Faults Detection and Failures Prediction Using

Vibration Analysis
Tristan Plante, Ashkan Nejadpak, and Cai Xia Yang
University of North Dakota, Grand Forks, ND

Abstract — In industrial applications, the uptime of machines different signatures yet, at the same time, patterns emerge
can be enhanced through equipment monitoring. This minimizes suggesting a specific problem within the system. Over time,
the risks of unpredicted failures and consequent plant outages. certain patterns can become more evident suggesting a machine
Since all failure modes can cause an increase in machine vibrations, may fail if left uncorrected. Recognizing and categorizing these
monitoring this area is the predominant and most widely used patters before equipment failure is the objective of fault detection
method to determine equipment condition, and to predict failures. and predictive maintenance, and allows corporations and
The objective of this study is to detect faults in rotating equipment industries to reduce spending in equipment repair and
with the use of vibration analysis. A motor condition monitoring replacement. This concept correlates to the method of predictive
experiment is set up, and the motor’s operational speed is
based maintenance.
controlled by an AC motor drive. The vibration of the motor is
measured and monitored. The measured vibration data is analyzed II. BACKGROUND INFORMATION
using spectrum analysis software and a MATLAB program. The
overall vibration level is monitored, the vibration severity is Fault detection
compared with the standard severity table and is used to determine The goal of this experiment was to find evidence regarding
the condition of the motor. The specific natural frequency
vibration patterns associated with specific electric motor faults.
corresponds with which kind of fault or failure mode is identified.
Specifically, the objective of the experiment was to determine the
This information provide insight on the condition of the machine.
validity of using vibration analysis to conduct predictive based
I. INTRODUCTION maintenance. Based on previous research, there are several
common motor faults that can be identified using vibration
Scheduled and run-to-failure maintenances are commonly analysis such as imbalance, mechanical looseness, and bearing
used by industries, but both tend to incur much higher costs. faults [1]. Each fault condition’s severity and type can be
Predictive or Condition-Based Maintenance based on known assessed based on the amplitudes of the corresponding peaks as
condition is used to predict (and therefore assist in avoiding) well as their respective locations on the frequency spectrum.
unplanned equipment failures. During observation of the Additionally, certain types of faults can be determined based on
vibration modes, a relationship was found between the ranges of the location were data was recorded on the equipment. In other
natural frequency of vibration and the failure modes. By words, some faults display a higher level of severity when the
measuring and analyzing measured vibration data, engineers are accelerometer is placed on various locations of the motor. To
able to retrieve valuable information on the status of the demonstrate the effects faults have on the motor’s corresponding
equipment, predict machine failure patterns, and plan timely vibration levels, multiple tests were conducted on a three phase
maintenance operations. To progressively extend the time inverter duty induction AC motor.
between failures for the monitored equipment, the trend of
vibration in frequency domain needs to be observed frequently. In addition to above mentioned methodologies for fault
The trend of the spectrum will provide information on what type analysis, some researchers have proposed fault diagnosis
of faults are present within the system, the severity of the fault, methods based on terminal voltage and current measurements. In
and will help determine the remaining lifespan of the machine. [2] – [4] the effects of bearing and winding faults on the stator
current have been studied. It is shown that using the frequency
Understanding the concepts behind vibration data allow response analysis one can perform health monitoring or life time
engineers to detect faults and predict failures caused by prediction on the motor.
equipment defects, or deterioration such as unbalanced rotors,
bearing defects, a lack of lubrication, coupling issues, and Stages of Bearing Failure
misaligned axles before they lead to catastrophic failure. To Bearing faults are considered the most common case when
understand how vibration analysis can be used to identify motor conducting rotating machinery maintenance; however, unlike
faults, one must first understand that all mechanical systems more basic faults, bearing faults appear in four stages. During
vibrate. This vibration retains a unique signature which, given stage one, bearings operate at normal conditions, and can be
proper analysis, can tell an operator how the system is responding considered undamaged. At stage two, bearing defect frequencies
to its operating conditions. Altering these conditions may reveal begin to appear as peaks on the frequency spectrum. According

978-1-4799-8190-8/15/$31.00 ©2015
Authorized licensed use IEEE
limited to: UQAT Campus. Downloaded on May 01,2024 at 20:06:48 UTC from IEEE Xplore. Restrictions apply.
to the article “Rolling Element Bearing Analysis” by Brian Mechanical looseness
Graney and Ken Starry, bearing defect frequencies can be
calculated using equations (1) – (4) according to [5]. The Vibration patterns resulting from mechanical looseness were
amplitudes of these frequencies hint toward the conditions of the also studied. By loosening mounting bolts on the three phase
bearing, and often increase over time. As the bearing deteriorates, electric motor, the body of the motor was allowed to move more
it reaches stage three where multiples of the bearing defect freely therefore altering the motor’s vibrational patterns. On the
frequencies begin to appear as peaks in the frequency spectrum. frequency spectrum, peaks corresponding with mechanical
It is common practice to replace these bearings after reaching this looseness are considered to appear as many multiples of the
stage. Finally, at stage four, bearing defect frequencies disappear motor’s running speed. Additionally, these peaks appear on a
from the spectrum and replaced by random noise in the low raised noise floor and display random amplitudes [7]. Similarly
frequencies spectrum [6]. At this stage, the bearing is at the risk to the unbalance experiment, the three phase motor was operated
of undergoing catastrophic failure which can cost companies at 1725 RPM (30Hz), and the vibration data was recorded using a
thousands in machine repair and/or replacement. By replacing data acquisition device.
damaged bearings before they fail, industries can drastically Bearing Fault
reduce the cost of replacing vital machinery therefore outlining
the importance of predictive based maintenance on high value In addition to mechanical looseness and unbalance, the
equipment. patterns relating to bearing failures were also studied. Using a
bearing from the three phase induction motor, a defect was
III. EXPERIMENT SETUP created on one of the bearing balls. Figure 2 shows the generated
defect of the rolling element within the motor’s bearing. This
Three fault conditions were studied in a series of experiments.
damage was done by striking one of the bearing balls with a
Each of these experiments were run using a three-phase, inverter- center punch.
duty, AC electric motor. This motor, was driven using a GS1-
10P2 AC drive operated at 1725 RPM. Using an accelerometer
placed in the vertical axis of the motor, vibration data was
recorded using a NI PXI-4498 data acquisition device and was
analyzed using the Sound and Vibration Assistant software.
Unbalance
To study the unbalanced rotor condition, a steel bolt was
mounted to one end of a three phase induction motor’s flywheel.
According to previous research, a motor with an unbalanced rotor
will display a large amplitude peak at one times the running
speed [1]. Operating at 1725 RPM (30Hz), the motor’s vibration
data was recorded using a data acquisition device and graphed
using an FFT in MATLAB. Figure 1 shows the setup for this
experiment.

Fig. 2. Top view of defected bearing

The bearing defect frequencies were also calculated for the

motor’s 6203-2RS bearing. The values for which can calculated
using the following equations:
1 ) (1)

1 (2)

1 (3)

1 (4)
Table 1 shows the calculated frequencies values for each
bearing fault. Much like the previous tests, the motor was
operated at 1725 RPM (30Hz). By doing this, it was predicted the
Fig. 1. Experiment setup (Unbalanced condition)
bearing’s corresponding frequency spectrum would exhibit traits
correlating to one of four stages of bearing failure thus

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:06:48 UTC from IEEE Xplore. Restrictions apply.
supporting the validity of using vibration analysis to conduct Compared to the 30 Hz peak seen in Figure 3, which displays the
predictive based maintenance. The spectrum plots that have been motor operating under normal conditions, the 30 Hz peak of
used in this analysis is based on the algorithm proposed in [8] Figure 4 shows a substantial increase in amplitude. Additionally,
and [9]. the overall noise in Figure 4 appears to have changed. These two
observations outline can be linked to the motors unbalance
condition.
TABLE I. Bearing Fault Frequencies

Bearing Frequency Types Frequency (Hz)

Shaft Speed Frequency 28.750
Inner race defect frequency (BPFI) 142.223
Outer race defect frequency (BPFO) 87.777
Cage defect frequency (FTF) 10.972
Ball spin frequency (BSF) 57.323
Rolling element defect frequency 14.656

IV. VIBRATION DATA ANALYSIS

Ideal condition
Figure 3 displays the FFT graph for a three phase
Fig. 4. Unbalance condition
induction motor operating at 1725 RPM (30 Hz). This data
was taken to act as the healthy/ideal condition. To clarify, no Because the motor’s balanced state was the only condition
fault conditions are placed on the motor. By monitoring the altered during this experiment, it can be stated that the
condition of an ideal motor, comparisons can be made differences between Figures 3 and 4 support the presence of
between motors under fault conditions and that of an ideal unbalance within the system. The unbalance fault condition can
motor. be associated with a large increase in the operating speed
frequency as well as a raised noise floor. These two conditions
agree with the findings of previous research stated in the
introduction as well as the experiment setup section of this report.
Mechanical looseness
Figure 5 displays several peaks appearing in the low
frequency spectrum. What is most notable of these peaks is that
their frequency values are multiples of the running speed.
Additionally, these peaks possess a variety of amplitudes each
large enough to be considered hazardous to the motor’s overall
health. If allowed to operate over longer periods of time, it is
likely the motor’s lifespan will be reduced. Fortunately,
mechanical looseness is often easy to address. In this case,
simply tightening the bolts on the motor’s mounting feet resolves
the issue. Figure 3 displays the motor’s vibration data with a
secure mount. Here, several of the running frequency multiples
are no longer present, and the amplitudes of each peak are
reduced. These differences show that the presence mechanical
looseness condition appears as several multiples of the motor’s
Fig. 3. Motor under normal operating conditions (0-1k Hz) running frequency as well as a raised noise floor in the spectrum
Unbalance Fault and therefore agree with the conditions stated in [1].
Based on the data presented in Figure 4, several peaks appear
to be present. The most notable of which is the peak at 30 Hz.
The 30 Hz peak correlates to the running frequency of the
motor/s drive axle and has an amplitude of approximately 0.14 g.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:06:48 UTC from IEEE Xplore. Restrictions apply.
 motor’s vibration data has undergone substantial change and is
no longer operating in a healthy state. This suggests two
possibilities, either the data collected was inaccurate, or the
bearing could have reached stage four of bearing failure.
While it may be easy to assume the data regarding the
damaged bearing was faulty, observing the motor’s vibration data
on a higher frequency suggests stage four of bearing failure.
Figure 7 shows the motor’s vibration data from zero to ten
thousand hertz before the bearing was damaged, while Figure 8
shows the same data after the bearing was damaged.

Fig. 5. Mechanical looseness condition

Bearing Fault
Based on the results from Figure 6, it appears the spectrum
lacks data relating to the specific bearing fault frequencies stated
in table 1; however, because of the severity of the damage placed
on the bearing, it is unlikely the motor can be considered to be
operating under normal conditions.

Fig. 7. Undamaged bearing (0-10k Hz)

Fig. 6. Damaged bearing (0-1k Hz)

Several notes can be taken from Figure 6. For instance, the

vibration spectrum displays a raised noise floor as well as a
number of low amplitude peaks appearing in the higher
Fig. 8. Damaged Bearing (0-10k Hz)
frequencies; however, what is interesting to note is that none of
these frequencies appear to be whole number multiples of the The main difference between Figure 6 and 8, is that Figure 6
running speed, or the bearing fault frequencies, nor do these consists of the vibration levels at lower frequencies (0-1k Hz). As
peaks appear to correspond with the unbalance, or mechanical stated earlier, Figure 6 has a minor raised noise floor compared to
looseness fault conditions, yet it is obvious the motor’s vibration the ideal case; however, the spectrum in Figure 8, which is set in
data has been affected by the damaged bearing. Comparing these the higher frequencies (0-10k Hz), shows a noticeable increase in
results to the spectrum in Figure 3, which shows the motor overall noise. This, outlines the existence of a possible fault
operating before the bearing was damaged, shows how the condition.

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:06:48 UTC from IEEE Xplore. Restrictions apply.
 Comparing Figures 7 and 8, one may note the distinctive Transactions on Industrial Electronics, vol. 47, pp. 984-993,
difference in the overall noise vibration of the motor. Because of October 2000.
this, the bearing could be in stage four of bearing failure. Stage [5]. B. Graney and K. Starry, “Rolling Element Bearing
four of bearing failure displays large amounts of noise in higher
Analysis,” Materials Evaluation. The American Society for
frequencies; however, at this stage, bearing defect frequencies no
Nondestructive Testing Inc. 2011.
longer appear [6]. This information provides an explanation as to
why the bearing defect frequencies are not present in the [6]. M. Copping, “Vibration Analysis Reporting – Bearing
spectrum. Should this be the case, depending on the importance Failure Stages and Responses,” Reliabilityweb.com, 2015.
of the motor, it is important to replace the bearing immediately. It [7]. “Introduction to Data Analysis Using Spectral Pattern
is common practice to prevent more vital machine bearings from Recognition Techniques,” Technical Associates of
reaching stage four of bearing failure, otherwise the bearing is at Charlotte, P.C., 2015.
the risk of experiencing catastrophic failure resulting in damage
to vital machine components. [8]. D. Southwick, “Using Full Spectrum Plots,” Orbit, vol. 14,
Bently Nevada Corporation, December 1993.
V. CONCLUSION [9]. D. Southwick, “Using Full Spectrum Plots Part 2,” Orbit,
The results from each of the three tests support the use of vol. 15, Bently Nevada Corporation, June 1994.
vibration analysis in predictive based maintenance. By
comparing the vibration data for each fault case to that of a
healthy motor, shown in Figure 3, the patterns corresponding
with each fault condition are outlined. This, therefore, shows how
certain faults in rotating mechanical systems can be determined
using vibration analysis. Future research will involve the analysis
of vibration trends. In other words, this research will involve
predicting how, and when the rotating equipment will fail and
how these faults would develop over time. By developing a time
dependent procedure towards assessing motor faults, motors can
be operated for the maximum allowed time before being repaired
thus reducing overall maintenance costs.
ACKNOWLEDGEMENT

The work reported in this paper was funded by ND EPSCoR

New Faculty Start-up Award 43700-2725-UND0019805 and ND
EPSCoR Advanced Undergraduate Research Award. We thank
our colleague Dr. Zahui from UND who provided insight and
expertise that greatly assisted the research. We would also like to
thank Kevin Olson from Great River Energy/Coal Creek Station
for sharing his pearls of wisdom on equipment maintenance with
us during the course of this research.

REFERENCES
[1]. S. Pickens, “Vibration Trouble-shooting Field Guide,”
PDM Engineering.
[2]. N. Mehala, “Condition monitoring and fault diagnosis of
induction motor using motor current signature analysis,”
Ph.D. dissertation, National institute of technology
Kurukshetra, India, 2010.
[3]. W. Thomson and M. Fenger, “Current signature analysis to
detect induction motor faults,” IEEE Industry Applications
Magazine, vol. 7, pp. 26-34, 2001.
[4]. M. El Hachemi Benbouzid, “A review of induction motors
signature analysis as a medium for faults detection,” IEEE

Authorized licensed use limited to: UQAT Campus. Downloaded on May 01,2024 at 20:06:48 UTC from IEEE Xplore. Restrictions apply.