Professional Documents
Culture Documents
What - Is - Random Signal PDF
What - Is - Random Signal PDF
/
H
z
)
Acceleration Profile
Demand
Control
Figure 3 Amount of power per unit (density) of frequency (spectral) as a function of the
frequency
The PSD demonstrates how hard the shaker is working. It doesnt give any direct
information about the forces experienced by the DUT. This is important to remember.
Since the PSD is the result of an averaging method that produced the statistical
approximation of the spectrum, an infinite number of real-time waveforms could have
generated such a PSD. Thus, at any time during a test, it is impossible to know
specifically from the PSD what forces the DUT is experiencing. The need for the PSD is
that it aids the tester in making an appropriate test profile for the shaker that will come
close to real-life vibrations that the DUT will experience.
The idea that an infinite number of real-time waveforms could generate a
particular PSD can be seen from the following graphs (Figures 4 through 7) produced from
data collected at VRC on June 28 and 30, 2005. Note that the PSD spectra formed from
the data in both cases is exactly the same, yet generated from different waveforms.
0 100 200 300 400 500 600 700 800
-15
-10
-5
0
5
10
15
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n
(
G
)
Acceleration Waveform
Ch1
Figure 4 Waveform for Body and IP Profile Lightbulb-4 for trial 2005Jun28 1330
0 100 200 300 400 500 600 700 800
-15
-10
-5
0
5
10
15
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n
(
G
)
Acceleration Waveform
Ch1
Figure 5 Waveform for Body and IP Profile Lightbulb-4 for trial 2005Jun30 1110
5 10 100 1000
-5
1x10
-4
1x10
-3
1x10
-2
1x10
-1
1x10
Frequency (Hz)
A
c
c
e
l
e
r
a
t
i
o
n
(
G
/
H
z
)
Acceleration Profile
Demand
Control
Ch1
Figure 6 PSD spectrum for trial 2005Jun28 - 1330
5 10 100 1000
-5
1x10
-4
1x10
-3
1x10
-2
1x10
-1
1x10
Frequency (Hz)
A
c
c
e
l
e
r
a
t
i
o
n
(
G
/
H
z
)
Acceleration Profile
Demand
Control
Ch1
Figure 7 PSD spectrum for trial 2005Jun30-1110
Probability Density Function (PDF) Graphs:
An examination of the acceleration waveforms (Figures 4 and 5) will indicate that
much of the random vibration acceleration values are nearly the same ( 5 G). However,
some of the acceleration values are quite large compared to the normal values. To help
illustrate the rangeof acceleration values, the Power Spectrum Density is converted into
an amplitude probability density graph (PDF) (similar to Figure 8). Notice how much of the
acceleration values fall near the average acceleration value (represented by 0 Sigma). In
fact, much of the vibrations in the real-world approximate a Gaussian distribution, that
is, a distribution in which the vast majority of the data is in the 3 sigma range.
-5 -4 -3 -2 -1 0 1 2 3 4 5
10
-5
10
-4
10
-3
10
-2
10
-1
10
0
Sigma
P
r
o
b
a
b
i
l
i
t
y
D
e
n
s
i
t
y
Probability Density Function for K = 3
2005Jun30 -1117k3
Figure 8 Probability density for a Light bulb test using Gaussian distribution (k=3)
There are some real-life cases in which there are more high acceleration values
than a Gaussian distribution would show. Unfortunately, most modern techniques covert
the PSD into a PDF that assumes that the majority of the data is in the 3 sigma range
(i.e. Gaussian distribution). This assumption removes from the real-time data the
accelerations that were of extremely large magnitude. These higher accelerations, which
are present in real-life scenarios, are omitted from the probability density graphs of all
those who use the traditional Gaussian distribution method. Consequently, present-day
methods of random testing are unrealistic because they fail to take into account these
higher level accelerations. Furthermore, random testing with Gaussian distribution will
result in a longer time-period to test the product to failure because the higher
accelerations responsible for failure have been omitted. Therefore, random testing, for all
its advantages over traditional sine testing, has its own disadvantages, and a better
method of testing products would prove valuable. Vibration Research Corporation has
developed ways to improve upon traditional Gaussian distributions in random testing.
This new patent-pending technique is called Kurtosion.
Overall, random testing is an excellent tool for vibration testing. It is more
efficient, more precise, and more realistic than sine testing. And, although random
testing is not perfectly realistic and can be improved upon, testing industry ought to use
random testing extensively in their testing procedures.
Field-Data Replication (FDR):
Development in vibration research has resulted in newer methods that come
closer and closer to real-life data replication. Random testing is a great improvement on
sine testing but still does not perfectly represent what happens in real-life. In response to
this, vibration research companies developed Sine-on-Random testing which
overlapped sine spectra with random spectra. The goal of this testing is to include some
peaks that occur in real-life scenarios into the random spectra. This method has been
somewhat successful in bringing tests closer to reality.
More recent development has included a method of recording real-life data and
turning it directly into a spectrum to be used in lab. This method, called field-data
replication (FDR), is very helpful in accurately representing in a test setting what is
happening on the field. This is like shaped random in a way, because the spectrum is
the same as seen in the real application. This method is good, but also has its
shortcomings. It is difficult to find a representative waveform, especially in aerospace
applications. When one is obtained, it is representative of a particular situation of the
product. Unfortunately, it is probably not representative of the entire life of the product.
Gaussian Distribution vs. Kurtosis Distribution:
One final helpful distinction is that relating to the probability distributions of a
DUTs vibratory accelerations. As mentioned earlier, a probability distribution shows the
reader how the data points compare with the average data point. Most of the data points
will center near the average with a number of outliers. Generally, as more data points are
collected the probability distribution forms a nice smooth bell-shaped curve.
Gaussian distribution is the normal probability distribution of random data. The
probability distribution curve takes on the classic smooth bell-shaped curve. Consider
the Probability Density Function (PDF) graph shown below for a set of data with
Gaussian distribution.
-5 -4 -3 -2 -1 0 1 2 3 4 5
10
-5
10
-4
10
-3
10
-2
10
-1
10
0
Sigma
P
r
o
b
a
b
i
l
i
t
y
D
e
n
s
i
t
y
Probability Density Function for K = 3
2005Jun30 -1117k3
Figure 9 Probability density for a Light bulb test using Gaussian distribution (k=3)
With the use of statistics, one can find a number of interesting things about a set
of collected data. For example one can easily compute the mean and the standard
deviation of a data set statistical concepts familiar to most people communicating the
average of the data set and the range in which most of the data points fall. But a less
familiar statistical concept is the kurtosis of the data set. Kurtosis is a measure of the
peakyness of the probability distribution of the data. For example, a high kurtosis
value indicates the data is distributed with some very large outlier data points, while a
low kurtosis value indicates most data points fall near the mean with few and small
outlier data points. In the previous figure (Figure 9), the data set has a kurtosis value of 3
(Gaussian distribution) and is a smooth curved distribution with few large amplitude
outliers. However, Figure 10 shows a data set with a kurtosis value of 5. Note how the
tails extend further from the mean (indicating large number of outlier data points).
The contrast between the PDFs of a Gaussian distribution and a higher kurtosis
distribution is clearly seen in Figure 11.
-8 -6 -4 -2 0 2 4 6 8
10
-5
10
-4
10
-3
10
-2
10
-1
10
0
Sigma
P
r
o
b
a
b
i
l
i
t
y
D
e
n
s
i
t
y
Probability Density Function for K = 5
2005Jun30 - 1320k5
Figure 10 Probability density for Light bulb test using Kurtosis Control (k=5)
-8 -6 -4 -2 0 2 4 6 8
10
-4
10
-3
10
-2
10
-1
10
0
Multiple of Sigma
P
r
o
b
a
b
i
l
i
t
y
D
e
n
s
i
t
y
Probability Density Function
Kurtosis = 3
Kurtosis = 7
Figure 11 A comparison of kurtosis values 3 and 7. Note how the higher kurtosis value includes
higher sigma values (higher peak accelerations).
Therefore, the fundamental difference between a Gaussian distribution and a
kurtosis distribution is that, although the two data sets may have the same mean, standard
deviation and other properties, yet the Gaussian data set has its data points closely
centered on the mean while the kurtosis distribution has larger tails further from the
mean.
Implications of Kurtosis Distribution to Vibration Testing:
The obvious question is what significance does kurtosis distribution and
Gaussian distribution have to vibration testing? Present-day methods of random testing
assume a Gaussian mode of distribution of random data. Modern controllers run random
vibration tests with the majority of the RMS values near the mean RMS level, thus
vibrating the product only for a short time at peak RMS values. In fact, a Gaussian
waveform will instantaneously exceed three times the RMS level only 0.27% of the time.
When measuring field data, the situation can be considerably different, with amplitudes
exceeding three times the RMS level as much as 1.5% of the time. This difference can be
significant, since it has also been reported that most fatigue damage is generated by
accelerations in the range of two to four times the RMS level.
4
Significantly reducing the
amount of time spent near these peak values by using a Gaussian distribution can
therefore result in significantly reducing the amount of fatigue damage caused by the test
relative to what the product will experience in the real world. Gaussian distribution,
therefore, is not very realistic.
A better method of testing products than using the Gaussian distribution of data is
to adjust the distribution of data to more closely fit the real-world data by adjusting the
kurtosis level. The difference between the Gaussian distribution and a higher kurtosis
value is simply the amount of time spent at or near the peak levels. Adjusting the
kurtosis level to match the measured field level will result in a more realistic test.
A latest modification in random vibration testing is a closed-loop method of
kurtosis control, developed by Vibration Research Corporation (patent-pending). This
method will permit the adjustment of the kurtosis levels while maintaining the same
testing profile and spectrum attributes. With this new technique, in addition to the
standard random test PDF and RMS parameters, a kurtosis parameter is now defined to
produce a test in the lab. This is similar to current random tests but is one step closer to
the vibrations measured in the field. This kurtosis parameter can be easily measured from
field data in the same manner as the RMS and PDF are currently determined from field
data.
1
Tustin, Wayne. Random Vibration and Shock Testing. Equipment Reliability
Institute, Santa Barbara, CA, 2005, pg 205.
2
Ibid, pg 234-235.
3
Ibid, pg 224.
4
Connon, W.H., Comments on Kurtosis of Military Vehicle Vibration Data,
Journal of the Institute of Environmental Sciences, September/October 1991, pp.
38-41.