-
Acoustic Emission Fatigue Crack Monitoring of a Simulated
Aircraft Fuselage Structure
Jeremy Lucas EmbryRiddle Aeronautical University 136 Forest Lake
Blvd, Apt 901 Daytona Beach, FL 32119 269/382-8834
[email protected] Co-Author(s): Eric Hill, EmbryRiddle Aeronautical
University, Michael Marsden, Weldon Thornton
I. Introduction
As commercial aircraft are pushed into longer and longer service
by operators trying to minimize costs, fatigue crack growth in
these aircraft will become an ever increasing problem. Fatigue
crack growth has in the past led to multiple catastrophic failures,
one of the most notable incidents being the 1988 Aloha Airlines
Flight 243[1], where an eighteen foot long section of fuselage
separated from a Boeing 737-200 in flight due to corrosion fatigue.
Airlines have attempted to mitigate the risk associated with
fatigue crack growth and avoid repeating these failures by
implementing programs to periodically inspect and replace critical
parts that are likely to undergo fatigue failure. Additionally,
past research has been done on the use of acoustic emission (AE) to
provide real time, in flight monitoring of aircraft. For example,
in flight testing was used with good results on the C-130 Hercules
aircraft [2]. The basic idea behind AE nondestructive testing (NDT)
is to record the sound of crack growth as it occurs in a given
material under load. As a crack grows, it gives off energy that is
transmitted throughout the object in the form of a sound wave. By
placing piezoelectric transducers on the object, it is possible to
monitor these sound waves. Since there are almost always other
sources of sound in the object other than just fatigue crack
growth, it is necessary to classify these sounds based on what
caused them. The different sources of acoustic emission signals
give off different waveforms. By using a Kohonen self-organizing
map (SOM) neural network to classify these waveforms into
appropriate categories and understanding the physics behind each of
the sources of noise, it is possible to determine which AE hits
corresponded to each noise source. Thornton [3] and Marsden [4]
each wrote a thesis on this subject. Thornton focused his research
on using power spectrums of the AE waveforms to classify the
various acoustic emission sources. Although this approach seemed to
have had reasonable success, it was clear that the methods used to
classify the signals would need to be improved upon. Marsden
continued this research a year later using the AE waveform
quantification parameters instead of the waveforms themselves in
order to classify each hit. This led to more promising results,
although a direct comparison between the two was difficult. The
current approach being undertaken on this subject has focused on
using the AE parameters to classify the data, as this has been more
widely used in other AE applications in the past. The resulting
classification of this data seemed to be much cleaner than the
results originally obtain by Marsden. Next, the AE waveforms of
each of the classifications was examined in order to validate the
results. Finally, a source location algorithm was used to determine
the source of each of the AE hits, allowing for a second
verification of the data. This paper will detail the technical
background behind acoustic emission testing and self organizing
maps. It will then explain the setup of the current experiment, and
give the results of these research efforts. II. Technical
Background
This section gives an overview of both acoustic emission
nondestructive testing and self organizing map neural networks.
These technologies will be described as they relate to the
monitoring of fatigue crack growth in an aircraft fuselage.
Acoustic Emission Nondestructive Testing As a crack grows in a
structure, it gives off energy that propagates throughout the
structure as a sound wave. By placing an AE transducer on the
surface of an object undergoing fatigue crack growth, it is
possible to record the signals given off. However, much of the data
recorded will be background noise given off by sources
59
Nondestructive Evaluation of Aerospace Materials and Structures
II: Program Papers and Abstracts [Saint Louis, MO, May 2010]: pp
59-63. Copyright 2010, 2011, American Society for Nondestructive
Testing, Columbus, OH.
-
other than fatigue crack growth. In the current experimental
setup, this background noise will come primarily from either metal
rubbing or rivet fretting. However, due to the highly sensitive
nature of the transducers, noise will also be picked up from
sources external to the experimental setup, such as other machines
operating in the same room, the door to the room closing, and
various other sources of noise. Further, electromagnetic
interference from other surrounding electrical devices can also
cause unwanted noise in the data. The transducers are connected to
an AE data recording device, which can capture both the waveforms
and the AE parameters.
There are five primary parameters commonly used to classify an
acoustic emission hit as shown in Figure 1. The first of these is
the amplitude, which is the maximum amplitude that a signal reaches
in decibels [dB]. The duration of the signal is the signals length,
measured in microseconds [s]. The counts is the number of times the
signal rises above the preset threshold, and the energy is the
measured area under the rectified signal envelope. The fifth, and
less commonly used, parameter is the rise time. This measures the
length of time it takes from the first threshold crossing to the
maximum amplitude, again in microseconds [s]. Other acoustic
parameters will also often be used for various purposes, but these
are in fact combinations of these five primary parameters. For
example, average frequency is used herein, it is the counts divided
by the duration. The threshold used in the current research effort
has been set at 30 dB. This level is intended as a prefilter to
remove some of the unwanted low amplitude noise, while leaving the
AE data almost entirely intact. The AE data analyzer system will
also record the waveforms of the AE signals, which can be used in
conjunction with the AE parameters to classify the various signals.
One of acoustic emissions strengths is that it can be used as a
passive technique, not requiring the removal or destruction of a
part undergoing testing. This allows for monitoring of a part while
it is in service, reducing the costs associated with the testing.
However, this method can only be used while the crack is actually
growing, which will only occur when the part is placed under
stress. This means that it is only possible to monitor an aircraft
while it is actually in flight experiencing aerodynamic-loading; it
cannot be used by maintenance personnel on the ground. However, if
implemented properly, it can at least provide a warning that there
is fatigue crack growth occurring, which can then be visualized by
other nondestructive testing techniques.
Figure 1. Acoustic Emission Parameters
Self-Organizing Maps The Kohonen self-organizing map (SOM) is a
type of artificial neural network (ANN), used for classification of
data An ANN is a mathematical tool that operates on the same
principles as the human brain. These can be used to classify large
amounts of nonlinear data relatively quickly, using a series of
artificial neurons, or processing elements (PEs). These PEs use the
weighted inputs of the data to classify it into various categories.
The SOM used in this research has fairly simple architecture. The
data is entered in the input layer, with each input PE
corresponding to one of the AE parameters used for classification.
Since various input variables have different units and therefore
different ranges of data, the input values are normalized to all
fall between negative and positive one, so as to avoid an AE input
parameter with a higher range of values from overriding other AE
input parameters with smaller ranges. This operation is done
automatically by most neural network computer programs. Each of
these input PEs are connected to each output PE by a weight. This
weight is initially assigned a random
60
Nondestructive Evaluation of Aerospace Materials and Structures
II: Program Papers and Abstracts [Saint Louis, MO, May 2010]: pp
59-63. Copyright 2010, 2011, American Society for Nondestructive
Testing, Columbus, OH.
-
value between zero and one. The number of output PEs corresponds
to the number of expected classifications. For example, if it is
assumed that there are three primary sources of AE signals, three
output PEs are used. The running of a SOM is broken down into two
phases. The first is the training phase, wherein the weights
connecting the input and output PEs are constantly changing. During
the training phase, each time a data point is classified to a
specific output classification, the weight that connects the input
parameter and output classification is updated. Here the weight is
changed to be closer to the value of the input parameter, so that
other data points similar to the one just classified are more
likely to be classified into the same output classification. The
training phase typically consists of the data being run through the
SOM multiple times in order to train the network properly. The
second phase is the testing phase. Here the weights are held
constant, and the trained network classifies new input data. III.
Testing Procedure
This section outlines the experimental setup and procedure used
in this research. This experiment was performed in the Structure
and Instrumentation Lab in the Lehman Engineering and Technology
Center at the Daytona Beach campus of Embry-Riddle Aeronautical
University. Fabrication For this research, an aluminum cylinder was
used to simulate an aircraft fuselage. The current approach used
three cylinders made out of 2024-T3 aluminum and three made out of
7075-T6 aluminum. Each of these cylinders were made from a single
sheet of 0.0040 inch thick aluminum, and were 12 inches long and 12
inches in diameter, with a single lap joint secured with a line of
rivets on the back. In order to simulate a defect in an aircraft
fuselage that would likely produce fatigue crack growth, a one inch
hole was drilled into the front of the cylinder, with a triangular
notch filed out at the top. This defect was then covered from the
inside with an aluminum patch that was riveted into place,
representing a typical repair done to an aircraft in service. A
picture of the defect used in testing is shown in Figure 2. A
bladder made of PVC rubber was made, and placed inside the aluminum
cylinder in order to prevent water leakage. Two end plates were
created out of steel, with rubber attached to one side using
adhesive. The cylinder was clamped between these endplates, with
the rubber being used as a compressive seal to prevent water
leakage. Three ports were installed in the endplates. One port was
used for filling the apparatus with water, one for pressurization,
and one for a pressure gauge. The testing apparatus is shown in
Figure 3.
Figure 2. Intentional Defect Figure 3. Testing Apparatus
Testing For testing, the apparatus was filled with water. A hose
was attached to the port for pressurization, with the other end
attached to a piston. The piston was actuated by a hydraulically
operated tension/compression machine. The piston was cycled at 1
Hz, and created a cyclic pressure load that went from approximately
45 psi at a minimum to 75 psi at a maximum. Four acoustic emission
transducers were used to collect data for this experiment.
These
61
Nondestructive Evaluation of Aerospace Materials and Structures
II: Program Papers and Abstracts [Saint Louis, MO, May 2010]: pp
59-63. Copyright 2010, 2011, American Society for Nondestructive
Testing, Columbus, OH.
-
sensors were placed on a line at two inches above the defect
(Figure 3), and were positioned around the cylinder in order to be
able to determine the location of any acoustic emission signal. All
of the sensors used were 150 Hz resonant transducers, and were
attached to the metal cylinder using hot melt glue. The hot glue
acted to not only hold the transducers to the cylinder, but also to
effectively transmit the acoustic emission signals from the metal
to the transducer. The data from these transducers were captured by
a multi-channel AE analyzer that was connected to a laptop which
recorded the data. Results Due to the sensitivity of the
piezoelectric transducers, a large part of the data that was
captured during the testing of these cylinders was due to the
results of noise sources other than fatigue crack growth. Some of
the sources of noise that were present during this testing were
water entering and leaving the cylinder during each compression
cycles, the PVC bladder rubbing against the aluminum cylinder,
leakage of water form the cylinder, electromagnetic interference,
and multiple hit data. Multiple hit data occurs when the system
records two signals as though they were one signal. It was
important to filter out as much of this data as possible before
using the SOM to classify the data. The method used to do this was
a variation of the method used by Suleman, et al [5]. Each source
of acoustic emissions tends to have a narrow frequency band at
which it is emitted. Therefore, it was possible to determine, from
past research and preliminary results, the typical frequency range
of the data of interest. Using this knowledge, it was possible to
remove data that were above this average frequency. The data that
were removed was anything with an average frequency at or above
1000 kHz. Once this was accomplished, the data were run through a
SOM, using average frequency, duration and amplitude as the
classification parameters. This data were sorted into three
distinct classifications. One of these classifications was seen to
be further noise data. This noise was also removed, and the SOM run
again, this time classifying into two categories. One of these
categories was clearly a combination of metal rubbing and rivet
fretting, and the other was a combination of fatigue cracking and
plastic deformation. The duration vs. counts for these data is
shown in Figure 4, with the points labeled Mechanism 1 representing
the combination of fatigue cracking and rivet fretting. This
category was then run through another SOM, and the two categories
were separated out. The amplitude histogram of the sorted data is
shown in Figure 5.
Figure 4. Duration vs. Amplitude Initial SOM Figure 5. Amplitude
Histogram Final SOM Validation of the data occurred through two
methods. First, the waveforms of the data that were suspected to be
fatigue cracking were examined. These waveforms had the expected
shape of a clean acoustic emission signal, which can easily be
differentiated from noise waveforms. An example of a typical
waveform is shown in Figure 6. Second, the source of the acoustic
emission signals was also examined. The signals that were expected
to be from the fatigue cracking were shown to be in the same area
of the stress concentration, therefore giving another validation to
the results of this analysis.
62
Nondestructive Evaluation of Aerospace Materials and Structures
II: Program Papers and Abstracts [Saint Louis, MO, May 2010]: pp
59-63. Copyright 2010, 2011, American Society for Nondestructive
Testing, Columbus, OH.
-
Figure 6. Typical Fatigue Crack Waveform
IV. Conclusions
It is clear that acoustic emission NDT, combined with proper
data analysis , yields feasible method of in-flight monitoring of
aircraft. However, it is critical to understand the noise sources
that will be present in the data as well as the physics of the
failure that is of interest. Without a proper understanding of the
noise sources, the true data can easily be tainted by the noise,
and the classification of the data will be both extremely difficult
and probably incorrect. A misunderstanding of the physics of
failure will lead to inaccurate descriptions of the classifications
that the SOM produces. However, if each of these is understood
fully and applied correctly, this can be a powerful tool to help
reduce the threat of aircraft failure due to fatigue cracking.
More research will have to be done before this can be
implemented for monitoring of an aircraft. A potential next step
would be to incorporate information on the water pressure on the
cylinder when each signal is emitted. This could lead to more
insight into how the crack forms, and potentially to better
classifications. Finally, in-flight data collection would be the
biggest obstacle in future work. The first step would be to place
transducers on an aircraft where fatigue cracking is unlikely to
collect the noise associated with normal flight operations. After
that, it should be possible to use the knowledge gained in the lab,
combined with this noise data, to perform actual in-flight tests,
and eventually, in-flight monitoring.
V. References
1. OLone, R.G., Safety of Aging Aircraft Undergoes Reassessment,
Aviation Week and Space Technology,
2. McBride, S.L., and Machlachlan, J.W., Acoustic Emission Due
to Crack Growth, Crack Face Rubbing, and Structural Noise in CC-130
Hercules Aircraft,,
Vol. 128, Issue 20, May 16, 1988, pp. 16-18.
Journal of Acoustic Emission3. Thornton, Weldon P.
"Classification of Acoustic Emission Signals from an Aluminum
Pressure Vessel Using a
Self-Organizing Map." MSAE Thesis. Embry-Riddle Aeronautical
University, Daytona Beach, FL, 1995.
3 (April 1984), pp. 1-9.
4. Marsden, Michael L. "Detection of Fatigue Crack Growth in a
Simulated Aircraft Fuselage." MSAE Thesis. Embry-Riddle
Aeronautical University, Daytona Beach, FL, 1996.
5. Suleman, Jamil, Okur, Muhammed A., and Eric v. K. Hill.
Neural Network Fatigue Life Prediction in Aluminum from Acoustic
Emission Data. Technical Paper for Aging Aircraft Conference 2009
-- Embry-Riddle Aeronautical University, April 2009.
6. Acoustic Emission Testing (Nondestructive Testing Handbook
(3rd ed.) Vol. 6). New York: American Society for Nondestructive
Testing, Columbus, OH, 2005.
7. "Fractures&Joints.lec1." UC Santa Cruz - ITS -
Instructional Computing. Web. 18 Mar. 2010. .
8. Rovik, Christopher L. "Classification of In-Flight Cracks In
Aircraft Structures Using Acoustic Emission and Neural Networks."
MSAE Thesis. Embry-Riddle Aeronautical University, Daytona Beach,
FL, 1996.
63
Nondestructive Evaluation of Aerospace Materials and Structures
II: Program Papers and Abstracts [Saint Louis, MO, May 2010]: pp
59-63. Copyright 2010, 2011, American Society for Nondestructive
Testing, Columbus, OH.