DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN

INDUCTION MACHINES S. Bachir

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND

ROTOR FAULTS OCCURING IN INDUCTION MACHINES

S. Bachir*,**, S. Tnani*, J. C. Trigeassou*, G. Champenois*

* Laboratoire d’Automatique et d’Informatique Industrielle
40, avenue du Recteur Pineau, 86020 Poitiers Cedex, France.
** Moteurs LEROY SOMER, Direction Scientifique, 16015 Angoulème Cedex, France
Phone : (33) 5 49 45 36 43 / Fax : (33) 5 49 45 40 34
E-mail : Smail.Bachir@esip.univ-poitiers.fr

Keywords
Diagnosis, parameter estimation, modeling, short circuit model, broken rotor bars, induction motors.

Abstract
The authors propose a new diagnosis method for simultaneous stator and rotor faults in induction
motors. This method is based on parameter estimation of the new Park’s model, developed to take into
account inter turn short-circuits of stator windings and broken rotor bars. A diagnosis procedure has
been validated on real faulty experiments.

1. Introduction
Recently, extensive research has been carried out in fault detection and diagnosis of induction motors.
In literature, many proposed techniques are based on spectral analysis of stator current, stator voltage
and electromagnetic torque [1][4][7][8]. These methods are based on detection of sidebands at certain
frequencies using Fourier’s analysis (FFT software). For these latter methods, it is assumed that the
current measurements can be modelized like multi-component mixtures whose magnitudes change
when a failure occurs. Moreover, even in safe conditions, the frequency depends on speed and power
supply frequency. So, the usual techniques, based on spectral analysis, are not well-adapted and only
parametric methods tackle faults detection for adjustable speed motor drive. Recently, continuous
identification has been used to perform diagnosis procedure [2][10][11][13]. These techniques study
the deviation of parameters to detect and localize faults.
In this paper, we present a new model of squirrel cage induction motor dedicated to inter turn short-
circuit and broken rotor bars. First, the short circuit model proposed by [11] has been extended to the
general case. The basis model can explain the stator faults only when they occur on one phase. Thus,
we dedicate to each phase a short circuit element to take into account eventual faults in several phases.
Moreover, we apply Concordia transformation on the multi-loops squirrel cage rotor [1][7] to obtain
the defective rotor resistance according to the healthy one. This expression makes it possible to
calculate the number of broken rotor bars.
The parameters of this model have been identified by Output-Error technique [10][12][13] with
appropriate integration of physical knowledge to perform parameter estimation for diagnosis purpose.
A special machine has been designed and constructed in order to simulate true faulty experiments. The
experimental results show good agreement and demonstrate the possibility of diagnosing simultaneous
faults (inter turn short-circuit and broken rotor bars).

2. A new modeling of induction motors for stator/rotor faults detection

2.1. Stator faults model
In order to take into account the presence of inter turn short circuit winding in stator on the
asynchronous motor, reference [11] proposed an original model which comprises an additional shorted
winding in the Park’s d q axes (fig. 1). In the faulty case, this winding creates stationary field in stator.

EPE 2001 – Graz P. 1

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

Two parameters are introduced to define the stator fault: the angle θ cc between the inter turn short
circuit winding and the first stator phase. This parameter allows the localization of faulty winding and
can take only the three values 0, 2π / 3 and 4π / 3 corresponding respectively to a fault on the real
stator phase a, b and c. The short circuit parameter η cc is equal to the ratio of inter turn short circuit
on the total number of inter turns in one healthy phase. η cc allows the detection of an eventual fault
and the calculation of the number of inter turn short circuit in the stator.

ds
Short circuit
windings
θcc

qr

qs
θ

dr

Fig. 1: Stator windings with inter turn short-circuit

This model, very simple to implement, makes it possible to explain the fault in only one phase with the
simple short circuit element. On the other hand, it is inappropriate in the case of simultaneous faults in
several stator phases. Thus, we propose the generalization of this model by dedicating a short circuit
element to each phase. In presence of defect in several phases, each short circuit element allows the
detection and localization of inter turn short circuit of the corresponding phase.

i dqs i 'dqs Rs ω m P( π / 2 )φ Lf
dqs

i dqr
i cc1 i cc2 i cc3

η cc η cc η cc Lr Rr
U dqs 1 2 3

im

Fig.2: Short circuit model of induction motor

Fundamentally, in the faulty case, an induction machine can be characterized by two models
(corresponding to common and differential modes). The common model corresponding to the healthy
dynamics of the machine (Park’s model) and the differential model explaining faults. Thus, we write
the equations of the global model (Fig. 2)
 dφ
dqs
U dqs = R s .i'dqs + + ω m .P( π / 2 ).φ
Stator (common model) :  dt dqs (1)
φ
 dqs = φ fs + φ ms = ( L f + Lr ).i dqs + Lr .i dqr
'

 dφ
dqr
 U dqr = 0 = Rr .i dqr +
Rotor (common model) :  dt (2)
φ
 dqr = φ ms = Lr .( i dqs + i dqr )
'

EPE 2001 – Graz P. 2

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

2.η cck
Differential model : i cc = P( −θ ).Q( θ cck ).P( θ ).U dqs , k = 1,3 (3)
k 3.R s
3
Resultant current : i dqs = i'dqs + ∑ i cc (4)
k
k =1
where
i cc is the short-circuit current in the kth phase.
k
R s , L f , Rr and Lr are respectively stator resistance, global leakage inductance referred to the stator,
rotor resistance and rotor inductance. ω m and θ are respectively mechanical speed and rotor position.
Number of int er turns short circuit windings
η cck = (5)
Total number of int er turns in healthy phase

cos( θ ) − sin( θ )
P( θ ) =   : rotational matrix
 sin( θ ) cos( θ ) 
 cos( θ cck ) 2 cos( θ cck ). sin( θ cck )
Q( θ cck ) =   : matrix depending on short circuit angle
cos( θ cck ). sin( θ cck ) sin( θ ) 2 
 cck 
θ cck (if the inter turn short-circuit is at the phase a (resp. b and c), then the angle θ cck is 0 rad
(resp. 2π / 3 and 4π / 3 )
2.2. Broken rotor bars model
Most of available methods for detecting rotor bar defects in fixed speed are based on various
identification of side bands around the main frequency components of line current. For detection in
adjustable speed motor drive, these methods are not well-adapted because signals are not stationary.
So, an appropriate dynamical model is necessary. In [1][7], a transient model of rotor cage induction
motor is proposed. This model makes it possible to simulate a broken rotor bar by increasing there
resistance to set its currents equal to zero [1] or by decreasing their number [7]. The small drawback of
this model is that it requires a knowledge of the electrical parameters (bar resistance, end-ring
resistance,…).
In this work, we propose a new formulation of the Park’s rotor resistance according to the rotor bars
number. This formulation introduces a new faulty parameter in order to explain the variation of Park’s
rotor resistance. So, under broken rotor bars, the value of the faulty parameter obtained by parameters
estimation allows to know the number of rotor bars in defect.
It is assumed that the rotor cage can be replaced by a set of mutually coupled loops, each loop being
composed by two bars and end ring portions [1][7]. The equations of loops currents and flux are
given [1] :
dφ
r
0 = R .J r + (6)
dt
where
J r = [ J r0 ! J rN −1 ] T is the vector of currents loops and φ is the vector of flux loops.
b r
 Re
 2.( Rb + N ) for i = j
 b
R ∈ ℜ N b × Nb : resistance matrix where : Rij = − Rb for i = j ± 1 (7)
 0 else


where N b is the number of rotor bars. Rb and Re are respectively resistance of rotor bar and end
ring.

EPE 2001 – Graz P. 3

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

The multi-loops rotor is the system with N b phases. Then, application of Concordia transformation to
equation (Eq. 6) allows to write the Park’s rotor resistance according to the rotor bars as :
dφ dφ dqr
0 = ( T2 Nb .R .T2 Nb −1 ).i dqr +
dqr
= Rr .i dqr + (8)
dt dt
where :
2 cos( 0 ) " cos( kα ) " cos(( N b − 1 ).α )
T2 ,Nb =  sin( 0 ) " sin( kα ) " sin(( N − 1 ).α )  (9)
Nb  b 
T2 ,Nb is a N b phases to two phases Concordia transform and α = 2π / N b a rotor loop angle.
If one neglects the end ring resistance between bar resistance ( Re / N b << Rb ) , the Park’s rotor
resistance is given by
Rr ≈ 2.Rb .( 1 − cos( α ))
In the case of rotor faults, when N bb bars are broken between N b healthy ones, the number of rotor
bars decreases to N b' = N b − N bb [7]. Thus, the expression of faulty resistance R'r according to the
healthy rotor resistance Rr becomes :
1 − cos( α ′ )
Rr ′ = Rr ≈ Rr .( 1 + η bb ) (10)
1 − cos( α )
where :
2
N b 2 − N b'
η bb = : Faulty parameter (11)
2
N b'
α ′ = 2π / N b' : Faulty rotor loop angle.
The broken rotor bars model of squirrel cage induction motors is given by (Fig. 3)

idqs Rs ω m P( π / 2 )φ Lf idqr
dqs

Rr
Lr
U dqs
ηbb .Rr
im

Fig 3: Broken rotor bars model of induction motors

Then, the estimation of η bb by parameters identification allows to obtain the number of broken rotor
bars with
Nb
N bb = N b − (12)
1 + ηbb

2.3. Global state-space faulty induction motor model

In previous sections, two models of stator and rotor faults were presented. For a global simulation and
detection of simultaneous stator/rotor faults, we propose the global faulty model including
• Park’s model with the electrical parameters ( R s , Rr , Lr , L f ) .
• Stator faulty model with the three additional parameters ( η cc1 ,η cc2 ,η cc3 ) .

EPE 2001 – Graz P. 4

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

• Rotor faulty model with broken rotor bars parameter η bb .

For simulation, it is necessary to write the faulty model in state space representation. If mechanical
speed ω m is assumed to be quasi stationary with respect to the dynamics of the electric variables
[10][13], the model becomes linear but not stationary with four order differential equations. For
simplicity, the state vector is chosen composed of two-phase components of the dq stator currents and
the rotor flux. Then, the continuous time model of the faulty induction motor, expressed in the
mechanical reference frame, is
 x#( t ) = A( ω m ).x( t ) + B.u( t )
 (13)
Y = C .x( t ) + D.u( t )
with :
[
x = i ds i qs φ dr φ qr ]T : state vector

U ds  i ds 
u=  et Y =   : input and output.
U qs  i qs 
 R s + R r .( 1 + η bb ) R r .( 1 + η bb ) ωm 
− ωm 
 Lf L f .L r Lf 
 R + Rr .( 1 + η bb ) ω R r .( 1 + η bb ) 
 −ωm − s − m 
Lf Lf L f .Lr
A= ,
 R r .( 1 + η bb ) 
 R r .( 1 + η bb ) 0 − 0 
 Lr 
 R r .( 1 + η bb ) 
0 R r .( 1 + η bb ) 0 −
 
 Lr 
T
 1 
L 0 0 0
 , C = 
1 0 0 0
B=
f
 1  ,
0 1 0 0 
 0 Lf
0 0
 
3 2.ηcck
D= ∑ P( −θ ).Q( θ cck ).P( θ ) (14)
k =1 3.Rs
The discrete time model is deduced from the continuous one by second order series expansion of the
transition matrix [10].

3. Identification Algorithm
3.1. Introduction
Output Error techniques are based on iterative minimization of an output error quadratic criterion by a
Non Linear Programming (NLP) algorithm, the faulty model presented (Eq. 13-14) depending only on
four electrical parameters and four faulty parameters. Usually, users have a good knowledge of the
electrical induction motors parameters, so it is very interesting to introduce this information in the
estimation process to provide more certainty on the uniqueness of the optimum. For this purpose, we
have modified the classical quadratic criterion [10][13], in order to incorporate physical knowledge.
3.2. Compound quadratic criterion
Consider the state-space induction motor representation
 x#( t ) = A( θ ).x( t ) + B( θ ).u( t )
Y = C( θ ).x( t ) + D( θ ).u( t ) = f ( θ , u( t )) (15)


EPE 2001 – Graz P. 5

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

[
with θˆ = R s Rr Lr Lf η cc1 η cc2 η cc3 ηbb ]T (16)
A predicted output ŷ is obtained by numerical simulation of the state space model
k
ŷ = f ( θˆ , u k ) (17)
k

where θ̂ is an estimation of true parameter vector θ .

Assume that we have measured K values of input-output ( u( t ), y* ( t ) with t = kTe ; where Te is the
sampling period). Then defining the output prediction error
ε k = y*k − ŷ k ( u ,θˆ ) (18)
In order to incorporate physical knowledge or a priori information, classical quadratic criterion has
been modified. The solution is to consider a compound criterion J c mixing a priori estimation θ 0
(weighted by its covariance matrix M 0 ) and the classical criterion (weighted by the variance of output
noise δ̂ 2 ). Then, compound criterion is usually defined as [10]:
K
∑εk
2

J c = ( θˆ − θ 0 )T M 0 −1 ( θˆ − θ 0 ) + k =1 (19)
δˆ 2
Thus, optimal parameter vector minimizing J c is the mean of prior knowledge and experimental
estimation weighted by their respective variances.
3.3. Minimization of compound criterion
We obtain the optimal values of θ by Non Linear Programming techniques. Practically, we use
Marquardt’s algorithm [9] for off line estimation

{
θˆ i +1 = θˆ i − [J θθ }
′′ + λ .I ]−1 .J θ′ θˆ =θ
i
(20)

where
 K 
 ∑ ε k .σ k ,θ 
 
J θ′ = 2. M 0 −1 ( θˆ − θ 0 ) − k =1  : gradient
 δˆ 2

 
 K
T 
 ∑ σ k ,θ .σ k ,θ 
′′ ≈ 2. M 0 −1 + k =1 
J θθ   : hessian
 δˆ 2

 
λ : monitoring parameter.
∂ŷ
σ k ,θ = k : output sensitivity function
∂θ i

4. Diagnostic procedure
Firstly, a set of experiments and identification of only electrical parameters is used in order to
constitute a reference or nominal values of Park’s parameters (including their variances and variance
of the output disturbance). It is very important to note that only electrical parameters in the vector θ
(Eq. 16) are weighted by their variance in the compound criterion J c . No information is introduced
on faulty parameters η cck and η bb because it is evident that we have no knowledge on the state of the
machine.

EPE 2001 – Graz P. 6

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

Then, during industrial operation, sequential electrical data acquisitions are performed. Each set of
datas is used to get a new estimate. Thus, electrical parameters estimation corresponds to dynamical
state of the motor (resistance values, time constant, that is to say the common mode model) whereas
the faulty parameters indicate a number of inter turns short-circuit at each phase and the number of
broken rotor bars in the induction motor (that is to say the differential faulty mode model).

5. Experimental results
5.1. Data acquisition
In order to validate diagnosis procedure, a set of experimental tests were carried out. The tested motor
used in the experimental investigation of the occurrence of inter turn stator winding faults was a three
phase, 50 Hz, 4-poles, 1.1 kW squirrel cage induction machine. In order to allow inter turns short
circuit of different levels, the stator winding has been modified by addition of number of tapings
connected to stator coils in one of the phases. Different rotors, with broken bars, are used to simulate a
bar breakage occurring during operation. The induction machine is driven by field oriented vector
algorithm included in a speed control closed-loop. To permit motor operation under different load
levels a DC generator was coupled to the motor shaft.
The data acquisition system has a sampling period equal to 0.7 ms. Measured variables are passed
through a 4th order butterworth analogic filter whose cut-off frequency is 500 Hz.
Identification algorithm needs persistent excitation to provide appropriate information. This excitation
is realized with a P.R.B.Sequence equal to 90 rpm added to the reference of the speed loop equal to
750 rpm.
5.2. Experimental results
Different tests (10 realizations by manipulation) have been performed (Table I) :
1. Healthy motor.
2. A short circuit of 3.88 % (18 inter turn short circuit ) in 1st phase.
3. A short circuit of 12.5 % (58 inter turn short circuit) in 2nd phase.
4. A simultaneous short circuit of 12.5 % in 1st phase and 6.25% in 2nd phase.
5. One broken rotor bar.
6. Simultaneous 12.5 % of inter turn short circuit in 1st and 2nd phases with two broken rotor bars.

Estimation results
Experimental manipulations
mean of 10 acquisitions
η cc1 ,η cc 2 ,η cc 3 (%), N bb η̂ cc1 η̂ cc2 η̂ cc3 N̂ bb
-3
1) Healthy motor 0, 0, 0 (%), 0 1.30 0.50 - 9.10 0.13
2) 3.88, 0, 0 (%), 0 3.66 1.30 - 0.55 0.11
3) 0, 12.50, 0 (%), 0 0.31 10.50 0.99 0.13
4) 3.88, 12.50, 0 (%), 0 3.11 10.60 0.42 0.14
5) 0, 0, 0 (%), 1 1.49 0.60 0.23 0.93
6) 12.50, 12.50, 0 (%), 2 10.12 10.75 - 0.95 1.86

Table I. Estimation results of stator and rotor faults

As observed in table 1, there is good agreement between a real fault and its estimation. All the faulty
parameters vary to indicate the values of inter turn short circuit in the three-stator windings and the
number of broken rotor bars. Estimation error of inter turn short circuit in all case is negligible and
doesn’t exceed five inter turn. In the healthy case, estimation N̂ bb indicates that no bar is in defect,
but in the cases 5) and 6), N̂ bb shows a good estimation of the number of broken rotor bars.

EPE 2001 – Graz P. 7

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

Figure 4 presents the evolution of inter turn short circuit estimation in one phase for several
experiments. We observed that all the estimation results exhibit the good approximation of the fault.
Figure. 5 presents the dispersion of the 10 estimations in different situations of rotor faults. It is
evident that the mean of these estimation is the good indication from the number of broken rotor bars.

η cc Nbb

0.18
o estimation 2
+ real faults o estimation
0.14 12.5 % real faults
1.5

1
0.1
6.25 % 0.5
0.06
3.88 % 0

0.02 0% -0.5

0
Fig. 5: Broken rotor bars estimation
Fig. 4: Estimation of inter turn short circuit

5.3. Spectrum analysis

The identification residual corresponding for the comparison of real dq stator currents and their
optimal estimation conveys the “faulty” information. Thus in [10], it is shown that using conventional
Fourier’s analysis, it is easy to distinguish stator faults from rotor ones when conventional Park’s
model is used. This model is not able to explain the faults because it is very restrictive, spectrum lines
are found at frequencies F>50 Hz, corresponding to the stator fault (Fig. 6.a).

40
20 Usually Park’s
0 model
-20
-40
-60
-80

Hz
0 50 100 150 200

Fig. 6.a: Spectral density (dB) of the identification residuals (without short circuit elements)

In figure 6.b, FFT analysis of identification residuals shows that these sidebands are eliminated when
we use the short circuit model. This proves that this faulty model is able to explain the faults.

20

0 Short circuit model

-20

-40

-60

-80

Hz
0 50 100 150 200

Fig.6.b: Spectral density (dB) of the identification residuals (with short circuit elements)

EPE 2001 – Graz P. 8

DIAGNOSIS BY PARAMETER ESTIMATION OF STATOR AND ROTOR FAULTS OCCURING IN
INDUCTION MACHINES S. Bachir

In fact, the currents of shorted windings have fundamental frequency equal to 2.F0 , where F0 is
power supply frequency and if we develop the expression of the short circuit currents (Eq. 3), we find
that the frequency of their fundamental is equal to 2.F0 . This explains the absorption of the sidebands
at this frequency.

6. Conclusion
In this paper, a new model dedicated to squirrel cage induction machines has been presented for the
realistic identification and detection of stator and rotor faults. With this tool, broken bars and inter turn
short circuit windings can be simulated under varying speed. Firstly, inter turn short circuit windings
have been modeled by a short circuit elements. Each element has been dedicated to a stator phase in
order to explain the fault. Secondly, Park’s rotor resistance is originally expressed to allow the
decreasing of the number of rotor bars in faulty situation.
Parameter estimation is used to perform fault detection and localization. Prior information of only
electrical parameters of Park’s model (common model) has been introduced in the optimum search.
Experimental tests illustrate the efficiency of this technique for use in off-line stator/rotor faults
diagnosis. The estimates of the number of inter turns short circuit windings and broken rotor bars in
different realizations give a good approximation of the fault level in the machine.
Finally, the ultimate objective will be to detect and localize the faults without the use of a speed
sensor. So, it will be necessary to develop a speed estimator with only electrical signals.

References
1. Baghli. L., Hein. D., Razik. H., Rezzoug. A. Modelling rotor cage induction motors for default detection.
IEEE International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives
SDEMPED’97, September 1997.
2. Coirault. P., Trigeassou. J. C., Gaubert. J. P., Champenois. G. Parameter identification of an induction
machine for diagnosis. IFAC Symposium Safeprocess’97,Kingston, vol.1, pp. 276-281, August 1997.
3. Grellet. G., Clerc. G. Actionneurs électriques. Principes, modèles et commande. Eyrolles, 1997.
4. Innes. A. G., Langman. R. A. The detection of broken bars in variable speed induction motors drives.
ICEM’94.
5. Jemni. A. Estimation paramétrique des systèmes à représentation continue, application au génie électrique.
Ph.D. Thesis. Université de Poitiers. France 1997.
6. Ljung. L. System identification : Theory for the user. Prentice Hall, 1987.
7. Makki.A, Ah-jaco. A, Yahoui. H, Grellet. G. Modelling of capacitor single-phase asynchronous motor under
stator and rotor winding faults. IEEE International SDEMPED’97, pp. 191. September 1997.
8. Maléro. M. G., et AL. Electromagnetic torque harmonics for on-line interturn shortcircuits detection in
squirrel cage induction motors. EPE’99. Lausanne 1999.
9. Marquardt. D. W. An Algorithm for least-squares estimation of non-linear parameters. Jour. Soc. Indust.
Appl. Math., vol. 11, No. 2, pp. 431-441, 1963.
10. Moreau. S. Contribution à la modélisation et à l’estimation paramétrique des machines électriques à courant
alternatif: Application au diagnostic. Ph.D. Thesis. Université de Poitiers, France 1999.
11. Schaeffer. E. Diagnostic des machines asynchrones: modèles et outils paramètriques dédiés à la simulation
et à la détection de défauts. Ph.D. Thesis. Université de Nantes. France 1999.
12. Trigeassou. J.C, Gaubert. J.P, Moreau. S, Poinot. T. Modélisation et identification en génie électrique à partir
de résultats expérimentaux. Journées 3EI’99, Supelec Gif-sur-Yvette, march 1999.
13. Trigeassou J C., Moreau S., Champenois G., Gaubert J P. Diagnosis of Induction Machines: A procedure for
electrical fault detection and localization. IEEE International SDEMPED’99, pp. 225-230. Gijon, September
1999.

EPE 2001 – Graz P. 9