MOTOR CURRENT SIGNATURE ANALYSIS

bY
P.Pillay* and z .xu
*Clarkson University, ECE Dept, P.O.Box 5720, Potsdam, N’I 13699-5720
Tel(3 15) 268-6509, Fax (3 15) 268-7600
Email - [email protected]
IDM Controls, Georgia.

Abstract - This paper shows how the speed of an induction d) Offers means for separating one form of disorder from
motor (IM) can be extracted from the motor currents of an another,
induction motor with a Labview implementation of Motor e) Can be performed rapidly and as frequently as desired by
Current Signature Analysis (MCSA). The implementation relatively unskilled personnel using portable, inexpensive
is first applied to an IM driven by the 60 Hz mains. The equipment,
estimation system consists of data acquisition, demodulation, f) Is equally applicable to high-powered and fractional
and FFT spechum analyses. The estimation is then extended horsepower machines, ac and dc motors.
to PWM inverter-fed LM drives. For PWM drives, the
presence of stator current harmonics and a variable MCSA can be implemented using time domain or
fundamental frequency demands filtering and a more fi-equencydomain analyses. Whde the technique is powerful
powerful demodulation technique. The above estimation enough to detect mechanical (degradations,the contribution
algorithms are implemented in Labview for Windows. in this paper is a real-time Labview implementation for
speed detection. Both line-fed and inverter-fed motors are
I. INTRODUCTION considered. The paper is organized as follows: Section I1
presents the fhdamentals of MCSA. Section I11
Motor Current Signature Analysis (MCSA) is an
electric machinery monitoring technology developed by the 11. FUNDAMENTALS OF MCSA
Oak Ridge National Laboratory (ORNL) [1-41. It provides
a highly sensitive, selective, and cost-effective means for on- MCSA is based on the recognition that a
line monitoring of a wide variety of heavy industrial conventional electric motor powering a machine also acts as
machinery [5-lo]. It has been used as a test method to an efficient and permanently connected transducer, detecting
improve motor bearing wear assessment for inaccessible small timedependent motor Ic~advariations generated within
motors during plant operation. In 1989, ORNL used it to the mechanical system and converting them into electric
monitor a variety of electric motor driven devices at the current signals that flow along the cable supplyingpower
Philadelphia Electric Company Eddystone Generating to the motor. These signals, though small in relation to the
Station for detecting the degradation in aging power plant average current drawn by the motor, can be extracted
equipment. In a comprehensive assessment of the aging of reliably and nonintrusively and processed to provide
motor operated valves (MOVs), MCSA has shown to be indicators of the condition (signatures) of the motor. The
capable of detecting, differentiating, and tracking the trend of these signatures can be determined over time to give
progress of the MOV abnormalities,such as abnormal line information concerning the motor and the load.
voltage and worm gear tooth wear. Extensive test data Current signals can be analyzed in the time-domain
support that MCSA has a number of inherent strengths, the or the frequency-domain.Tlie former is also capable of
most notable being that it: analyzing systems during transients, such as during the
initial or fmal operation of the system. MCSA requires
a) Provides nonintrusivemonitoring capability at a location amplitude information of the motor currents. The currents
remote from the equipment, have imbedded information on the driven loads, with the
b) Provides degradation and diagnosticinformation informationbeing available 11x1 the frequency domain or the
comparable to conventional instrumentation, time domain. To obtain rotor speed, fi-equency-domain
c) offers high sensitivity to a variety of mechanical disorders analysis is chosen. Fig.2.1 shows a general procedure for
affecting operational readiness, MCSA.

$5.000 1996 IEEE

0-7803-3544-9196
587
 the results can be viewed on the computer screen.
Data acquisition b) The VI receives insh-uctionsfrom a block diagram, which
is comtmctedin G. The block diagram is a pictorial solution
to a programming problem. The block diagram is also the
source code for the VI.
Demodulation
A G program for data acquisition is shown in
Fig.2.2. Its front panel is shown in Fig.2.3. Since Labview
controls several devices (interfaces) and each device has a
lumber of
FFT of current
signature ,..............................................

Fig. 2.1 General procedure for MCSA

Motor current signals can be obtained from the outputs of

current transducers which are placed nonintrusively on one
of the power leads. The resulting raw current signals are
acquired by computers after they go'through conditioning
circuits and data interfaces. The signals are processed by
Labview in the computer after acquisition.

A. Labview Implementation

Labview is a program development application, Fig. 2.2 G program for acquisition

much like various C* or BASIC development systems, or .
National Instruments Labwindows. However, Labview is
different from those applications in one important respect.
Other programming systems use text-based languages to p~ transposed vollaps waph
create lines of code, while Labview uses a graphical $. cl
................................................................... :
programming language,(;, to create programs in block
diagram form. Labview uses terminology, icons, and ideas
familiar to researchers and relies on graphic symbols rather
than textual language to describe programming actions.
Labview has extensive libraries of functions and subroutines
for most programming tasks. Labview also contains
application-specificlibraries for data acquisition, analysis,
presentation and storage. Labview includes development
tools designed speclfically for data acquisition and
instrument control. Labview programs are called virtual n
...
instruments (VIS)because their appearance and operation
imitate actual instrument. However, they are analogous to
functions fi-om conventional language programs. VIS have
both an interactive user interface and a source code
equivalent and accept parameters from higher level VIS.
These three VI features are discussed below. Fig. 2.3 Front Panel for acquisition

a) The interactive user interface of a VI is called the front channels (VO ports), an acquisition port must specified. The
panel, because it simulates the panel of a physical subroutine named AI CONFIG is built to choose a device,
instrument. The fiont panel can contain knobs, push buttons, channel, and size of buffers which are used to the store
gmphs, and other controls and indicators. After data is input, acquired data. A choice can be easily made by typing

588
required numbers on the controller in the panel. The Al the mechanical and electrical systems. f, is 60 I-k,which is
START subroutine starts acquisition and stores acquired the frequency of ac power supply. (2.1) is rewritten as
data into a buffer. Al READ reads the data from the buffer
and shows them on a graph which is a screen in the panel. i ( t )= m ( t ) c o s ( 2 n f o t ) (2.2)
The output of Al READ can be transferred to other VIS or
stored in a spreadsheet. The data can be stored in a
spreadsheet using SGL or other file subroutines. where m(t) is the amplitude ccf the stator current. Compared
with (2. I), m(t) is
A stator current signal acquired from a motor
driven by 60 Hz mains is shown as a graph on the front of m( t )=k,+k,cos ( 2 l I f I t ) +k,co
the panel in Fig 2.3. It is acquired at 4800 points per second. ... +kiC0S (2nf,t)
To show that the amplitude of current is variable in time, the
positive peak portion is redrawn in Fig.2.4. Since rotors of
IMs are loads, spectra of the signals which are demodulated The frequency of each term in (2.3) is lower than f,. To
from corresponding current signals must contain the extract m(t) from the stator current signals, three methods
componentsrefmed to the rotor speed. have been developed. They ,are described below.

C. Method One
A stator current signal, as expressed in (2.1), is
first squared to yield
i ( t )x i ( t )=m(t ),cos2 ( 2 n f 0 t ) (2.4)
Wantposed voltape praph
IJJj
...........................................................

Rewriting (2.4)

( i( t )) 2 = -1m ( t )2 +
2
1
-cos ( 2 x 2 n f O t ) m ( t ) * ( 2 . 5 )
2

&a fib, :
0
......................................................................................
lid c h a r d

Fig. 2.5 Procedure for demoidulation

Fig. 2.4 Positive peak portion of an acquired signal
Since the fi-equency of each component in m(t) is lower than
B. Demodulation of Current Signals f, ,a lowpass filter can be used to filter the second term of
(2.5). m(t) is extracted after i(t)* goes through a lowpass
Regardless of thermal noise, the stator current filter and then a square-rooted operator. The cutoff
signals can be expressed as frequency of the lowpass filter must be lower than 2f0 in
order to filter components with frequency of 2f,. The
i ( t )=ko+k,cos( 2 n f 1 t ) +k2cos( 2 n f 2 t )+ procedure of Method One is shown in Fig. 2.5. Both the
square operation and square-rooted operator are
...+kicos ( 2 n f i t)c o s ( 2 n f 0 t )(2.1)
implemented using Labview functions. Fig. 2.6 is a G
program based on Method ( h e . The current signal i(t) is
sent to a multiplier to multiply itself and then i(t)* is
where k,are constants, are frequencies,which depend on obtained. To filter signals with kquency componentshigher

589
than 2f0, i(t)' is converted into Fourier transforms and
relative components set to zero. Converting this Fourier
transforms back from the ji-equency domain to the time
domain gives m(t)'. Finally, amplitude information is
separated from the current signal after m(t)' is square
rooted.

D.Method "WO
An acquired current signal is shown in Fig. 2.7. Its
sampling frequency is 4800 Hz ,i.e., there are 80 samples
in a cycle of ac. In order to obtain amplitude information, the
acquired current signal is converted to its amplitude series
in which a sample is the maximum

Fig. 2.6 An acquired current signal

nethod Two I

J IMaqnitude serie:
IDBL]

Fig. 2.6 G program for Method One

value in each cycle. This amplitude series contains the same

information as m(t) in (2.3). Fig. 2.7 shows a G program
which converts an acquired current signal into an amplitude
series using the maximum values of each cycle. A loop is
executed 625 times for an acquired signal array with 50000
samples. For each iteration, it chooses 80 successive
samples from the array, finds a sample with the maximum
value, and outputs the sample. An amplitude series is
produced after the iteration is finished.

Fig. 2.7 G program for Method Two

E. Method Three
In the Method Two, the maximum value in each
cycle of ac is used as a sample in an amplitude series. To
increase the range of spectra, the negative amplitude (its

590
absolute value) can also be treated as a sample. For the
acquired signal shown in Fig. 2.6, both positive and negative
amplitude values of each cycle are used as samples. For an
amplitude series converted in this way, it has double the
length of that converted using Method Two. I

IMaqnitude series J

4zl Fig. 2.9 G program to display the spectrum of m(t)

the complex form of the Flourier Transform to a polar

pdy expression. The Subset function chooses an appropriate

range to display. The Bundle function builds a convenient
X scale for display.

1-1 A model to implement MCSA is shown in Figs.

2.10 and 2.1 1, which includes acquisition, demodulation
(Method One), and the FFT of m(t). A spectrum of m(t) is
drawn in Fig. 2.1 1. The spectrum shows that the rotor speed
is 29.3 Hz and the slip frequency is 2.4 Hz.An experiment
Fig. 2.8 G program for Method Three to demonstrate the ability of the: three models to detect speed
has been done.
Therefore, the range of its spectrum is increased. A G
program for Method Three is shown Fig. 2.8. A Loop I11 SPEED ESTIMATION OF MOTOR
iterates 1250 times. For each iteration, it searches the A Introduction
sample for the maximum amplitude in 40 successive
samples. An absolute function is used here to convert In the previous section, three methods for
negative amplitude to its absolute value. extracting the amplitude information from current signals
have been developed. The spectra produced by the three
To read the spectrum of m(t), the G program different methods have the same characteristics. To find the
shown in Fig. 2.9 is built. A FFT subroutine converts m(t) motor speed, the speed spectrum must be searched in the
to its Fourier Transforms. The Complex to Polar function range of ( fk to 2fo/p ), where: p is the number of poles and
converts 2fdp is the synchronous mechanical speed 2fdp is the upper
limit of the rotor speed. fk can be determined by

where Rr is rotor resistance, Rs is stator resistance, Xis is

stator leakage inductance, and Xh is rotor leakage
inductance. The component with the maximum amplitude
value in that range correspcnds to the rotor speed. The
program in Fig. 3.3 is built to find fr.
The FFT subroutine outputs the Fourier transform
of m(t). It is a complex array and is converted to a polar

59 1
 array by the Complex to Polar function. The Subset
function picks up a range for searching. The Max & Min
~ c t i o nfinds
s the component with the maximum value and
gives its position in the range. Subsequently,its position in
the Fourier transform array is found. and E, is obtained by
using its position multiplied by the resolution of frequency,

tofk I
where i is the position of the component associated with $..
96
I 1
]Correspondingto 2fOIp I
MlN .
I I I I

IR esolution of Freq

Fig. 3.3 G program to search f,

For normally operated IMs connected to the

6OHz supply, 2 is 60 Hz and the harmonics in the current is
negligible. When the motor is driven by a PWM inverter,
harmonics appear in the current signals. A current signal
acquired from an IM dnven by a PWM inverter is shown in
Fig. 3.4. The harmonic peaks superposed on the
fundamental wave will give the wrong amplitude
information. Therefore, it is necessary to filter those
harmonics before demodulating the current signals. The
Fig. 3.1 G program of MCSA demodulation method needs to be modified, if Method One
I is chosen. During PWM control, the IM operates at different
speeds. When 2 is changed, the cutoff frequency (2fb) of the
lowpass filters also has to change.

;
15.0-1

10.0-1
5.0-1

0.04
0.0 510 1d.O I d 0 d.0 2g.O 3d.O 3g.O 4dO -H

I fslip=2.4 i fr=29.3

Fig. 3.2 Panel of the G program in Fig. 3.1. Spectrum

indicates: E,=29.3,f,,=2.4.

Fig. 3.4 A Stator Current Signal Acquired from a PWM-

driven Motor

592
B. M i t e Impulse Response Filters(1IR) or autoregressive movingaverage (ARMA) filters. It is
another reason to use IIR fillers in this application. Fig. 3.6
is a Labview implementation of an IIR filter.

A ,Y

Fig. 3.6 IIR of Labview

The inputs of the filter is a signal array that requires

procesSing,forward coefficients, and feedback coeficients.
Fig. 3.5 General block diagram for IIR Its output is a filtered signal array. Fig. 3.7 shows a filtered
signal for the signal with harmonics shown in Fig. 3.4.
The IIR is chosen here, because it requires fewer
coefficients to perform the same filtering operation. Thus,
IIR filters execute much faster and require less memory.
Another reasonis that MCSA in this thesis does not require
the phase information of each component. The infiiite
impulse response filters (IIR) in Labview are digital
filters whose impulse response can theoretically be
infinite in duration. Fig. 3.5 shows a general block diagram
for IIR filters.
The digital characterizationbetween the input
sequence x and output sequence y is
-2.04
o 2'5 5'0 7'5 do 115 160 1;5 2m 2k Goln
j."

I
where n is the polynomial order of the forward branch
labeled A, and m is the polynomial order of the feedback
branch labeled B. Fig. 3.7 A filtered Signal
The delay associated with the feedback branch of
the filter contipration causes the infinite impulse response. C. Speed Estimation for Adjustable Speed PWM-driven
Even when no more external samples feed into the system, Motors
the feedback portion continues to send samples to the filter The demodulation methods developed in the
structure. In practical filter applications,the input sequence previous sections are applied when the fundamental
is finite, and the output sequence is truncated to a finite hquency of the stator current is 60 &. For variable speed
number of samples. The disadvantageof IIR filters is that drives, the synchronous frequency fois adjustable. fogives
the phase response is nonlinear. Since the application used the high cutoff fi-equency of 2f0, for the lowpass filter in
here is simplesignal monitoring and does not require phase Method One, and determines a range to search for $, The
information, IIR filters are well suited. For linear phase filndamental fiequency can be obtained using the FFT of the
response applications, finite impulse response filters are stator current signal. Mer f, is found, it is sent to a
generally used. IIR filters are also known as recursive filters demodulation block to deternline the high cutoff fkequency.

593
 A labview program for speed estimation of the presentation of three different methods to demodulate the
PWM-driven IM is shown below. Its panel indicates the fundamental the component so that the components of
synchronousfi-equency&, motor speed $., and slip frequency interest can be detected with greater ease. While the
fdip. technique can be applied to a range of problems associated
with condition monitoring, particular attention was paid to
the detection of the speed signal for closing the speed loop.
The applicability to PWM inverter-fed motors was also
examined.

ACKNOWLEDGEMENTS
Dr P.Pillay would like to thank Entergy Corporation for
funding his research and Ms R.D.Pillay for help with the
document preparation.

REFERENCES
[11H.D.Haynes, D.M.Eissenberg,“Motor current signature
analysis method for diagonosingmotor operated devices, “

US Patent no 4965513, Oct 23,1990.

[2] H.D.Haynes, “Application of signature analysis for
determining the operational readiness of motor operated
valves under blowdown test condition”, Nuclaer
Engineering and design, vol 118, 1990, pp 399-408.
[3] R.C.Kryter and H.D.Haynes, “Condition monitoring of
machmery using motor current signature analysis”, Sound
and vibration, 1989, pp 14-21.
[4] L.P.Gradin, W.B. Cartwright and N.M.Burstein, “Test
Fig. 3.8 G program to estimate rotor speed of IMs dnven by method improves motor bearing wear assessment at Calvert
PWMS Cliffs”, Power Engineering, June 1994, pp 32-33.
[5] G.B.Kliman and R.A.Koeg1, “Noninvasive detection of
broken bars in operating induction motors”, IEEE Trans. On
Energy Conversion, vol3, no.4 ,Dec 1988.
[6] R.R.Schoen and T.H.Habetler, “Effects of time-varying
loads on rotor fault detection in induction machines”, IEEE
125.0- IAS Annual Meeting, 1994, pp 324-330.
[7] A.J.Penman, J.C.Tait and W.E. Bryan, “A software-

,j,d,,
100.0-
75.0- based approach to machine condition monitoring”, 2nd
50.0-
International Conference on Electrical Machmes-Design and
Applications, London, Sept, 1985.
25.0-
[SI F.Filipetti, G.Franceschini, C.Tassoni and P.Vas,
0.0-,
0.0 i0 ld.0 1i.O 2d.0 2i.O 3d.O “Broken bar detection in induction machines: comparison
between current spectrum approach and parameter
estimation approach, IEEE 1994 IAS Annual Meeting, pp
95-102.
[9] F.Filipetti, M.Matelli, G.Franceschini and C.Tassoni,
“Development of expert system knowledge base to on line
Fig. 3.9 Panel of the G program in Fig. 3.8, f0=28.12, diagnosis to rotor electrical faults of induction motors”,
f;=13.93 IEEE IAS Annual Meeting, Oct 1992.
[lo] J.R.Cameron, W.T.Thompson and A.B.Dow,
IV CONCLUSIONS “Vibration and current monitoring for detecting airgap
eccentricity in large indction motors”, Proc IEE, vol 133, Pt
Ths paper has developed a Labview implementation of B, no 3, pp 155-163.
motor current signature analysis. A key feature is the

594