European Association for the International Conference on Renewable Energies and Power Quality

Development of Renewable Energies, Environment (ICREPQ’12)

and Power Quality (EA4EPQ) Santiago de Compostela (Spain), 28th to 30th March, 2012

Development of Diagnosis System for Rolling Bearings Faults on Real Time Based
on FPGA

M. Kashiwagi1, C. da Costa2 and M. H. Mathias1

1
Department of Mechanical
UNESP - Univ Estadual Paulista
Campus of Guaratinguetá – Guaratinguetá, São Paulo (Brazil)
Phone:+0055 12 31232849, e-mail: [email protected], [email protected]
2
IFSP – Fedral Institute of Science Education and Technology
Campus of Cubatão – Cubatão, São Paulo (Brazil)
Phone:+0055 11 76407234, e-mail: [email protected]

Abstract. The real-time monitoring of events in an industrial Machine vibration analysis is one of the important tools
plant is an advanced technique that presents the real conditions of for rolling bearing faults identification. In fact, large
operation of the machinery responsible for the manufacturing rotating machinery systems are often equipped with
process. A predictive maintenance program includes various sensors based on mechanical quantities. In many
rotating machinery condition monitoring techniques of the situations, vibration monitoring methods are used to
machine to determine the conditions of failure. To increase the detect the presence of incipient failures in rolling
operational reliability and to reduce preventive maintenance, it is bearings [9].
necessary an efficient tool for analysis and process monitoring, in
real time, enabling the detection of incipient faults for rolling
In this context, a variety of sensors could be used to
bearings. Over the past few years there has been a major collect measurements from a rotating machine for the
technological developments related to digital system, including purpose of preemptive failure monitoring. These sensors
innovations in both hardware and software. These innovations might measure stator voltages and currents, air-gap and
enable the development of new design methodologies that take external magnetic flux densities, rotor position and speed,
into account the ease of future modifications, upgrades and output torque, internal and external temperature and case
expansions of the designed system. This paper presents a study of vibrations, among others. In addition, a real time
new design tools for embedded digital systems based on open vibration measurement and analysis instrument could
hardware architecture with reconfigurable logic. Will be monitor a variety of failures.
discussed a case study in the area of fault detection of rolling
bearings, as well as its implementation and testing.
For the purpose of detecting such fault-related signals,
many diagnostic methods have been developed so far.
These methods to identify faults in rotating machines,
Key words may involve several different types of fields of science
and technology [10- 11].
Mechanical vibration, Embedded system, Digital signal The contribution of this paper is to present new design
processing and FPGA. tools for embedded digital systems based on open
hardware architecture with reconfigurable logic for real
1. Introduction time vibration measurement and analysis. Besides,
special purpose digital signal processing (DSP) hardware,
Predictive maintenance by vibration monitoring of implemented into a low-cost field programmable gate
rotating machine is a scientific approach that becomes array (FPGA) is developed to provide on-line detection,
the new route t o the maintenance management. Rotating measurement and data analysis for rolling bearing faults.
machine, even new ones generate some level of
vibration. Small levels of ambient vibrations are 2. Rolling Bearing Faults
acceptable. However, higher levels and increasing trends
are symptoms of abnormal machine performance. Many The common faults of rolling bearings include corrosion
condition monitoring methods have been proposed for in inner race, outer race and rolling elements, fatigue
different type of rotating machine faults detection and pitting and cage damage. Any faults of inner race, outer
localization [1-8]. race and rolling elements will cause modulation
phenomenon. If there is a fault in either inner or outer

https://doi.org/10.24084/repqj10.377 545 RE&PQJ, Vol.1, No.10, April 2012

race or rolling elements, mechanical impulse with higher domain analysis can give qualitative information about
amplitude will be incurred while shaft rotating. This the machine condition. Generally a machine vibration is a
impulse will motivate the nature frequency of inner, outer stationary signal composed of random vibration and
race and rolling elements [12]. noise. Traditionally, fast Fourier transform (FFT) is used
For a particular bearing geometry, inner race, outer race to perform such analysis. If the level of random
and rolling element faults generate vibration spectra with vibrations and the noise are high, inaccurate information
unique frequency components. These frequencies, known about the machine condition is obtained. Noise and
as the defect frequencies, are functions of the running random vibrations may be suppressed from the vibration
speed of the motor and the pitch diameter to ball diameter signal using signal processing applications such FIR
ratio of the bearing. Outer and inner race frequencies are filters, averaging, correlation and convolution [14].
also linear functions on the number of balls in the bearing. According to property of convolution, the convolution in
Given the geometry of the bearing in Fig. 1, for an angular time domain is equal to the product in frequency domain,
contact ball bearing in which the inner race rotates and the so the fault signal becomes modulated signal after
outer race is stationary, the four characteristic frequencies through accelerator sensor.
is presented in Table 1. Where the outer race is fixed, fi is Some works [15], [16] have used higher order spectra to
the rotation frequency of shaft in Hertz, D is the pitch detect the fault frequency from modulated frequencies.
diameter, d is the ball diameter, α is the contact angle, Z is This approach is useful when there is a simple
the number of balls. Assume the contact between balls and modulation. However, in complex modulations it is hard
inner race and outer race is pure rolling contact [13]. to get a good result in this way.
In this case study, a diagnostic system for identification
of bearing failures was developed. In the first stage the
virtual instrument system was developed in MATLAB
software. In the second stage, the model of the virtual
instrument system was converted to VHDL code using
SIMULINK/DSP Builder software and embedded on
FPGA. The design issues was captured and presented in a
way that allowed the VHDL code to be generated
automatically

4. Virtual Instrument System

The proposed structure for the virtual instrumentation

Fig. 1- Ball bearing geometry system consists of three steps: (i) calculation of the
theoretical frequency characteristics of bearing failures;
Fault frequency of Z d (ii) analysis of the time domain to calculate the RMS
inner race fbi = fi × (1 + cos α ) value, peak value, crest factor and kurtosis, (iii) analysis
2 D
of high frequency bands (the technical envelope or
Fault frequency of Z d HFRT) in the frequency domain, identifying the
outer race fbo = fi × (1 − cos α ) frequency of failures (peak values of the spectrum). The
2 D
virtual instrumentation system was implemented in
Fault frequency of D d MATLAB. Fig. 2 shows the graphical representation of
rolling elements fbs = fi × (1 − ( )2 cos 2 α ) the virtual instrument system implemented.
2d D A computational routine calculates the theoretical fault
Fault frequency of 1 d frequency, using the geometric data of the rolling
cage f c = fi × (1 − cos α ) bearing. In the analysis in time domain, routine calculates
2 D the parameters that demonstrate the existence of faults in
rolling bearings, such as RMS value, peak value, crest
Table 1- Characteristic fault frequencies factor and kurtosis. In the analysis of high frequency
bands, the technique of the envelope is obtained through
It is important that the measurements of vibration are taken four stages: (i) filtering the signal with an elliptic filter
to diagnose the state of the machine before it enters into a (band pass) in order to eliminate some noise, (ii)
fault mode. It is further necessary to do so in real time by correction signal in the time domain, which consists of
continuously monitoring the machine variables. removing the negative part of the signal, (iii) application
of fast Fourier transform to obtain the spectrum, (iv)
3. Faults Identification Approaches identification of the five largest amplitude peaks of the
spectrum, to locate the frequency of failure.
Real time vibration measurement and analysis instrument The experimental procedure was performed so that the
is one important tool for rolling bearing faults virtual instrumentation system acquired data online, at a
identification. There are two types of analysis: time frequency of 10 kHz, at intervals of 4 seconds, comprise
domain and frequency domain. The frequency domain a total of 40,000 samples for processing and analysis by
analysis is more attractive because it can give more computer routines developed, operating in the MATLAB
detailed information about the status of the machine. Time software. The bearing NSK 6205 was used to test, with

https://doi.org/10.24084/repqj10.377 546 RE&PQJ, Vol.1, No.10, April 2012

previously known faults and generated in the laboratory by
the manufacturer NSK. The bearings were classified by the
manufacturer according to the location of the fault (inner
and outer race). Table 2 shows the geometric data of the
6205 NSK bearing, which were used for the theoretical
calculations of the frequency characteristics of bearing
failures.

Geometrical Parameters Value

Number of Ball 9
Ball diameter 7.96 mm
Inner raceway diameter 31.04 mm
Outer raceway diameter 46.72 mm

Table 2- Geometrical Parameters of the 6205 NSK bearing.

Fig. 3 – Mounting structure for bearing and related motor.

Fault Frequency Harmonic Harmonic Harmonic

(Hz) 1X 2X 3X
Inner race 135.40 270.80 406.20
Outer race 89.60 179.20 268.80
Ball 58.87 117.74 176.61
Cage 9.95 19.90 29.85

Table 3- Characteristic frequencies of bearing failures.

Fig. 4 shows the graphic of the vibration signal in time

domain and spectral components produced by bearing
defect (inner race). In the vibration spectrum presented in
Fig. 4 it is not so ease the identification of the bearing
Fig. 2 - Schematic diagram of the virtual instrument system. failure.
In the analysis of high frequency bands, the technique of
A. Laboratory Test Setup the envelope is obtained. Fig. 5 shows a spectrum of
vibration signal demodulated with the identification of
A laboratory test setup was prepared to examine the characteristic frequency of inner race fault. It can be
theoretical results, which has been focused on the seen that the desired frequency components can be
development and testing of algorithms and methods distinguished easily.
suitable for real-time detection of rolling bearing faults
identification. A test bench was created to provide a
representative model of a real situation where the bearing
could be mounted in its housing and the active forces and
velocities were similar to those found in actual situations
of the industrial environment. The vibration sensor is an
accelerometer, with a bandwidth of more than 10 kHz, the
selected data acquisition card, is a 16 bit card with a
maximum sampling rate of 1.25 MS/s.
Fig. 3 shows the laboratory test setup that illustrates the
test environment used for the development of
computational algorithms for virtual instrumentation
system proposed. Basically, it consists of a
microcomputer, a data acquisition board, an amplifier, a
signal conditioner, an accelerometer and a three-phase
induction motor.

B. Experimental results
Fig. 4 – Vibration signal and the vibration spectrum of the
Table 3 shows the fault frequencies and harmonics bearing failure (inner race).
calculated by the virtual instrumentation system,
considering the rotation axis as 25 Hz (1500 rpm) and the
geometric parameters of the NSK 6205 bearing.

https://doi.org/10.24084/repqj10.377 547 RE&PQJ, Vol.1, No.10, April 2012

 The DSP Builder is a software tool to support the design
of systems based on digital signal processing (DSP),
developed by Altera Inc., which provides a library of
special blocks for use with SIMULINK, which are
directly compiled into logic configurable to FPGA
devices [20].

Fig. 5 – Spectrum of vibration signal demodulated with the

identification of the characteristic frequency.

5. Modeling Virtual Instrument System in Fig. 6 – Schematic diagram of model 2: frequency analysis of
SIMULINK Fourier transform in the frequency domain.

SIMULINK is a dynamic system and simulation software Fig. 7 shows the block diagram of model 2 modeled on
the DSP Builder software.
that offers an interactive scientific and engineering
environment for system modeling, analysis, and
simulation. This environment is useful for the rapid
prototyping of a digital signal processing application in
terms of functional blocks, and provides a high-level
simulation. A functional block is a basic structure that can
represent a function, or a specialized system, with defined
input and output ports and customized parameters [17, 18 e
19].
The SIMULINK allowed the modeling and simulation of
the proposed diagnostic system for identification of
bearing failures. The libraries of functional blocks of
SIMULINK allowed the construction of the basic structure
of the functional virtual instrumentation system, which
Fig. 7 – Block diagram of model 2 modeled in DSP Builder.
was divided into three models:
The block normalization of the signal was implemented
Model 1: measurement of the statistical parameters in time
using the library Altera DSP Builder Advanced
domain;
Blockset/ModelPrim and Blockset/Arithmetic. The block
Model 2: frequency analysis of Fourier transform in the
frequency domain; Filter was implemented using the library Altera DSP
Model 3: theoretical calculation of the frequencies of Builder Blockset/MegaCore FIR_Compiler_v10_0. The
failures and displays of geometric data of the bearing. block FFT was implemented using the library Altera DSP
Builder Blockset/MegaCore FFT_v10_0.
Once installed, along with the MATLAB / SIMULINK,
The system designer has to do the mapping the behavioral
the DSP Builder allowed: (i) modeling system developed
of electronic hardware to symbolic representation in
in SIMULINK; (ii) simulation of the model created; (iii)
SIMULINK model. The signal is coming of the vibration
once the development phase has been completed, the
sensor. The selected data acquisition card is a 16-bit card
conversion model for the RTL code in VHDL; (iv)
with a maximum sampling rate of 1.25 MS/s. The model
simulation of VHDL code, using the same test vectors
consists of one analog input block to drive the signal
used in the simulation with SIMULINK; (v) transfer to
sensor, a subsystem of normalization of the signal block, a
FPGA hardware and final test of the complete system.
6th order low-pass Butterworth filter block with a cut-off
After verifying the results obtained in the simulation
frequency of 5 kHz, a FFT calculating block implemented
environment in SIMULINK / DSP Builder, it was
in MATLAB and graphical display block which produces
possible to generate the VHDL code for all models built,
a graph of amplitude versus frequency of the band
as well as the synthesis and configuration of the FPGA
frequency of input signal. A functional block representing
VHDL code.
the block diagram of a model 2 is shown in figure 6.
7. Conclusion
6. Modeling Virtual Instrument System in
DSP Builder Software A new approach for modeling, simulation and
development of a real time vibration measurement and

https://doi.org/10.24084/repqj10.377 548 RE&PQJ, Vol.1, No.10, April 2012

analysis digital instrument based in FPGA was presented. instrument technology”. Journal of Physics: Conference
FPGAs provide the highest DSP performance available on Series, International Symposium on Instrumentation
Science and Technology, vol. 48, pp. 467-473, 2006.
a programmable platform, but optimizing a DSP algorithm
[13] A. Sadoughi, S. Tashakkor, M. Ebrahimi, E. Rezaei,.
in a FPGA can be difficult. Until recently, the algorithm “Fault diagnosis of bearings in rotating machinery based
needed to be ported to HDL and then RTL functional on vibration power signal autocorrelation”, SICE – ICASE
simulation would be verified to using the high-level International Joint Conference 2006, pp. 4709-4714, Oct
simulation test vectors. Modern design tools as 2006, Busan, Korea.
SIMULINK based package called DSP Builder, provide a [14] S. A. S. Al Kazzaz, G. K. Singh. “Experimental
higher level of design abstraction and productivity and investigations on induction machine condition monitoring
and fault diagnosis using digital signal processing
deliver the performance of traditional optimized HDL. techniques”. Electric Power Systems Research, vol.65, pp.
This approach uses a high-level behavioral description of 197-221, 2003.
the DSP algorithm. The design can be simulated and [15] A. Sadoughi, H. Behbahanifardl,. “A pratical bearing fault
deployed into FPGA hardware. It is not required to use diagnoser”, International Conference on Condition
HDL for programming. It is only required to build system Monitoring and Diagnosis, pp. 151-154, April 2008,
model under MATLAB/SIMULINK, and then the design Beijing, China.
can be completed after the parameters of each model are [16] J. R. Stack, R. G. Harley, T. G. Habetler,. “An amplitude
set. modulation detector for fault diagnosis in rolling element
bearings”, IEEE Transaction on Industrial Electronics, vol.
51,No 5, pp. 1097-1102, Oct. 2004.
[17] C. da Costa, M. H. Mathias, P. Ramos, P. S. Girão, “A
References new approach for real time fault diagnosis in induction
motors based on vibration measurement”, Instrumentation
[1] A. J. M. Cardoso and J. L. H. Silva, “Bearing failures and Measurement Technology Conference (I2MTC),
diagnosis in three-fhase induction motors by extended pp.1164-1168, May 2010, Austin, Texas.
park´s vector approach”. Industrial Electronics Society, 31st [18] . G. Xiong, X. Zhou and P. Ji. “Implementation of the
Annual Conference of IEEE, IECON 2005, Nov. 2005. Quadrature Waveform Generator Based on DSP Builder”,
[2] A. Sadoughi and H. Behbahanifardl,” A pratical bearing IEEE Computer Society, International Symposium on
fault diagnoser”. International Conference on Condition Intelligent Information Technology Application
Monitoring and Diagnosis, CMD 2008, pp. 151-154, April Workshops, IITAW´08, pp.773-776, Dez. 2008,China.
2008, Beijing, China. [19] . I. Grout, J. Ryan and T. O´Shea, “Configuration and
[3] P. N. Dobrodeyev, S. A. Volokhov, A. V. Kildishev and J. debug of field programmable gate arrays using
A. Nyenhuis, “Method for detection of broken bars in Matlab/Simulink”, Journal of Physics: Conference Series
induction motors”. IEEE Transactions on Magnetics, Vol. 15, pp.244-249, 2005.
36, no.5, pp. 3608-3610, Sept. 2000. [20] . Barlas, T.; Moallem, M. “Developing FPGA-based
[4] A. Clerc, S. Tashakkor, M. Ebrahimi and E. Rezaei, “Fault embedded controllers using. Matlab/Simulink”. In:
Diagnosis of Bearing in Rotating Machinery Based on SILVESTRE-BLANES,. J. (ed.) Tech, chapter 27, pp.
Vibration Power Signal Autocorrelation”. SICE-CASE 543-556, 2010.
International Joint Conference, pp. 4709-4714, Oct. 2006,
Korea.
[5] A. Lebaroud and G. Clerc, “Diagnosis of induction motor
faults using instantaneous frequency signature analysis”.
IEEE Proceedings of the 2008 International Conference on
Electrical Machines, ICEM 2008, 18 th International
Conference on, pp.1-5, Sept. 2008.
[6] R. R. Obaid, T. G. Habetler and J. R. Stack, “Stator current
analysis for bearing damage detection in induction motors”,
Symposium on Diagnostic for Electric Machines, Power
Electronics and Drives, SDEMPED 2003,pp. 187-
192,August 2003, Atlanta, GA, USA.
[7] J.F. Martins, V. F. Pires and A.J. Pires, “Unsupervised
neural-network-based algorithm for an on-line diagnosis of
three-phase induction motor fault”, IEEE Transactions on
Industrial Electronics, Vol.54, pp. 259-264, No. 1, February
2007.
[8] F. F. Costa, L. A. L. Almeida, S. R. Naidu and E. R. Braga,”
Improving the signal data acquisition in condition
monitoring of electrical machines”. IEEE Transactions on
Instrumentation and Measurement, Vol.53, pp. 1015-1019,
No. 4, August 2004.
[9] M. E. H. Benbouzid, “A review of induction motors
signature analysis as a medium for faults detection”, IEEE
Transactions on Industrial Electronics, Vol.47, no.5, pp.
984-993, October 2000.
[10] A. Sadoughi, M. Ebrahimi, M. Moalem and S. Sadri,
“Intelligent diagnosis of broken bars in induction motors
based on new features in vibration spectrum”. Diagnostic for
Electric Machines, Power Electronics and Drives, IEEE
International Symposium, pp. 106-111, Sept. 2007.
[11] S. Nandi, H. A. Toliyat and X. Li, “Condition monitoring
and fault diagnosis of electrical motors – a review”, IEEE
Transactions on Energy Conversion, vol. 20, no. 4, pp. 719-
729, Dez. 2005.
[12] F. Pan, S. R. Qin and L. Bo,” Development of diagnosis
system for rolling bearings faults based on virtual

https://doi.org/10.24084/repqj10.377 549 RE&PQJ, Vol.1, No.10, April 2012