.
The signal in Figure 8 is not pure sine, but it can be modeled as such for engineering
purposes. The number of cycles is nearly 20. The duration is 200 milliseconds. The
dominant frequency is thus
( 20 cycles / 0.200 seconds) = 100 Hz

(1)
This calculation confirms the observation of the 100 Hz peak in the power spectral
density plot in Figure 6.
The sine function in Figures 7 and 8 tended to remain stable with time, in terms of
amplitude and frequency. Either of these parameters, however, could have shifted with
time. One of the purposes of filtering is to study this behavior.
Consider the solid rocket motor in Figure 9.
8
Cavity
Solid Propellant
Figure 9.
The cavity has an acoustic pressure natural frequency. The cavity can be modeled as a
closed pipe, because the nozzle throat diameter is very small. The propellant is expended
during powered flight. The cavity volume increases as a result. The acoustic pressure
natural frequency tends to decrease as the volume increases. Again, this behavior could
be analyzed by filtering the data.
Example 2
Reconsider the time history in Figure 6. Bandpass filter the data from 10 Hz to 60 Hz.
The resulting time history is shown in Figure 10.
The bandpass filtered data in Figure 10 is “narrowband random.” Note that the overall
level is 0.090 GRMS. The frequency bandwidth is
60 Hz – 10 Hz = 50 Hz
(2)
The power spectral density amplitude for this band can be calculated as follows
PSD =
[0.090 GRMS]2 (3)
50 Hz
PSD = 0.00016 GRMS 2 / Hz (4)
By convention, the unit is abbreviated as
PSD = 0.00016 G 2 / Hz (5)
9
1.5
1.0
0.5
ACCEL (G)
-0.5
-1.0
-1.5
0 0.2 0.4 0.6 0.8 1.0
TIME (SEC)
Figure 10.
Now compare the level in equation (5) with plot in Figure 6 over the domain from 10 Hz
to 60 Hz. The plot amplitude tends to agree with the calculation from the bandpass
filtering operation.
Power Spectral Density Summary

Example 2 demonstrates an important point: a complete power spectral density function
can be constructed via bandpass filtering in successive bands.
For example, the first band could be taken from 0 Hz to 10 Hz; the second from 10 Hz to
20 Hz, the third from 20 Hz to 30 Hz; and so on in 10 Hz increments. The overall GRMS
value is then calculated for each band. The level is then squared. The square is divided
by the bandwidth, which is 10 Hz in this example. The power spectral density level is
thus calculated for each band.
This bandpass method altogether bypasses the Fourier transform step.
10
The Fourier transform method remains more efficient for computational purposes. The
bandpass filter method, however, is easier to understand.
References
1. T. Irvine, An Introduction to the Filtering of Digital Signals, Vibrationdata
Publications, 2000.
2. Stearns and David, Signal Processing Algorithms in Fortran and C, Prentice Hall,
Englewood Cliffs, New Jersey, 1993.
Homework
1. File shock.txt is a shock pulse which occurred during the flight of a rocket vehicle.
The specific event was the tail-off, or burnout, of a rocket motor. The data is
composed of several dominant frequencies. The dominant frequencies could be
forcing frequencies, natural frequencies, or some combination of both. Use program
poweri.exe to take a power spectral density function of the data. Use a rectangular
window with 4096 samples per segment. Plot the resulting power spectral density
function. List the frequencies of the dominant peaks that are below 200 Hz.
2. One of the dominant frequencies occurs near 120 Hz. The next goal is to filter the
time history to clarify the behavior of this frequency. Use program filter.exe.
Bandpass filter the data using frequency limits of 100 Hz and 150 Hz. Plot the
results. Note the beat frequency effect, suggesting two closely-space frequencies.
11
SHOCK AND VIBRATION RESPONSE SPECTRA COURSE Unit 19
Force Shock: Classical Pulse
By Tom Irvine
____________________________________________________________________
Introduction
Consider a structure subjected to a force shock pulse.
For example, an object might be purposely struck with an impulse hammer. The force
impulse usually takes the form of a half-sine pulse. The object's resulting displacement is
typically a decaying sinusoidal pulse. The natural frequency and damping ratio of the
object can thus be determined. This is one form of modal testing.
As another example, a certain rocket vehicle must withstand the force shock from a
motor with a short burn time. The thrust versus time curve might have either a half-sine
or rectangular shape.
The same vehicle may also be required to withstand the force shock pulses from a
attitude control system, which uses bursts of nitrogen gas to reorient the vehicle during
coast periods.
The purpose of this Unit is to consider the case where the force input is in the form of a
classical pulse, such as a half-sine or rectangular pulse. The force input is applied
analytically to a single-degree-of-freedom system.
The maximum displacement response of a system with a variable natural frequency is
plotted as a force shock response spectrum. The shock response spectrum is useful for
evaluating the damage potential of the shock pulse.
Derivation of Equations
Consider a single-degree-of-freedom system.
f(t)
&y&
m
k c
where
m is the mass,
1
c is the viscous damping coefficient,
k is the stiffness,
y is the absolute displacement of the mass,
f(t) is the applied force.
Note that the double-dot denotes acceleration.
The free-body diagram is

f(t)
&y&
m
ky c&y
Summation of forces in the vertical direction
∑ F=m&y& (1)
m&y& = − cy& − ky + f ( t ) (2)
m&y& + cy& + ky = f ( t ) (3)
Divide through by m,
c k 1

&y&+   y& +   y =  f ( t ) (4)
m m m
By convention,
(c / m) = 2ξω n (5)
( k / m) = ω n 2 (6)
where
ω n is the natural frequency in (radians/sec),
ξ is the damping ratio.
2
By substitution,
1
&y&+ 2ξω n y& + ω n2 y = f (t) (7)
m
Now assume a sinusoidal force function.
f ( t ) = f o sin(ωt ) (8)
The governing equation becomes.
1
&y&+ 2ξωn y& + ωn2 x = fo sin(ωt ) (9)
m
The right-hand-side can be rewritten as
2
ω
&y&+ 2ξωn y& + ωn2 y = n fo sin(ωt ) (10)
k
3
Equation (10) can be solved via Laplace transforms. Details are given in Reference 1.
The resulting displacement is
  ξω  
y(t ) = y(0) e − ξω n t cos(ωd t ) +  n  sin (ωd t )
  ωd  
 1  − ξω n t
+ y′(0)  e sin (ωd t )
 ω d 
 
 2 
 ωn f o / k   2 2 
+ − 2ξω n ω cos(ωt ) −  ω − ω n  sin (ωt )
  2 2 2 2  
  
  ω − ω n  + (2ξ ω ω n )  

   
 
 
{ }{ }
2
 ω ωn f o / k  − ξω n t
+ e 2ξω n ωd cos(ωd t )
  2 2 2  
 ωd  ω − ω n  + (2ξ ω ω n )  
2
   
 
 
{ }ω2 + ωn2 [− 1 + 2ξ 2 ]  sin(ωd t)
2
+ e
  2 2
2 
2 
 ωd  ω − ω n  + (2ξ ω ω n )  
   
(11)
4
Equation (11) gives the response for a steady-state sine input. It is also valid for the first
half-cycle of a half-sine input. Thus, modify equation (11) for the half-sine input case.
Also, assume zero initial displacement and zero initial velocity.
y(t ) =
 
 2 
 ωn f o / k   2 2 

  2 2 2 
2     
  ω − ω n  + (2ξ ω ω n )  

   
 
 
{ }{ }
2
+ e 2ξω n ωd cos(ωd t )
  2 2
2  
 ωd  ω − ω n  + (2ξ ω ω n )  
2
   
 
 
{ }ω2 + ωn2 [− 1 + 2ξ 2 ]  sin(ωd t) ,
2
+ e
   2 2
2 
2 
 ωd  ω − ω n  + (2ξ ω ω n )  
   
for 0≤t≤T
(12)
Note that T is the half-sine duration and that
π
ω= (13)
T
5
The velocity is
y& (t ) =
 
 2 
 ωn f o / k   2 2 
  2 2
2
2    
  ω − ω n  + (2ξ ω ω n )  

   
 
 
{ } { }{ }
2
+ − ωd  e 2ξω n ωd sin(ωd t )
  2 2
2 
2 
 ωd  ω − ω n  + (2ξ ω ω n )  
   
 
 
{ } { } [ ]
2
 ω ωn f o / k  − ξω n t  2 2 2  
+ ωd  e ω + ω n − 1 + 2ξ  cos(ωd t )
  2 2 2
2    
ω
 d   ω − ω n  + (2 ξ ω ω n )  

   
 
 
{ }{2ξωn ωd cos(ωd t)}
2
+ {− ξω n } e
  2 2 2 
2 
 ωd  ω − ω n  + (2ξ ω ω n )  
   
 
 
{ }ω2 + ωn2 [− 1 + 2ξ 2 ]  sin(ωd t) ,
2
+ {− ξω n } e
  2 2 2 
2 
 ωd  ω − ω n  + (2ξ ω ω n )  
   
for 0≤t≤T
(14)
The acceleration can be found by taking the derivative of the velocity equation. A more
expedient method for computational purposes, however, is to simply take
2
ωn 2
&y& = f o sin(ωt ) − 2ξω n y& − ω n y , for 0 ≤ t ≤ T (15)
k
6
For t > T, the free vibration equation may be used to determine the velocity.
  y& (T) + (ξω n )y(T)  

y( t ) = exp[− ξω n (t − T )][ y(T) ]cos[ωd (t − T )] +   sin[ωd (t − T )] ,
  ωd  
t>T
(16)
Equation (16) is taken from Reference 2.
The velocity is found by taking the derivative.
 ω 
y& ( t ) = exp[− ξω n (t − T )]  y& (T) cos[ωd (t − T )] − n {ξ y& (T) + ω n y(T)}sin [ωd (t − T )],
 ωd 
t>T
(17)
The acceleration can be found by taking the derivative of the velocity equation. A more
expedient method for computational purposes, however, is to simply take
&y& = − 2ξω n y& − ω n2 y , for t > T (18)
Finally, the nondimensional acceleration can be found by multiplying the acceleration by

a factor of [m / f o ] .
Equations (13) through (18) thus provide a method for determining the response of
single-degree-of-freedom system to a half-sine force input. These equations can be
readily implemented in a computer program.
7
Example
Consider the example in Table 1. The input force pulse is shown in Figure 1. The
calculations were made using equations (13) through (18).
Table 1.
Force Shock Response Spectrum, Q=10, m=1 kg,
Force Input = 0.010 sec, 1 N, Half-sine Pulse
Natural Peak Positive Peak Negative

Frequency Acceleration Acceleration Figure
(Hz) (m/sec^2) (m/sec^2)
10.00 0.96 -0.37 2
20.00 0.89 -0.72 -
30.00 0.88 -1.03 -
40.00 1.09 -1.28 -
50.00 1.25 -1.45 -
60.00 1.34 -1.43 -
70.00 1.36 -1.25 3
80.00 1.32 -1.13 -
90.00 1.22 -1.04 -
100.00 1.06 -0.91 -
110.00 0.87 -0.75 -
120.00 0.66 -0.60 4
130.00 0.44 -0.53 -
140.00 0.31 -0.47 -
150.00 0.29 -0.42
160.00 0.27 -0.38
170.00 0.28 -0.35
180.00 0.35 -0.32 -
190.00 0.38 -0.33 -
200.00 0.38 -0.32 -
Note that only the peak positive and negative values are retained for each time history
response. The peak values are found via a simple search method rather than a calculus
method.
Furthermore, note that the peak response can occur either during or after the half-sine
pulse.
The overall shock response spectrum is shown in Figure 5. It is constructed by plotting
the peak positive and negative acceleration amplitudes versus natural frequency in (Hz).
8
The are other types of shock response spectra which could be plotted. For example, the
absolute value acceleration response spectrum could be plotted, instead of the individual
positive and negative spectra.
Furthermore, the peak displacement or peak velocity could be plotted versus the natural
frequency.
In addition, this process could be repeated for other classical pulses, such as trapezoidal,
sawtooth, and rectangular pulses.
INPUT FORCE PULSE 0.010 SEC, 1 N HALF-SINE PULSE
1
FORCE (N)
-1
-2
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 1.
9
RESPONSE TO INPUT FORCE PULSE ( 0.010 SEC, 1 N HALF-SINE PULSE)
NATURAL FREQ = 10 Hz Q = 10 MASS = 1 kg
4
2 2
Max = 0.96 m/sec Min = -0.37 m/sec
3
2
ACCEL (m/sec )
2
-1
-2
-3
-4
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 2.

4 2 2
3
2
ACCEL (m/sec )
2
-1
-2
-3
-4
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 3.
10
4
2 2
3
2
ACCEL (m/sec )
2
-1
-2
-3
-4
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 4.
11
SHOCK RESPONSE SPECTRUM Q=10 MASS = 1 kg
FORCE = 0.010 sec, 1 N HALF-SINE PULSE
10
Negative
Positive
PEAK ACCEL (G)
0.1
1 10 100 1000
Figure 5.
Note that the spring stiffness approaches zero as the natural frequency decreases to zero.
Rigid-body dynamic effects thus occur at very low natural frequencies. The problem
simplifies to F=ma at these low frequencies.
The peak response, however, occurs at an intermediate frequency.
References
1. T. Irvine, Total Response of a Single-degree-of-freedom System to a Harmonic

Forcing Function, Vibrationdata.com Publications, 1999.
2. T. Irvine, Free Vibration of a Single-degree-of-freedom System, Vibrationdata.com
Publications, 1999.
12
Homework
1. Repeat the example for a half-sine force input of 0.008 seconds, 1 N. Assume a
damping value of Q = 10 and a mass of 1 kg. Use program fhsine.exe for both the
time history response and shock response spectra calculations.
2. What is the natural frequency which has the highest absolute value response for
problem 1?
13
Force Shock: Arbitrary Pulse
By Tom Irvine
____________________________________________________________________
Introduction
Vibration is typically measured in terms of acceleration. An accelerometer can usually
be placed at any location on an object's surface. Note that acceleration is analogous to
voltage. The voltage between any two points of a circuit is usually easy to measure.
Force, on the other hand, is analogous to electrical current. Note that an electrical circuit
must be broken in order to measure current. The current meter itself becomes part of the
circuit. A "mechanical circuit" is often impractical to break. Thus, dynamic force
measurements are seldom made.
Nevertheless, there are some exceptions.
Consider a rocket motor mounted horizontally in a test stand for a static fire test. A force
transducer might be placed between the forward end of the motor and the test stand. The
force transducer measures the thrust of the motor. The thrust versus time curve typically
has a rectangular shape. An oscillation may be superimposed on this shape, however.
Furthermore, wind and turbulent airflow are typically modeled in terms of pressure,
which is proportional to force.
The purpose of this Unit is to consider a single-degree-of-freedom system subjected to an
arbitrary force shock.
The maximum response of a system with a variable natural frequency is plotted as a force
shock response spectrum. The response may be in terms of displacement, velocity, or
acceleration.
Consider a single-degree-of-freedom system.
f(t)
&y&
m
k c
1
where
m is the mass,
c is the viscous damping coefficient,
k is the stiffness,
y is the absolute displacement of the mass,
f(t) is the applied force.
Note that the double-dot denotes acceleration.
m&y&+ cy& + ky = f ( t ) (1)
1
&y&+ 2ξω n y& + ω n2 y = f ( t) (2)
m
where
ω n is the natural frequency in (radians/sec),
ξ is the damping ratio.
The intermediate steps of the derivation are given in Reference 1. The response can be
represented in terms of a digital recursive filtering relationship, where the index
represents time.
The displacement x at index i is
xi =
+ 2 exp[
− ξω n ∆t ]cos[
ω d ∆t ]x i − 1
− exp[
− 2ξω n ∆t ]x i − 2
 ∆t 
−   exp[
− ξω n ∆t ]{sin [
ω d ∆t ]}f i − 1
 mω d 
(3)
where ∆t is the time step,
x i is the response at time t,
x i − 1 is the response at time t-∆t,
2
x i − 2 is the response at time t-2∆t,
f i − 1 is the force input at time t-∆t,
The acceleration &x& at index i requires a two-step calculation. The first step is
&& i =
w
+ 2 exp[
− ξω n ∆t ]cos[
ω d ∆t ]x i − 1
− exp[
− 2ξω n ∆t ]x i − 2
 − 2ξω n ∆t 
+ f i
 m 
− ∆t exp [

− ξω n ∆t ]
2 (
ω n − 1 + 2ξ
2 )sin [ω d∆t]+ 2ξωn cos[ωd∆t] f i− 1
m ωd

 
  m  

(4)
The second step is
&& i + [f i / m ]
&x&i = w (5)
Example
Consider a single-degree-of-freedom system subjected to a sinusoidal force input as
shown in Figure 1. The system has a natural frequency of 100 Hz and a damping ratio of
0.05, equivalent to Q=10. The system's mass is 1 kg.
3
FORCE INPUT
1
FORCE (N)
-1
-2
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 1.
The corresponding response displacement and acceleration are shown in Figures 2 and 3,
respectively. These responses are calculated using equations (3) through (5). These
equations are implemented using program arbit_f.exe.
The shock response spectrum is shown in Figure 4. It was calculated using program
fsrs.exe.
4
DISPLACEMENT TIME HISTORY
0.05
Max = 0.015 mm Min = -0.015 mm
0.04
0.03
DISPLACEMENT (mm)
0.02
0.01
-0.01
-0.02
-0.03
-0.04
-0.05
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 2.
20 2 2
15
ACCELERATION (m/sec )
10
2
-5
-10
-15
-20
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 3.
5
SHOCK RESPONSE SPECTRUM Q=10 FORCE INPUT FROM FIGURE 1.
100
Negative
Positive
10
PEAK ACCEL (G)
0.1
10 20 50 100 200
Figure 4.
Note that the exact solution can be determined either by Laplace transforms or by other
methods. Thus, this problem presents a good opportunity to judge the accuracy of the
equations presented in this Unit. Assume steady-state behavior. The comparison is
shown in Table 1.
Table 1. Peak Response Values

Parameter Digital Recursive Steady-State
Filter
Displacement ±0.015 ±0.016
(mm)
Acceleration ±9.98 ±10.00
(m/sec^2)
Recall that the steady-state method for a sinusoidal force input was given in Unit 9.
The comparison results are good. Higher accuracy could be achieved by using a higher
sample rate. The sample rate was 2000 samples/second. In addition, the input consisted
of 20 cycles. Additional cycles would be required to achieve the steady-state ideal.
6
Reference
1. T. Irvine, An Introduction to the Shock Response Spectrum, Vibrationdata
Publications, 2000.
Homework
1. Assume a single-degree-of-freedom system with a natural frequency of 70 Hz and an
amplification factor Q=10. Assume mass = 1 kg. Calculate the acceleration response
to a 1 N, 0.010 symmetric sawtooth force pulse, as given in file: ssaw.txt. Use
program arbit_f.exe
2. Calculate the shock response spectra for ssaw.txt using program fsrs.exe.
Note that the symmetric sawtooth pulse is a classical pulse. The homework problems
could thus be solved by Laplace transforms. Nevertheless, a classical pulse is a special
example of an arbitrary pulse.
7
Base Excitation Shock: Classical Pulse
By Tom Irvine
____________________________________________________________________
Introduction
Consider a structure subjected to a base excitation shock pulse. Base excitation is also
referred to as support motion.
The pulse is typically defined in terms of acceleration, although it could be defined in
terms of displacement of velocity.
Examples of base excitation are
1. An automobile traveling down a washboard road.
2. A building subjected to an earthquake or seismic motion.
3. A crystal oscillator mounted on a circuit board subjected to shock.
For simplicity, the automobile, building, or crystal oscillator can be modeled as a single-
degree-of-freedom system. The response of the object to a given base excitation can then
be calculated given the natural frequency and the damping ratio. Note that the mass
value is not explicitly required.
Typically, the peak response acceleration is the desired output parameter. The natural
frequency of the object, however, may be unknown. Thus, the response calculation can
be performed for a series of natural frequencies. The peak acceleration response can then
be plotted against the natural frequency. The resulting function is a shock response
spectrum.
The shock response spectrum is useful for evaluating the damage potential of the shock
pulse.
The purpose of this Unit is to present the shock response spectrum for the case of a
classical base input pulse, such as a half-sine pulse.
Consider the single-degree-of-freedom system subjected to base excitation shown in
Figure 1. The free-body diagram is shown in Figure 2.
1
&x&
m
k c &y&
The variables are

m = mass,
k = stiffness,
&x&
m
k(y-x) c (y& − x&)
Define a relative displacement z as follows.

Let z=x− y
z& = x& − y&
&z& = &x& − &y&
&x& = &z&+ &y&
&z&+ 2ξω n z& + ω n 2 z = − &y& (1)
Consider the half-sine pulse given by equation (2).
2
  πt 
A sin  T , 0 ≤ t ≤T
  
&y&( t ) = 
 0, t > T


(2)
The equation of motion becomes
 πt 
&z&+ 2ξω n z& + ω n 2 z = − A sin  , 0 ≤ t ≤T (3)
T 
Let
π
ω= (4)
T
&z&+ 2ξω n z& + ω n 2 z = − A sin (ω t ), 0 ≤ t ≤T (5)
Equation (5) may be solved via Laplace transforms, as shown in Reference 1.
After many steps, the absolute acceleration during the pulse is

&x&(t ) = − ω n exp (− ξω n t )[

ω
[ ( ) ] 
ω n z(0) + 2ξ z&(0)]cos(ω d t )+ n − ξω n z(0) + 1 − 2ξ2 z&(0 ) sin (ω d t )
ωd 
Aω 2 −  
ω − ω n 
 sin (ω t )− (2ξω ω n )cos(ω t )
2 2
+   
 2 2  
2
2
ω − ω n + (2 ξ ω ω )
n 

  

+
 2
A ω ω n [exp(− ξω n t )] 
 2{ } ω n 
ω d 
2 2 { 2 ( 

 2ξ ω cos(ω d t )+   − ω n + ω 1 − 2ξ sin (ω d t ) )}
2 2 2  

ω − ω n  + (2ξ ω ω n ) 

  

+ A sin (ω t ), for 0 ≤ t ≤T
(6)
3
The absolute acceleration after the pulse is
&x&( t ) =
− ω n exp (− ξω n (t − T )){[
ω n z(T) + 2ξ z&(T)]cos(ω d (t − T ))}
ω
ω d
[ ( ) ] 
− ω n exp (− ξω n (t − T ))  n − ξ ω n z(T) + 1 − 2ξ2 z&(T ) sin (ω d (t − T )) , for t > T

(7)
Example
Consider the example in Table 1. The base pulse is shown in Figure 3. The calculations
were made using equations (6) and (7), as implemented in the ahsine.exe program.
Table 1.
Shock Response Spectra, Q=10,
Acceleration Base Input = 0.010 sec, 10 G, Half-sine Pulse

(Hz) (G) (G)
10.00 3.69 -3.15 4
20.00 7.18 -6.13 -
30.00 10.26 -8.77 -
40.00 12.79 -10.93 -
50.00 14.62 -12.50 -
60.00 15.74 -13.42 -
70.00 16.31 -13.63 -
80.00 16.51 -13.18 5
90.00 16.44 -12.15 -
100.00 16.24 -10.65 -
110.00 15.90 -8.74 -
120.00 15.51 -6.60 6
130.00 15.10 -4.42 -
140.00 14.64 -2.45 -
150.00 14.21 -1.30
160.00 13.76 -1.91
170.00 13.33 -2.84
180.00 12.92 -3.52 -
190.00 12.50 -3.84 -
200.00 12.12 -3.78 -
4
Note that only the peak positive and negative values are retained for each time history
response. The peak values are found via a simple search method rather than a calculus
method.
Furthermore, note that the peak response can occur either during or after the half-sine
pulse.
The are other types of shock response spectra which could be plotted. For example, the
absolute value acceleration response spectrum could be plotted, instead of the individual
positive and negative spectra.
Furthermore, the peak displacement or peak velocity could be plotted versus the natural
frequency.
In addition, this process could be repeated for other classical pulses, such as trapezoidal,
sawtooth, and rectangular pulses.
BASE INPUT: 0.010 sec, 10 G HALF-SINE PULSE
20
10
ACCEL (G)
-10
-20
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 3.
5
SDOF RESPONSE: FN = 10 Hz, Q=10
20
15
10
5
ACCEL (G)
-5
-10
-15
-20
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 4.

20
15
10
5
ACCEL (G)
-5
-10
-15
-20
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 5.
6
20
15
10
5
ACCEL (G)
-5
-10
-15
-20
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14
TIME (SEC)
Figure 6.
SHOCK RESPONSE SPECTRUM Q=10 BASE INPUT: 0.010 msec, 10 G HALF-SINE PULSE
100
Negative
Positive
PEAK ACCEL (G)
10
1
10 100 1000
Figure 7.
7
Note that the positive shock response spectrum in Figure 7 converges to the peak input
amplitude at higher natural frequencies, say above 200 Hz.
Component Shock Testing
Some shock tests are specified in terms of a base input half-sine pulse. The test may be
performed on a drop tower or on a shaker table.
There is a significant difference between the positive and negative shock response spectra
of a half-sine pulse, however, as shown in Figure 5. Thus, these specifications typically
require that the test be performed in each direction of each axis.
Testing on shaker table presents a further challenge. The net displacement and net
velocity of the shaker table must each be zero. Thus, pre and post pulses must be used to
achieve these requirements.
References
1. T. Irvine, Response of a Single-degree-of-freedom System Subjected to a Classical

Base Excitation, Vibrationdata.com Publications, 1999.
Homework
1. Repeat the example for a half-sine base input of 0.008 seconds, 10 G. Assume a
damping value of Q = 10. Use program ahsine.exe for both the time history response
and shock response spectra calculations.
2. What is the natural frequency which has the highest absolute value response for
problem 1?
3. A component has already been tested to a 50 G, 0.011 sec base input half-sine pulse.
A new requirement is that the component must withstand a 100 G, 0.002 sec half-sine
pulse. Compare the specifications in terms of their corresponding shock response
spectra. What conclusions can be made about the need for retesting?
4. Read tutorial srs_intr.pdf.
8
Base Excitation Shock: Arbitrary Pulse
By Tom Irvine
____________________________________________________________________
Introduction
Unit 21 discussed a structure subjected to a base excitation shock pulse, where the pulse
was a classical pulse such as a half-sine pulse.
The advantage of the classical pulse is that the response can be analyzed in terms of
deterministic functions, derived via Laplace transforms.
Some real-world shock events indeed take the form of a half-sine pulse. Most pulses,
however, have a complex, oscillating shape, as shown in the example in Figure 1.
PYROTECHNIC SHOCK PULSE
200
150
100
ACCEL (G)
50
-50
-100
-150
-200
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
TIME (SEC)
Figure 1.
The time history in Figure 1 is measured data from a rocket vehicle shroud separation
test. The shroud is the nosecone of the vehicle. It opens up like a clamshell prior to the
payload release.
The damage potential of the shock pulse in Figure 1 can be analyzed in terms of a shock
response spectrum, as described in previous Units.
1
The purpose of this Unit is to present the shock response spectrum for an arbitrary base
input pulse.
As a review, consider the single-degree-of-freedom system subjected to base excitation
shown in Figure 2. The free-body diagram is shown in Figure 3.
&x&
m
k c &y&
The variables are

m = mass,
k = stiffness,
&x&
m
k(y-x) c (y& − x&)
The equation of motion is
&& = c (y& − x&) + k ( y − x)

mx (1)
Define a relative displacement z. Let z= x− y.
2
&z&+ 2ξω n z& + ω n 2 z = − &y& (2)
Equation cannot be solved exactly for the case where the base input is an arbitrary pulse.
A convolution integral approach is needed as described in Reference 1.
After many steps, the response can be represented in terms of a digital recursive filtering
relationship, as derived in Reference 1.
&x&i =
+ 2 exp[− ξω n ∆t ]cos[
ω d ∆t ]&x&i − 1
− exp[− 2ξω n ∆t ]&x&i − 2
+ 2 ξω n ∆t &y&i

ω
(
+ ω n ∆t exp[− ξω n ∆t ] n 1 − 2ξ2 )sin[ω d∆t]− 2ξcos[ωd∆t]&y&i− 1
ω d
 
(3)
where
ξ is the damping ratio,
ω d = ω n 1− ξ2 ,
∆t = time step,
x i is the response at time t,
x i − 1 is the response at time t-∆t,
x i − 2 is the response at time t-2∆t,
y i is the base input at time t,
y i − 1 is the base input at time t-∆t.
3
Example
Calculate the shock response spectrum of the pulse in Figure 1, using equation (3) as
implemented in programs arbit.exe and qsrs.exe. Program arbit.exe generates a complete
time history for a given natural frequency.
Program qsrs.exe generates a complete shock response spectrum for a family of natural
frequencies.
Sample response values are shown in Table 1.
Table 1.
Shock Response Spectra of Pyrotechnic Shock Pulse, Q=10,
(Hz) (G) (G)
50 6.5 -7.3 4
100 6.5 -5.6 -
150 5.0 -6.5 -
200 7.1 -6.2 -
250 13.2 -11.5 -
300 22.5 -20.5 5
350 17.0 -17.3 -
400 14.2 -16.4 -
450 14.4 -16.1 -
500 15.2 -14.2 6
Note that the duration of the input pulse in Figure 1 extends to 0.190 seconds. The
response of the 50 Hz system is extended to 0.230 seconds, however, to allow for a few
cycles of free vibration.
Also note that only the peak positive and negative values are retained for each time
history response. The peak values are found via a simple search method rather than a
calculus method.
Each of the coordinates in Table 1 is represented in Figure 7.
The positive and negative spectral curves in Figure 7 are nearly equal, suggesting that the
data quality is good per the guidelines in Reference 1.
The absolute value shock response spectrum is shown in Figure 8. The absolute value
curve envelops both the positive and negative curves.
4
SDOF RESPONSE (fn=50 Hz, Q=10)
TO PYROTECHNIC SHOCK PULSE IN FIGURE 1
10
5
ACCEL (G)
-5
-10
0.01 0.03 0.05 0.07 0.09 0.11 0.13 0.15 0.17 0.19 0.21 0.23
TIME (SEC)
Figure 4.

30
20
10
ACCEL (G)
-10
-20
-30
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
TIME (SEC)
Figure 5.
5
30
25
20
15
10
ACCEL (G)
5
0
-5
-10
-15
-20
-25
-30
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
TIME (SEC)
Figure 6.
6
SHOCK RESPONSE SPECTRUM Q=10
PYROTECHNIC SHOCK PULSE FROM FIGURE 1.
1000
Negative
Positive
100
PEAK ACCEL (G)
10
0.1
10 100 1000 10000

Figure 7.
ABSOLUTE VALUE SHOCK RESPONSE SPECTRUM Q=10

PYROTECHNIC SHOCK PULSE FROM FIGURE 1.
1000
100
PEAK ACCEL (G)
10
0.1
10 100 1000 10000

Figure 8.
7
The local peaks in Figure 8 may correspond to structural natural frequencies which were
excited by the pyrotechnic shock. Further analysis and testing would be required to
resolve this, however.
The spectra in Figures 7 and 8 show that the response amplitude tends to rise in some
proportion to the natural frequency. This is typical of pyrotechnic shock.
Consider an avionics component that is to be mounted near the measurement location

corresponding to the data in Figure 1. A design engineer may wish to design the
component to have the smallest possible natural frequency in order to minimize the
component’s response to the shroud separation event in flight.
Other flight environments and design constraints must be considered, however, in the
design process.
References
1. T. Irvine, An Introduction to the Shock Response Spectrum, Vibrationdata.com

Publications, 2000.
Homework
1. Use program arbit.exe to calculate the time history response for file: pyro.txt. this
file contains two columns: time (sec) and accel (G). This is actual measured data
from the previously described shroud separation test. Assume a natural frequency of
50 Hz and Q=10. What are the peak positive and negative amplitudes? Program
maxfind.exe can be used to determine the peaks.
2. Calculate the complete shock response spectrum of pyro.dat using program qsrs.exe.
Again, assume Q=10. Plot the positive and negative spectra on one graph. Plot the
absolute value spectrum on another.
3. How do the peak values in problem 1 relate to the spectral curves in problem 2?
4. Read the NASA pyrotechnic shock document, NasaPyro.zip, available at:

http://www.vibrationdata.com/tutorials/.htm
8
Unit 23. Integration of Accelerometer Data to Velocity
By Tom Irvine
________________________________________________________________________
Introduction
Pyrotechnic shock data is difficult to measure accurately. Saturation effects, which are
explained in this report, may degrade the data. A saturated signal must be either
discarded or repaired.
The present Unit focuses on diagnosis methods. Future Units will cover repair
techniques.
A specific diagnosis technique is to integrate the accelerometer signal to velocity.
A velocity signal which diverges from the zero baseline indicates a poor accelerometer
signal.
A velocity signal which oscillates about the zero baseline tends to indicate a good
accelerometer signal although, additional criteria must also be satisfied.
Baseline Shift
Numerous problems can affect the quality of accelerometer data during pyrotechnic
shock events. A baseline shift, or zero shift, in the acceleration time history is perhaps
the most common error source.
Chu notes in Reference 1 that this shift can be of either polarity and of unpredictable
amplitude and duration. He has identified six causes of zero shift:
a. Overstressing of sensing elements
b. Physical movement of sensor parts
c. Cable noise
d. Base strain induced errors
e. Inadequate low-frequency response
f. Overloading of signal conditioner.
Accelerometer resonant ringing is a special example of causes "a" and "e.” This is a
particular problem if the accelerometer has a piezoelectric crystal as its sensing element.
A piezoelectric accelerometer may have its amplification factor Q well above 30 dB at
resonance. This resonance may be excited by high-frequency pyrotechnic shock energy.
Resonant ringing causes higher element stresses than expected.
Chu notes that this may cause the signal conditioner to overload, as follows
When a signal conditioner attempts to process this signal, one of its stages is
1
driven into saturation. Not only does this clipping distort the in-band signals
momentarily, but the overload can partially discharge capacitors in the
amplifier, causing a long time-constant transient.
This overload causes zero shift in the acceleration time history. This shift distorts the
low-frequency portion of the shock response spectrum.
Piersol gives criteria for determining whether a signal has a zero shift in References 2 and
3. Briefly, it involves integrating the acceleration time history to obtain a velocity time
history. The velocity time history is then inspected for any unexpected net change.
Examples
Good Signal
An accelerometer signal from a pyrotechnic shock test is shown in Figure 1. The
acceleration signal oscillates about the zero baseline. This is a good sign, but further
analysis is needed to validate the signal.
The velocity time history is shown in Figure 2. The mean value tends to drift in the
positive direction with time. The velocity criterion is not met in the ideal sense.
Nevertheless, the velocity signal is stable. For practical purposes, the velocity signal is
considered good.
The shock response spectra are shown in Figure 3. The positive and negative spectral
curves are nearly equal. This agreement completes the signal validation.
Saturated Signal
An accelerometer signal from a separate pyrotechnic shock test is shown in Figure 4.
The signal appears well behaved at first inspection.
The velocity time history is shown in Figure 5. The signal clearly diverges from the zero
baseline. This divergence represents some form of saturation. For example, the
accelerometer's own natural frequency may have been excited.
The corresponding shock response spectra are shown in Figure 6. The positive and
negative spectra diverge at certain natural frequencies. For example, the difference is 6.6
dB at 250 Hz.
Comparison
The good signal in Figure 1 had a peak acceleration of 110 G. The saturated signal in
Figure 4 had a peak acceleration of 12,000 G. This difference is not coincidental. The
probability of saturation tends to be proportional to the input level. The frequency
content is also important in this regard.
Data is precious, even if it is saturated. Some filtering or trend removal technique could
be used to remove the spurious signal from the saturated accelerometer time history in
Figure 4. These techniques will be covered in future Units.
2
Preventative Solution
An analog lowpass filter can be placed before the first input stage of the signal
conditioner to prevent overloading the electronics. Note that this filter can also be used
to prevent aliasing, as discussed in previous Units.
A more effective solution, however, is to use an accelerometer which has a mechanical
low pass filter. The purpose of this filter is to prevent excitation of the accelerometer
natural frequency. The mechanical filter is located between the accelerometer and the
measurement surface.
3
ACCELERATION TIME HISTORY SHROUD SEPARATION TEST
150
Good Signal
100
50
ACCEL (G)
-50
-100
-150
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 1.
VELOCITY TIME HISTORY SHROUD SEPARATION TIME
10
Good Signal
5
VELOCITY (IN/SEC)
-5
-10
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 2.
4
SHOCK RESPONSE SPECTRA Q=10 SHROUD SEPARATION TEST
1000
Negative Good Signal
Positive
100
PEAK ACCEL (G)
10
0.1
10 100 1000 10000
Figure 3.
5
ACCELERATION TIME HISTORY LINEAR SHAPE CHARGE
15000
10000
5000
ACCEL (G)
-5000
-10000
-15000
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 4.
VELOCITY TIME HISTORY LINEAR SHAPE CHARGE
1500
Saturation effect
1000
VELOCITY (IN/SEC)
500
-500
-1000
-1500
0 0.05 0.10 0.15 0.20
TIME (SEC)
Figure 5.
6
SHOCK RESPONSE SPECTRUM Q=10 LINEAR SHAPE CHARGE
5
10
Negative Saturated signal
Positive
4
10
PEAK ACCEL (G)
3
10
2
10
1
10
10 100 1000 10000
Figure 6.
References
1. Anthony Chu, “Zero Shift of Piezoelectric Accelerometers in Pyroshock
Measurements,” Paper presented at the 57th Shock and Vibration Symposium, 1986.
2. Allan Piersol, “Pyroshock Data Acquisition and Analysis for U/RGM-109D Payload
Cover Ejection Tests NWC TP 6927,” Naval Weapons Center, China Lake, CA,
1988.
3. Allan Piersol, "Recommendations for the Acquisition and Analysis of Pyroshock
Data," Sound and Vibration; April 1992.
Homework
1. File rv_sep.txt is an acceleration time history with dimensions: time(sec) and accel
(G). It represents measured flight data from the re-entry vehicle separation of a
suborbital rocket vehicle. Linear shape charge was the source device. The
measurement was made near the source. Integrate the signal to obtain velocity.
Multiply by a scale factor to convert the amplitude to either (inches/sec) or
(meters/sec). Also, calculate the shock response spectrum. Did saturation occur?
Note: use programs integ.exe and qsrs.exe from previous Units.
7
Unit 24. Simple Trend Removal
By Tom Irvine
________________________________________________________________________
Introduction
Measured accelerometer data may offsets or trends. These effects are usually spurious,
although an exception is considered as a homework problem.
An offset could occur due to accelerometer saturation. Or an offset could occur due to
some electrical drift inside the data acquisition system.
There are several methods for removing offsets and trends.
One method is to perform a highpass filter operation, using the method discussed in Unit
18.
Another is to perform a polynomial curve. The polynomial curve is then subtracted from
the data. The purpose of this Unit is to present polynomial trend removal.
Theory
The polynomial coefficients are found via a least-squares method.
Typically, the polynomial order is less than or equal to 2. The coefficient calculation is
prone to numerical instability for higher order polynomials.
Note that a second-order polynomial is a parabola. A first-order polynomial is a line with
a slope. A zero-order polynomial is simply a constant.
The details of the least square calculation are found in standard numerical methods
textbooks, such as Reference 1.
Example
The following example shows how trend removal can be used to refine a shock pulse.
The acceleration time history in Figure 1 is actual measured data from a rocket vehicle
pyrotechnic shock test. The data quality appears to be good at first inspection.
Now assume the velocity time history is desired. Note that some models of shock
damage are based on velocity rather than acceleration.
The velocity time history is shown in Figure 2. The velocity signal diverges severely
from the zero baseline, indicating that the acceleration time history has a problem.
1
Now perform a simple first-order curve fit on the acceleration time history, by seeking a
polynomial of the form
y ( t ) = a 0 + a1 t (1)
The curve-fit is performed using program trend1.exe. The program returns the following
equation for the acceleration time history.
y ( t ) = 2.7635 + 3.1820 t (2)
Equation (2) is shown in Figure 3.

Program trend1.exe also performs the trend removal. The resulting acceleration time
history is shown in Figure 4.
The refined acceleration time history is then integrated. The resulting velocity time
history is shown in Figure 5. It has a stable oscillation about the zero baseline.
The trend removal is thus successful.
2
ACCELERATION TIME HISTORY - RAW DATA
400
300
200
100
ACCEL (G)
-100
-200
-300
-400
0 0.02 0.04 0.06 0.08
TIME (SEC)
Figure 1.
VELOCITY TIME HISTORY - RAW DATA
100
50
VELOCITY (IN/SEC)
-50
-100
0 0.02 0.04 0.06 0.08
TIME (SEC)
Figure 2.
3
FIRST ORDER TREND FROM SHOCK TEST TIME HISTORY
3
ACCEL (G)
0
0 0.02 0.04 0.06 0.08
TIME (SEC)
Figure 3.
4
ACCELERATION TIME HISTORY - REFINED DATA
First-order trend removed.
400
300
200
ACCELERATION (G)
100
-100
-200
-300
-400
0 0.02 0.04 0.06 0.08
TIME (SEC)
Figure 4.
VELOCITY TIME HISTORY - REFINED DATA

First-order trend was removed from acceleration time history prior to integration.
15
10
VELOCITY (IN/SEC)
-5
-10
-15
0 0.02 0.04 0.06 0.08
TIME (SEC)
Figure 5.
5
Reference
1. R. Hornbeck, Numerical Methods, Prentice-Hall, New Jersey, 1975.
Homework
File fairing.txt is an acceleration time history measured during the fairing separation
event of a rocket vehicle. The fairing separation occurred during powered flight. The
dimensions are: time(sec) and accel (G). The accelerometer was a variable capacitance
model which had the ability to measure frequencies down to zero. Thus the
accelerometer could measure rigid-body acceleration as well as elastic-body acceleration.
Note that Rigid-body acceleration is typically characterized by polynomial trends. On
the other hand, elastic-body acceleration is characterized by oscillation. The fairing.txt
file has both effects.
Use program trend1.exe to remove the rigid-body trend from the acceleration time
history. Plot the acceleration time history both before and after the trend removal. Also,
integrate the acceleration time history to obtain velocity, using program integ.exe.
Convert the amplitude to either m/sec or in/sec. Plot the resulting velocity time history.
6
Unit 25. Direct Saturation Removal
By Tom Irvine
________________________________________________________________________
Introduction
Again, measured accelerometer data may offsets or trends. These effects are usually
spurious. An offset could occur due to accelerometer saturation. Or an offset could
occur due to some electrical drift inside the data acquisition system.
Unit 24 presented a polynomial trend removal method, which is suitable for simple offset
cases.
Another candidate method is highpass filtering, which was discussed in Unit 18.
The purpose of this Unit is to determine whether highpass filtering is suitable for
removing saturation effects from shock data.
Example
Raw Data
An acceleration time history from a rocket vehicle frangible joint shock test is shown in
Figure 1. The signal clearly has a saturation effect, as evidenced by the offset. The
accelerometer’s own natural frequency may have been excited.
The shock response spectrum is shown in Figure 2. The spectral levels in the low
frequency domain near 10 Hz are artificially high due to the saturation effect.
The corresponding velocity time history is shown in Figure 3. The velocity clearly
diverges from the zero baseline.
Some experts would thus argue that the signal is invalid and that the data should be
discarded. Nevertheless, data is precious. Thus, use highpass filtering to remove the
saturation effects.
Filtered Data
A 50 Hz highpass filter is used in an attempt to recover the data. The highpass filtered
acceleration time history is shown in Figure 4. The filtering effectively removes the
offset. Nevertheless, it has the unwanted side effect of introducing a spurious trend prior
to time zero.
The shock response spectrum of the filtered data is shown in Figure 5. The filtering is
successful in that it brings the positive and negative curves together. Nevertheless, the
1
initial slope of each curve is unrealistically sharp. The filtering operation has thus
removed some of the “good data” at the low frequency domain.
Finally, the velocity time history of the filtered accelerometer data is shown in Figure 6.
Conclusion
Highpass filtering is somewhat successful for removing saturation effects from signals. It
removes the saturation effects but also removes some of the “good data.” It is thus a
brute force method.
A superior method will be introduced in Unit 26.
Homework
A sample acceleration time history from a linear shape charge test is given in lsc.txt. The
data format is: time (sec) and accel (G). Use this data to repeat the example in this text.
You may experiment with the highpass filtering frequency.
2
ACCELERATION TIME HISTORY FRANGIBLE JOINT SHOCK TEST
RAW DATA
2000
1500
1000
500
ACCEL (G)
-500
-1000
-1500
-2000
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
TIME (SEC)
Figure 1. Acceleration Time History, Raw Data

Ideally, the signal would oscillate about the zero baseline. An offset is
clearly present, however. The offset is beyond the scope of simple
polynomial trend removal. Highpass filtering will be attempted as an
alternative.
3
FRANGIBLE JOINT SHOCK TEST - RAW DATA
10000
Negative
Positive
1000
PEAK ACCEL (G)
100
10
10 100 1000 10000
Figure 2. Shock Response Spectrum of Raw Data

The negative and positive spectral curves clearly diverge from one another, particularly
below 500 Hz. Furthermore, each curve is unrealistically high in the low frequency
domain near 10 Hz. These characteristics are further evidence of saturation.
4
VELOCITY TIME HISTORY FRANGIBLE JOINT SHOCK TEST
RAW DATA
5000
VELOCITY (IN/SEC)
-5000
0 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090 0.100
TIME (SEC)
Figure 3. Velocity Time History, Raw Data

The integrated signal reaches nearly 5000 in/sec, which is very unrealistic. This is
another manifestation of the saturation effect.
5
50 Hz HP FILTER
2000
1500
1000
500
ACCEL (G)
-500
-1000
-1500
-2000
0 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090 0.100
TIME (SEC)
Figure 4. Signal Highpass Filtered at 50 Hz

The highpass filter operation removes the baseline shift effect. The filtered signal
oscillates about the zero baseline. On the other hand, the filtering introduces a spurious
trend leading up to time zero. This trend could be removed by diving the signal into
segments prior to and after time zero. Then the polynomial trend could be removed from
the pre-zero segment. Then, the segments could be reassembled.
6
FRANGIBLE JOINT SHOCK TEST - 50 Hz HP FILTERED
10000
Negative
Positive
1000
PEAK ACCEL (G)
100
10
10 100 1000 10000
Figure 5. Shock Response Spectrum of Filtered Data

The data was highpass filtered at 50 Hz prior to the shock response spectrum
calculation. The filtering brings the positive and negative curves closer together, yet it
also causes an unrealistically sharp slope below 50 Hz.
Note that the effect of the highpass filtering diminishes above 1000 Hz. The curves in
Figure 5 are nearly the same as those of the raw data in Figure 2 above 1000 Hz.
7
DATA HP FILTERED AT 50 Hz PRIOR TO INTEGRATION
150
100
VELOCITY (IN/SEC)
50
-50
-100
-150
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
TIME (SEC)
Figure 6. Velocity Time History of Filtered Data

The velocity time history now oscillates about the zero baseline. The initial velocity,
however, incorrectly appears as a non-zero value. The filtering process thus has
benefits, but also some limitations.
8
Unit 26. Indirect Saturation Removal
By Tom Irvine
________________________________________________________________________
Introduction
As a review, measured accelerometer data may offsets or trends. These effects are
usually spurious. An offset could occur due to accelerometer saturation. Or an offset
could occur due to some electrical drift inside the data acquisition system.
Unit 24 presented a polynomial trend removal method, which is suitable for simple offset
cases. Unit 25 presented a highpass filtering method, which has some drawbacks.
.
The purpose of this Unit is to present an indirect filtering method for removing saturation
effects from shock data. The indirect method tends to be superior to the other methods.
Example
Raw Data
An acceleration time history from a rocket vehicle frangible joint shock test is shown in
Figure 1. This is the same time history that was considered in Unit 25. The signal clearly
has a saturation effect, as evidenced by the offset. The accelerometer’s own natural
frequency may have been excited.
The shock response spectrum is shown in Figure 2. The spectral levels in the low
frequency domain near 10 Hz are artificially high due to the saturation effect.
The corresponding velocity time history is shown in Figure 3. The velocity clearly
diverges from the zero baseline.
Some experts would thus argue that the signal is invalid and that the data should be
discarded. Nevertheless, data is precious. Thus, recovery is attempted.
Recovery Scheme
The recovery scheme is the indirect filtering method outlined in Table 1.
1
Table 1. Indirect Filtering Method
Step Description
1 Lowpass filter the acceleration signal at 100 Hz.
2 Multiply the lowpass filtered signal by a scale

factor, such as 0.95.
3 Subtract the scaled, lowpass filtered signal from
the original signal.
4 Correct pre-shock.
5 Highpass filter the data at low frequency,

< 20 Hz.
6 Calculate the shock response spectrum.
7 Calculate the velocity time history.
The purpose of the method is to retain some of the low frequency data that would
otherwise be removed by a “brute force” highpass filtering method.
The method requires engineering judgment.
Note that some trial-and-error may be necessary to determine the optimum filtering
values in steps 1 and 5, as well as the optimum scale factor in step 2.
Recovered Data via the Indirect Method
The indirect method is implemented via program indirect.exe. The input time history is
the test data from Figure 1. The input parameters are shown in Table 2.
Table 2.
Input Parameters for Program: indirect.exe
Description
Lowpass filter at 100 Hz.
Scale Factor = 0.95.
Correct pre-shock? yes
Highpass filter at 16 Hz
2
The acceleration time history obtained via the indirect method is shown in Figure 4. The
signal oscillates about the zero baseline. The indirect method is successful thus far.
The shock response curves obtained from the resulting signal are shown in Figure 5. The
spectral curves have a consistent slope across the entire frequency domain. Furthermore,
the curves are nearly equal to one another over most of the domain.
The velocity time history is shown in Figure 6. Though imperfect, the velocity time
history has a stable oscillation about the zero baseline.
Finally, a comparison of absolute spectral curves from three methods is shown in Figure
7. The indirect method results are compared with two highpass filtering methods. The
indirect method curve has a more realistic slope than the other curves.
Conclusion
The indirect filtering method cannot recover the original acceleration signal with absolute
certainty. Nevertheless, it is a useful tool for estimating the original signal. It yields a
more realistic shock response spectrum than the brute force highpass filtering methods.
Homework
A sample acceleration time history from a linear shape charge test was given in the
previous Unit as lsc.txt. The data format is: time (sec) and accel (G). Use this data to
repeat the example in this text. Use program indirect.exe. Experiment with the input
parameters.
3
RAW DATA
2000
1500
1000
500
ACCEL (G)
-500
-1000
-1500
-2000
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
TIME (SEC)
Figure 1. Acceleration Time History, Raw Data

Ideally, the signal would oscillate about the zero baseline. An offset is
clearly present, however.
4
FRANGIBLE JOINT SHOCK TEST - RAW DATA
10000
Negative
Positive
1000
PEAK ACCEL (G)
100
10
1
10 100 1000 10000
Figure 2. Shock Response Spectrum of Raw Data

The negative and positive spectral curves clearly diverge from one another, particularly
below 500 Hz. Furthermore, each curve is unrealistically high in the low frequency
domain near 10 Hz. These characteristics are further evidence of saturation.
5
RAW DATA
5000
VELOCITY (IN/SEC)
-5000
0 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090 0.100
TIME (SEC)
Figure 3. Velocity Time History, Raw Data

The integrated signal reaches nearly 5000 in/sec, which is very unrealistic. This is
another manifestation of the saturation effect.
6
INDIRECT FILTER METHOD
2000
1500
1000
500
ACCEL (G)
-500
-1000
-1500
-2000
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
TIME (SEC)
Figure 4. Acceleration Time History, Processed via the Indirect Filter Method.
The processed acceleration time history oscillates about the zero baseline.
7
SRS Q=10 FRANGIBLE JOINT SHOCK TEST
10000
Negative
Positive
1000
PEAK ACCEL (G)
100
10
1
10 100 1000 10000
Figure 5. Shock Response Spectrum of Indirectly Filtered Data

The shock response spectra curves are approximately equal over most of the frequency
domain. The overall slope appears realistic. The indirect filter method has thus proven
successful, pending verification of the velocity time history.
8
100
50
VELOCITY (IN/SEC)
-50
-100
0 0.010 0.020 0.030 0.040 0.050 0.060 0.070 0.080 0.090 0.100
TIME (SEC)
Figure 6. Velocity Time History of Indirectly Filtered Data

The velocity time history now oscillates about the zero baseline. The velocity prior to
time zero has a spurious ramp. Nevertheless, the overall character of the velocity time
history is satisfactory.
9
ABSOLUTE SRS Q=10 FRANGIBLE JOINT SHOCK TEST
10000
100 Hz HP Filter
50 Hz HP Filter
Indirect Filter Method
1000
PEAK ACCEL (G)
100
10
1
10 100 1000 10000
Figure 7. Three Processing Methods
Three processing methods are shown in Figure 7.

Both of the highpass filtering methods yield spectral curves with unrealistically steep
slopes below 100 Hz.
The indirect method, however, yields a spectral curve with a slope consistent across the
entire frequency domain.
10
Unit 27. Sine Identification and Removal
By Tom Irvine
________________________________________________________________________
Introduction
Previous Units have given methods for identifying the amplitude and frequency of sine
functions in measured data.
One method is the direct inspection. Another method is the Fourier transform.
A third method is a curve-fit method that is performed in the time domain. This method
is useful for identifying the amplitude, frequency, and phase angle of a sine function. It
is more accurate than either the direct inspection method or the Fourier transform.
Nevertheless, the curve-fit method requires initial estimates of the frequencies. Thus, it
can be used in conjunction with the direct count or Fourier methods.
The time-domain curve-fit method is also useful for removing spurious sine signals from
data. These spurious signals may be due to ground loops, eddy currents, and other
electrical interference effects. In addition, analog tape records often introduce spurious
sine signals.
The purpose of this Unit is to demonstrate the time-domain curve-fit method.
Method
The method is semi-automatic. The user must estimate the sine frequency, or
frequencies. A computer program then uses random number generation to determine the
best fit in terms of amplitude, frequency, and phase.
The steps are in summary:
1. The user estimates one or more sine frequencies. The user also specifies the
number of iterations.
2. The computer program reads in the measured data signal and calculates its
standard deviation amplitude.
3. The program generates a trial amplitude, frequency, and phase angle using a
random number routine.
4. The program generates a trial signal based on the parameters in step 3.
5. The program subtracts the trial signal from the measured data signal and
computes the error.
6. The program repeats steps 2 through 5 many times per the user's initial
instructions.
7. The program then selects the amplitude, frequency, and phase angle which
gave the lowest error.
1
The above steps are implemented in program sinefine.exe.
As an important note, this method works best when the measured data has a sine function
with fairly constant amplitude. Also, it is appropriate for small segments of data, with
less than 10,000 data points. These guidelines are not absolute requirements, however.
Example 1
Recall the sine.txt file from Unit2A. Again, this is measured data taken on a floor
adjacent to a wafer polishing machine in a semiconductor facility. The measured time
history is shown in Figure 1.
SINE.TXT - MEASURED DATA FROM SEMICONDUCTOR FACILITY
0.2
0.1
ACCEL (G)
-0.1
-0.2
14.0 14.1 14.2 14.3 14.4 14.5 14.6
TIME (SEC)
Figure 1.
The dominant frequency is approximately 50 Hz, as estimated by counting the peaks and
then divided by the duration. Use the curve-fitting method to determine the amplitude,
frequency, and phase angle.
2
The following estimate is thus obtained using program sinefind.exe.
y ( t ) = A sin ( 2π f t − φ )
where
A = 0.0812 G
f = 48.381 Hz
φ = 6.173 radians
(1)
The input parameter to program sinefind.exe were:

40000 trials per frequency,
nominal frequency = 50 Hz with a tolerance of ±5 Hz.
Equation (1) is shown superimposed on the sine.txt time history in Figure 2. The
agreement is good. The error signal is shown in Figure 3.
0.2
Synthesized Time History
Measured Data
0.1
ACCEL (G)
-0.1
-0.2
14.0 14.1 14.2 14.3 14.4 14.5 14.6
TIME (SEC)
Figure 2.
3
ERROR SIGNAL FOR SINE.TXT
0.2
Error Signal = Measured Data - Synthesized Time History
0.1
ACCEL (G)
-0.1
-0.2
14.0 14.1 14.2 14.3 14.4 14.5 14.6
TIME (SEC)
Figure 3.
Example 2
Note that the time history in Figure 1 has a slight beat frequency effect. Thus, repeat the analysis
using two frequencies.
The following estimate is obtained using program sinefind.exe.
2
y ( t) = ∑ A i sin ( 2π f i t − φ i )
i= 1
where
A1 = 0.0812 G A 2 = 0.0179 G
f 1 = 48.381 Hz f 2 = 46.176 Hz
φ 1 = 6.173 radians φ 2 = 3.394 radians
(2)
The input parameter to program sinefind.exe were:
40000 trials per frequency,
nominal frequency 1 = 50 Hz with a tolerance of ±5 Hz,
nominal frequency 2 = 50 Hz with a tolerance of ±10 Hz.
4
Equation (2) is shown superimposed on the sine.txt time history in Figure 4. The
agreement is very good. The error signal is shown in Figure 5.
0.2
Synthesized Time History, with 2 frequencies
Measured Data
0.1
ACCEL (G)
-0.1
-0.2
14.0 14.1 14.2 14.3 14.4 14.5 14.6
TIME (SEC)
Figure 4.
ERROR SIGNAL FOR SINE.TXT FOR SYNTHESIS WITH 2 FREQUENCIES
0.2
Error Signal = Measured Data - Synthesized Time History
0.1
ACCEL (G)
-0.1
-0.2
14.0 14.1 14.2 14.3 14.4 14.5 14.6
TIME (SEC)
Figure 5.
5
Obviously, this synthesis process could be repeated for additional frequencies in order to
reduce the error.
Example 3
Again, measured data often contains spurious signal due to electromagnetic interference.
Data recorded in analog form on magnetic tape is particularly vulnerable. Consider the
time history in Figure 6, which was measured during a rocket fairing separation test. The
separation event consists of two pulses. The first is at time zero. The second is at 0.080
seconds.
The overall signal-to-noise ratio in Figure 6 is marginal. A “noise floor” is evident both
before and after the event. The noise floor appears to consist of broadband random noise
superimposed on a 30 Hz sinusoidal signal. The task is to remove the spurious sinusoidal
signal in order to clarify the data. Program sinefind.exe is used to perform this task. The
synthesized signal is shown in Figure 7. The error signal is shown in Figure 8. The error
signal is equal to the raw signal in Figure 6 minus the synthesized signal in Figure 7. The
error signal is retained as an estimate of the true measured signal.
RAW ACCELERATION TIME HISTORY FAIRING SEPARATION TEST
300
30 Hz spurious noise signal
200
100
ACCEL (G)
-100
-200
-300
-0.1 0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 6.
6
SYNTHESIZED SIGNAL FAIRING SEPARATION TEST
300
200
100
ACCEL (G)
-100
-200
-300
-0.1 0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 7.
PROCESSED ACCELERATION TIME HISTORY FAIRING SEPARATION TEST
300
30 Hz spurious noise signal removed.
200
100
ACCEL (G)
-100
-200
-300
-0.1 0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 8.
7
The synthesized time history in Figure 7 is given by the following equation, with
coefficients found via program sinefind.exe.
2
y ( t) = ∑ A i sin ( 2π f i t − φ i )
i= 1
where
A1 = 4.66 G A 2 = 0.790 G
f 1 = 30.314 Hz f 2 = 29.619 Hz
φ 1 = 5.612 radians φ 2 = 3.522 radians
(3)
The absolute shock response spectra of the raw signal and the processed signal are shown
in Figure 9. The need for accurate spectra drives the sine removal effort.
SHOCK RESPONSE SPECTRA Q=10

ACCELERATION TIME HISTORY FAIRING SEPARATION TEST
1000
Processed
Raw
PEAK ACCEL (G)
100
10
1
10 100 1000 10000
Figure 9.
8
Both the processed signal in Figure 8 and the spectral comparison in Figure 9 reveal that
further refinement could be made.
Homework
Use program sinefind.exe for the following problems. A good number of iterations per
frequency is 40000.
A good tolerance for the frequency estimate is 5 Hz or 10 Hz.
Note that program sinefind.exe runs rather slowly. A few minutes may be required for
each frequency, depending on the number of data points and iterations.
1. Recall file trailer.txt from Unit2B. This is actual data measured on a semi-trailer
during a proving ground test. Use program poweri.exe to calculate the power spectral
density. Plot the result. Identify the dominant frequency below 100 Hz.
2. Take the frequency determined in problem 1 as an estimate. Use two estimates of this
value within program sinefind.exe. Plot the raw signal and the synthesized signal.
3. The time history in Figure 6 is given in file fairing.txt. Repeat Example 3 as given in
the text but use three or four frequencies near 30 Hz to achieve a more thorough
removal of the spurious signal. Perform a shock response spectra comparison to
verify the removal. Use program qsrs.exe for the spectral calculation.
9
Shock Response Spectrum Synthesis via Wavelets
By Tom Irvine
______________________________________________________________________
Introduction
Mechanical shock can cause electronic components to fail. Crystal oscillators may
shatter, for example. Components such as DC-to-DC converters can detach from circuit
boards. Housings and other mechanical parts may develop fatigue cracks, even those
made from metal.
Mechanical shock can cause temporary malfunctions in addition to hard failures.
Mechanical relays can experience chatter, for example. Computer hard disk drives may
lock up, thereby requiring a re-boot.
Components should thus be subjected to shock tests in order to verify their design
integrity.
There are several common shock test methods. One method is to mount the component
on a shaker table and then subject it to a base excitation pulse.
The specification for a base excitation shock may be in the form of a shock response
spectrum. If so, a time history must be synthesized to satisfy the specification.
The purpose of this report is to present a synthesis method using wavelets.
Wavelet Theory
Again, a shock response spectrum can be met using a series of wavelets. The wavelets
are synthesized into a time history on a control computer. The control computer applies
the time history to an electromagnetic shaker. The shaker then applies the shock pulse to
the test item. The control computer then verifies the resulting shock pulse.
1
The equation for an individual wavelet is:
 0 , for t < t dm


  2πf m   Nm 
 A m sin  ( t − t dm ) sin [2 πf m ( t − t dm )] , for t dm ≤ t ≤  t dm + 
Wm ( t ) =   Nm   2f m 

  Nm 
 0 , for t > t dm + 
  2f m 

where
Wm ( t ) is the acceleration of wavelet m at time t,

Am is the wavelet acceleration amplitude,
fm is the wavelet frequency,
Nm is the number of half-sines in the wavelet,
t dm is the wavelet time delay.
(1)
Note than Nm must be an odd integer and must be at least 3.
The total acceleration at any time t for a set of n wavelets is
n
&x&( t ) = ∑ Wm ( t )
m= 1
(2)
Selection of the proper wavelet parameters to fulfill a given shock response spectrum is a
trial-and-error process. Prior experience is a valuable guideline. Note that the wavelet is
designed to have zero net velocity and zero net displacement.
A time history of sample wavelet is shown in Figure 1.
2
SAMPLE WAVELET AMP = 1.34 G, FREQ=100 Hz, 19 HALF-SINES
1.5
1.0
0.5
ACCEL (G)
-0.5
-1.0
-1.5
0 0.02 0.04 0.06 0.08 0.10
TIME (SEC)
Figure 1.
3
Example
Consider the specification in Table 1.
Table 1.
MIL-STD-810E Crash
Hazard, SRS Q=10
Natural Peak
Frequency Acceleration
(Hz) (G)
10 9.4
80 75
2000 75
Notes:
1. Require that both the positive and negative spectral curves meet the specification.
2. Assume tolerance bands of ± 3 dB.
3. Use 1/6 octave spacing.
4. Allow a 0.400 second duration.
Synthesize an acceleration time history which satisfies the specification via wavelets.
Optimize the time history to minimize the peak acceleration, velocity, and displacement
values.
A synthesis can be performed using the approach in Table 2.
4
Table 2. Synthesis Steps
Step Description
1 Generate a random amplitude, delay, and half-sine number for
each wavelet. Constrain the half-sine number to be odd. These
parameters form a wavelet table.
2 Synthesize an acceleration time history from the wavelet table.
3 Calculate the shock response spectrum of the synthesis.
4 Compare the shock response spectrum of the synthesis to the

specification.
Form a scale factor for each frequency.
5 Scale the wavelet amplitudes.
6 Generate a revised acceleration time history.
7 Repeat steps 3 through 6 until the SRS error is minimized or

until an iteration limit is reached.
8 Calculate the final shock response spectrum error.

Also calculate the peak acceleration values.
Integrate the signal to obtain velocity, and then again to obtain
displacement.
Calculate the peak velocity and displacement values.
9 Repeat steps 1 through 9 many times.
10 Choose the waveform which gives the lowest combination of

SRS error, acceleration, velocity, and displacement.
The method in Table 2 is a rough outline.

For the example, the time delays are arranged in a "reverse sine sweep" manner. In other
words, the highest frequency component has zero delay. Each successive wavelet
moving downward in frequency has a progressively longer delay. The delay step is
proportional to the wavelet frequency.
The two wavelet components having the lowest frequencies, however, are allowed to
begin at time zero.
This synthesis method produced the wavelet series shown in Table 3. The peak time
history values are shown in Table 4.
The acceleration, velocity, and displacement time histories are shown in Figures 2, 3, and
4, respectively. The shock response spectrum is shown in Figure 5. The spectral curves
satisfy the tolerance bands.
5
Table 3. Wavelet Parameters
Frequency Amplitude Half-sines Delay

(Hz) (G) (sec)
10.0 -1.84 7 0.000000
11.2 1.80 7 0.000000
12.6 0.51 5 0.155000
14.1 0.10 7 0.137990
15.9 0.44 7 0.122840
17.8 0.00 9 0.109350
20.0 4.85 11 0.097320
22.4 6.35 13 0.086610
25.2 4.23 15 0.077070
28.3 2.78 17 0.068570
31.7 2.40 17 0.060990
35.6 3.50 17 0.054240
40.0 4.11 17 0.048230
44.9 3.76 17 0.042870
50.4 4.81 17 0.038100
56.6 5.88 17 0.033850
63.5 5.93 17 0.030070
71.3 6.93 17 0.026690
80.0 6.91 17 0.023680
89.8 6.74 17 0.021010
100.8 7.34 17 0.018620
113.1 6.73 17 0.016500
127.0 7.23 17 0.014600
142.5 6.85 17 0.012910
160.0 6.52 17 0.011410
179.6 7.28 17 0.010070
201.6 6.72 17 0.008880
226.3 6.94 17 0.007820
254.0 7.06 17 0.006870
285.1 6.50 17 0.006030
320.0 7.33 17 0.005280
359.2 6.81 17 0.004610
403.2 6.79 17 0.004010
452.5 7.26 17 0.003480
508.0 6.50 17 0.003000
6
Table 3. Wavelet Parameters (continued)
Frequency Amplitude Half-sines Delay

(Hz) (G) (sec)
570.2 7.24 17 0.002580
640.0 7.02 17 0.002210
718.4 6.52 17 0.001870
806.3 7.40 17 0.001570
905.1 6.63 17 0.001310
1015.9 7.01 17 0.001070
1140.4 7.25 17 0.000860
1280.0 6.44 17 0.000670
1436.8 7.21 17 0.000510
1612.7 6.58 17 0.000360
1810.2 4.73 17 0.000220
2031.9 6.67 17 0.000000
Table 4.
Peak Values of Synthesized Time History
Parameter Maximum Minimum
Acceleration (G) 13.2 -12.9
Velocity (in/sec) 25.5 -23.3
Displacement (inch) 0.344 -0.122
Note that the reverse sine sweep method tends to produce lower acceleration, velocity,
and displacement values than a random delay method.
Minimization of these parameters is highly desirable due to shaker table limitations.
7
SYNTHESIS TO SATISFY MIL-STD-810E CRASH HAZARD SPECIFICATION
20
15
10
5
ACCEL (G)
-5
-10
-15
-20
0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 2.
VELOCITY TIME HISTORY

30
20
VELOCITY (IN/SEC)
10
-10
-20
-30
0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 3.
8
DISPLACEMENT TIME HISTORY
0.4
0.3
DISPLACEMENT (INCH)
0.2
0.1
-0.1
-0.2
-0.3
-0.4
0 0.1 0.2 0.3 0.4
TIME (SEC)
Figure 4.

1000
Negative
Positive
PEAK ACCEL (G)
100
10
1
10 100 1000 2000
Figure 5.
9
Homework
Consider the specification in Table 2.
Table 2. SRS Q=10

Natural Peak
Frequency Acceleration
(Hz) (G)
10 10
50 50
500 50
1000 100
2000 100
Notes:
1. Require that both the positive and negative spectral curves meet the
specification.
2. Assume tolerance bands of ± 3 dB.
3. Use 1/6 octave spacing.
4. Allow a 0.400 second duration.
Synthesize a time history pulse to satisfy the specification using program jsynth.exe. Plot
the acceleration, velocity, displacement, and shock response spectra. Optimize the
waveform by trying 300 cases.
10
Unit 2a
1. Calculate the RMS value for a constant amplitude signal x(t) = A.
1 T
X 2 = lim T → ∞ [a ( t )]2 dt
∫
T 0
1 T 2
X 2 = limT →∞
T ∫0
A dt
A2 T
X 2 = limT →∞
T ∫0
dt
2 A2
X = limT →∞ [T − 0]
T
X2 = A2
X RMS = A
2. Calculate the standard deviation for the signal in problem 1.
STD DEV^2 = RMS^2 – MEAN^2
The mean and the RMS are both equal to A. Thus the standard deviation is zero
1
3. A signal consists of two components:
 2πt 
x ( t ) = A + B sin  
 T 
what is the standard deviation?
2
2 1 T  2πt  
X = limT →∞
T 0 ∫
A + B sin 
 T 
  dt
2 
2 1 T  2  2πt    2πt   
X = limT →∞
T 0
 ∫
A + 2AB sin  
 T  
+ B sin 
 T 
  dt
 
1 T  2 2 B2
 2πt  B  4πt  
X 2 = limT →∞ ∫  A + 2 AB sin   + − cos    dt
T 0  T  2 2  T  

T
1  2 2TAB 2 TB2  4πt  
 2πt  B
X 2 = limT →∞  A t − cos  T  2 + t − sin  
T 2π   8π  T  
 0
1  2 B2 
X 2 = limT →∞ A T + T
T 2 
 
1
X 2 = A 2 + B2
2
Mean = A by inspection
STD DEV^2 = RMS^2 – MEAN^2
1
STD DEV^2 = A 2 + B2 − A 2
2
1
STD DEV^2 = A 2 + B2 − A 2
2
2
1 2
STD DEV^2 = B
2
1 2
STD DEV = B
2
1
STD DEV = B
2
1
STD DEV = B
2
4. Plot the data file sine.txt. This is measured data taken on a floor adjacent to a wafer
polishing machine in a semiconductor facility. It is a velocity time history with
amplitude in units of (in/sec). It is not a pure sine signal, but measured data never
really is. What is the dominant frequency?
48.3 Hz
5. The program maxfind.exe generates descriptive statistics for a time history with two
columns: time(sec) and amplitude(units). Copy this file to same directory as the
sine.txt file. Then type:
maxfind sine.txt
What is the average, standard deviation, RMS, and kurtosis values of this signal?
Note that the maxfind.exe program assumes amplitude units of G, but this is irrelevant.
Substitute in/sec for G in your notes.
3
6. Pogo is a type of sinusoidal vibration that occurs in certain rocket vehicles due to
combustion instability. Read tutorial pogo.pdf to learn more about this effect. This
tutorial is not concerned with descriptive statistics per se, but it gives a real-world
example of sinusoidal vibration.
Unit 2b
1. File trailer.txt is measured acceleration data from a semi-trailer tested at an

automotive proving ground oval. The data is not purely sinusoidal, but is reasonably
so. First, determine the frequency in Hertz. Then use equation (5) to estimate the
peak displacement. Remember to convert G to in/sec2, as part of the calculation.
You may use a hand calculation or Excel.
Fn = 18.1 Hz
Amplitude = 0.49 G, but peak value is 1.2 G
2. Check your calculation from problem 1 using program sine.exe. Use the pure sine
option in this program.
4
(again peak accel is 1.2 G).
3. Use sine.exe for this problem. A shaker has a displacement limit of 1 inch peak-to-
peak, or 0.5 inch zero-to-peak. What is the maximum velocity at a frequency of 10
Hz?

Shock and Vibration Response Spectra Study Course - Tom Irvine

Uploaded by

Copyright:

Available Formats

You might also like

Shock and Vibration Response Spectra Study Course - Tom Irvine

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Shock and Vibration Response Spectra Study Course - Tom Irvine

Uploaded by

Copyright:

Available Formats

SHOCK AND VIBRATION RESPONSE SPECTRA COURSE

Unit 1A. Natural Frequency Calculation

Some Simple Physics

Table 1. Energy Relationships for an Oscillating Mass

The natural frequency ω n for a pendulum is

Note that the period T is

Benefits of Natural Frequency

Drawbacks of Natural Frequency

Table 2. Methods of Excitation

Figure 4. Cypress Viaduct Failure

Ct is an empirical factor given in Table 3.

Table 3. Empirical Factor Ct

Earth's Natural Frequency

The Earth experiences seismic vibration.

Golden Gate Bridge

1. Earthquakes, primarily originating on the San Andreas and Hayward faults

Mode Type Period of Natural

Transverse 18.2 0.055

Circuit Boards are often modeled as Single-degree-of-freedom systems. The fundamental

Table 5. Sample Circuit Boards

Avionics Component Fundamental

A sample acceleration time history is shown in Figure 8.

1. A particular circuit board can be modeled as a single-degree-of-freedom system. Its weight

3. A rocket vehicle is carried underneath an aircraft. It experiences an initial displacement

5. A flagpole is made from steel pipe.

The height is 180 inches.

Determine the fundamental lateral frequency. Use program: beam.exe

6. Read References 1 and 2.

1. T. Irvine, Tacoma Narrows Bridge Failure, Vibrationdata Publications, 1999.

Free Vibration Equation

&y&( t ) = A exp( − ξ ω n t )sin ( ω d t + φ) (1)

For many systems, the damping ratio ξ < 0.25. Thus,

Equation (1) simplifies to

&y&( t ) ≈ A exp ( − ξ ω n t )sin ( ω n t + φ) (4)

The envelope equation in Figure 2 is

E(t ) = 1 exp( − 3.1 t ) (6)

Compare equation (5) with (6).

The natural frequency is 10 Hz.

The sample problem thus has an amplification factor of Q = 10.

4. Review tutorials damping.pdf and dcross.pdf.

Consider a single-degree-of-freedom system undergoing excitation from an applied force or base

The mean value equation becomes

where c is the number of cycles.

Apply a trig identity.

X RMS ≈0.707 A (13)

Note that σ2 is called the variance.

The mean-square value is

The standard deviation equation is still

An equivalent formula for the standard deviation is

Note that the kurtosis value appears dimensionless.

1. Calculate the RMS value for a constant amplitude signal x(t) = A.

2. Calculate the standard deviation for the signal in problem 1.

3. A signal consists of two components:

what is the standard deviation?

Note that each amplitude in this unit is taken as zero-to-peak.

The velocity x&( t ) is obtained by taking the derivative.

x&(t ) = ω X cos( ω t ) (2)

The acceleration &x&( t ) is obtained by taking the derivative of the velocity.