Article - PMSM - Diagnostic Court Circuit
Article - PMSM - Diagnostic Court Circuit
Article - PMSM - Diagnostic Court Circuit
Article
Real-Time Detection of Incipient Inter-Turn Short Circuit
and Sensor Faults in Permanent Magnet Synchronous Motor
Drives Based on Generalized Likelihood Ratio Test and
Structural Analysis
Saeed Hasan Ebrahimi * , Martin Choux and Van Khang Huynh
Department of Engineering Sciences, University of Agder, 4879 Grimstad, Norway; martin.choux@uia.no (M.C.);
huynh.khang@uia.no (V.K.H.)
* Correspondence: saeed.ebrahimi@uia.no; Tel.: +47-462-47-096
Abstract: This paper presents a robust model-based technique to detect multiple faults in permanent
magnet synchronous motors (PMSMs), namely inter-turn short circuit (ITSC) and encoder faults. The
proposed model is based on a structural analysis, which uses the dynamic mathematical model of a
PMSM in an abc frame to evaluate the system’s structural model in matrix form. The just-determined
and over-determined parts of the system are separated by a Dulmage–Mendelsohn decomposition
tool. Subsequently, the analytical redundant relations obtained using the over-determined part of the
system are used to form smaller redundant testable sub-models based on the number of defined fault
terms. Furthermore, four structured residuals are designed based on the acquired redundant sub-
models to detect measurement faults in the encoder and ITSC faults, which are applied in different
levels of each phase winding. The effectiveness of the proposed detection method is validated by
an in-house test setup of an inverter-fed PMSM, where ITSC and encoder faults are applied to the
Citation: Hasan Ebrahimi, S.; Choux,
system in different time intervals using controllable relays. Finally, a statistical detector, namely a
M.; Huynh, V.K. Real-Time Detection
generalized likelihood ratio test algorithm, is implemented in the decision-making diagnostic system
of Incipient Inter-Turn Short Circuit
and Sensor Faults in Permanent
resulting in the ability to detect ITSC faults as small as one single short-circuited turn out of 102, i.e.,
Magnet Synchronous Motor Drives when less than 1% of the PMSM phase winding is short-circuited.
Based on Generalized Likelihood
Ratio Test and Structural Analysis. Keywords: fault diagnosis; inter-turn short circuit; sensor fault; structural analysis; generalized
Sensors 2022, 22, 3407. https:// likelihood ratio test; PM synchronous motor
doi.org/10.3390/s22093407
Hilbert–Haung transform [14], wavelet transforms [15], and Cohen distributions [16]. These
signal-based methods face challenges of real-time implementations due to the computa-
tional burden, and missing fault characteristic signals does not guarantee that the machine
is healthy [8]. The data-driven approach such as an artificial neural network (ANN) [17]
and Fuzzy systems [18] requires a lot of historical data to train models and classify lo-
calized faults. Historical data is restricted in industry and producing a lot of historical
data in healthy and faulty conditions is costly and time-demanding [19]. Alternatively,
model-based techniques have been proposed to detect ITSC faults [20–22]. Among them,
the finite element method (FEM)-based models have been widely used due to the accuracy
and convenience of taking into account physical phenomena, e.g., saturation. FEM models,
known as time-demanding and computationally heavy ones, require deep knowledge
of the system, e.g., detailed dimensions and material characteristics. Other model-based
methods that use mathematical equations to model a motor’s behavior have been reported
to have challenges regarding validity when experiencing abnormal conditions such as
internal faults [8]. To address the mentioned challenges, structural analysis is proposed
as an alternative solution for detecting ITSC faults in electrical motors. The structural
analysis algorithm has been well studied and developed in the literature [23–25] and ap-
plied to different structures. The structural analysis approach has been able to successfully
detect faults in automotive engines [26–28], hybrid vehicle [29], and battery systems [30].
In [31,32]. The algorithm has successfully been applied on PMSM electric drive systems to
detect sensor faults such as voltage, current, encoder, and torque sensors. In our previous
study [33], it was proposed that the algorithm can be used on an electric drive system to also
detect common physical faults in PMSMs such as ITSC and demagnetization, and residual
responses were obtained by simulation. However, in previous studies, this algorithm has
not been implemented in real-time diagnosis of an industrial PMSM for detection of ITSC
faults. Implementing a structural analysis technique on a PMSM and drive, this paper aims
to achieve the following contributions:
• Detection of both internal motor faults and external measurement faults, namely ITSC
and encoder faults;
• Detection of the lowest level of ITSC fault, with one shorted turn in stator phase
winding;
• Early detection of an ITSC fault, i.e., considering a lower fault current in the degrada-
tion path as compared to shorted turns;
• Modeling of the noise in drive system measurement signals with unknown amplitude
and variance.
This paper presents a systematic fault diagnosis methodology based on structural
analysis for detecting multiple faults in PMSM drives, namely ITSC faults and encoder
fault. To achieve this, a healthy dynamic mathematical model of PMSM is defined in the
abc reference frame based on the dynamic constraints, measurements, and derivatives. To
model an ITSC fault in any phase, specific fault terms are added to the three-phase flux and
voltage equations. These fault terms include the deviations in the voltage, flux, and currents
of the stator winding caused by an ITSC fault, since a part of winding is shorted; hence,
three-phase voltage and flux signals are subjected to changes. In addition, fault terms are
added to the dynamic model to take into account the encoder faults, resulting in errors of
the angular speed and angle measurements. Subsequently, the analytical redundant part of
the structural model is extracted and divided into minimally over-determined sub-systems
from which three sequential residuals are obtained based on the error in the current signal
of each phase. Furthermore, a resultant residual is formed in the αβ frame to achieve a
better demonstration of different ITSC fault levels. Finally, a generalized likelihood ratio
test is developed to detect the faults in the resultant and encoder residuals under unknown
noise parameters assumptions, i.e., unknown amplitude and unknown variance.
Sensors 2022, 22, 3407 3 of 22
H0 : r (u, z) = 0 (1)
H1 : r (u, z) 6= 0
This methodology is especially effective for fault diagnosis of complex systems where a
prior deep knowledge of the whole system is neither needed nor affordable in terms of
computational burden and processing time. Instead, a small redundant part of the system is
selected and processed to obtain smaller redundant subsystems that can be used in forming
residuals for detecting each predefined fault. First, the structural model of a redundant
system is formed and represented by an incidence matrix with variables as columns and
equations as rows. The variables are categorized as unknown variables, known variables,
and faults, while the equations are categorized as dynamic equations, measurements,
and differential equations. Each row of the incidence matrix connects an equation to
the corresponding variables if they are present in that specific equation. Next, the just-
determined and over-determined parts of the system are separated by rearranging the rows
and columns in a way to form a diagonal structure that is known as Dulmage–Mendelsohn
(DM) decomposition. Using the analytic redundant part of this structure and based on the
degree of the redundancy, several smaller sets of ARRs are identified. These smaller sets are
called minimally over-constrained sets and have one degree of redundancy, holding exactly
one more equation than the number of variables. Subsequently, a fault signature matrix is
formed to demonstrate which fault can be detected or even discriminated. Finally, specific
diagnostic tests (residuals) are designed to detect faults. Here, a structural analysis of a
PMSM experiencing independent ITSC faults in each phase is presented, and diagnostic
tests are proposed to detect and discriminate them. Figure 2 shows the modeling diagram
of a faulty PMSM and the drive system where measurements are acquired by sensors and
faults are located inside the motor.
Sensors 2022, 22, 3407 5 of 22
dλ a
e1 :v a = R a i a + + f va
dt
dλ
e2 :vb = Rb ib + b + f vb
dt
dλc
e3 :vc = Rc ic + + f vc
dt
e4 :λ a = L a i a + λm cos θ + f λa
e5 :λb = Lb ib + λm cos (θ − 2π/3)+ f λb (2)
e6 :λc = Lc ic + λm cos (θ + 2π/3)+ f λc
e7 :Te = − pλm [i a sin θ + ib sin (θ − 2π/3)
+ ic sin (θ + 2π/3)]
dωm 1
e8 : = ( Te − bωm − TL )
dt J
dθ
e9 : = pωm
dt
Sensors 2022, 22, 3407 6 of 22
The known variables consist of the motor signals, which are measured for both control
purposes and fault diagnosis. Thus, in addition to the three-phase currents and angular
position, i.e., yia , yib , yic , and yθ that are necessary for the control system. Three-phase
voltages, i.e., yva , yvb , and yvc , are also measured to complete the diagnostic system. Equa-
tion (3) shows these known variables, where f θ and f ω fault terms are also added to account
for speed and angle measurement error.
m1 : yva = v a m4 : yia = i a m7 : yθ = θ + f θ
m2 : yvb = vb m5 : yib = ib m8 : yωm = ωm + f ωm (3)
m3 : yvc = vc m6 : yic = ic
In addition, since the dynamic model of PMSM includes five differential constraints in
the abc frame, these are needed to be defined as unknown variables. Equation (4) shows
the differential constraints for the structural model.
dλ a d dωm d
d1 : = (λ a ) d4 : = ( ωm )
dt dt dt dt
dλ d dθ d
d2 : b = (λb ) d5 : = (θ ) (4)
dt dt dt dt
dλc d
d3 : = (λc )
dt dt
diagnostic test is a set of equations (or consistency relations) extracted from the system
model, in which at least one equation is violated in case of the presence of a considered
fault. A system model is called a redundant model if the system model consists of more
equations than unknown variables. Assuming that model M = (C, Z ) contains constraints
(equations) C and variables Z, let unknown variables X be the subset of all variables Z in
model M (X ⊆ Z). The degree of redundancy of the model M is defined as:
ϕ( M ) = |C | − | X | (5)
where |C | denotes the number of equations, and | X | is the number of unknown variables
contained in the model M. According to bipartite graph theory, any finite dimensional
graph such as M = (C, Z ) can be decomposed into three sub-graphs as follows [37]:
• M− : structurally under-determined part of the model M, where fewer equations than
unknown variables lie, and the degree of redundancy is negative ϕ( M ) < 0.
• M0 : structurally just-determined part of the model M, where equal equations and
unknown variables lie, and the degree of redundancy is zero ϕ( M ) = 0.
• M+ : structurally over-determined part of the model M, where more equations than
unknown variables lie, and the degree of redundancy is positive ϕ( M ) > 0.
For diagnostic purposes, the over-constrained sub-graph is the interesting part because
it contains the important redundancy that is necessary for detecting a fault. According
to [38], a fault is structurally detectable if the equation that contains the fault variable lies
in the over-determined part of the whole model (e f ∈ M+ ). To obtain these sub-graphs, a
canonical decomposition of the main structure graph (M) is required. An example of this
canonical decomposition is shown in Figure 4, where three canonical sub-models of the
model M are obtained as {M− , M0 , M+ }.
This canonical decomposition is achieved after the rows and the columns of the main
structural graph (structural model incidence matrix) are rearranged so that the matched
variables and constraints appear on the diagonal. Therefore, having a decomposition tool
that analyzes the redundancy of the structural model and forms this diagonal structure is
very beneficial. Dulmage–Mendelsohn (DM) is a key decomposition tool that is applied on
a structural model directly and obtains a unique diagonal structure by a clever reordering
of equations and variables [39]. Figure 5 shows the DM decomposition for the PMSM
structural model, where the analytic redundant part is expressed in the bottom-right
part containing all the faults. Since this part includes redundancy and can be monitored,
diagnostic tests can be designed with the set of ARRs in M+ . As a result, if a fault is defined
in the model and is supposed to be detected by the diagnosis system, a residual that is
sensitive to the presence of that fault must exist.
Sensors 2022, 22, 3407 8 of 22
n −1 n
1
mD =
n+1
∑ ∑ D ( Vf i , Vf j ) (6)
i =0 j = i +1
2
where D (Vf0 , Vf j ) stands for the distance between the fault signature of f j and the healthy
case and measures the detectability of fault f j . D (Vf i , Vf j ) is the distance between two
fault signatures and is defined as the Hamming distance [41] between the two fault signa-
ture strings:
S
D ( Vf i , Vf j ) = ∑ | Vf i − Vf j | (7)
n =1
As can be seen in the fault signature matrix in Figure 7, MTES7 − MTES10 can be used for
forming such a residual because these four MTES sets all contain f va and f λa fault terms.
Among them, an MTES set is preferred that contains a lower number of fault terms because
it will be more isolated and less influenced by other faults. MTES7 and MTES8 contain
three fault terms, while MTES9 and MTES10 contain four fault terms. Therefore, either
MTES7 or MTES8 should be chosen, and MTES8 is preferred due to a lower number of
involved equations (MTES8 contains six equations, while MTES7 contains eight equations)
which leads to less complexity, as seen in Figure 6. MTES4 − MTES6 can be used for
forming residual R2 because they contain f vb and f λb fault terms. Among them, MTES5
is preferred because it contains a lower number of fault terms compared to MTES6 and
a lower number of equations compared to MTES5 . Similarly, MTES3 is chosen to form
residual R3 that is sensitive to an ITSC fault in phase-c winning and contains a lower
number of equations compared to MTES2 , given the fact that both contain f vc and f λc
fault terms. To form residual R3 that is sensitive to an encoder fault (angular velocity and
position measurements), an MTES set is preferred that contains both f θe and f ωm , and the
only MTES set that contains such fault terms is MTES1 . The combination of these four
MTES sets, i.e., MTES1 , MTES3 , MTES5 , and MTES8 yield a high diagnosability index as
m D = 1.88, and this maximizes the chance of discrimination of each fault from others. The
sequential residuals are obtained as follows:
1. R1 : MTES8 is used for deriving R1 based on the error between calculated and mea-
sured current of phase a winding, i.e m4 in (3):
m4 : R1 = i a − y i a (8)
SV1 : λ a = λ astate
m7 : θ = y θ
m1 : v a = y v a (9)
1
e4 : i a = (λ a − λm cos θ )
La
e1 : dλ a = v a − R a i a
2. R2 and R3 follow the same procedure mentioned for R1 based on the error between
calculated and measured currents of phase b and phase c using MTES5 and MTES3 ,
respectively.
3. R4 : MTES1 is used for deriving R4 based on the error between the calculated and
measured shaft’s angular speed, i.e m8 in (3):
m8 : R4 = ω m − y ωm (11)
m7 : θ = y θ
dθ d
d5 : = (θ ) (12)
dt dt
dθ
e9 : ω m = p
dt
(13)
Sensors 2022, 22, 3407 11 of 22
0.5
R1 (A)
0
-0.5
-1
0 5 10 15 20
time (s)
0.5
R2 (A)
-0.5
-1
0 5 10 15 20
time (s)
0.5
R3 (A)
-0.5
-1
0 5 10 15 20
time (s)
1
R4 (rad/s)
-1
-2
0 5 10 15 20
time (s)
Figure 10. Residual responses in abc phases.
6. Diagnostic Decision
Using the residual responses, a diagnostic decision making system was designed to
detect the ITSC faults based on statistical signal processing–detection theory. While R4 can
be directly used to detect encoder faults, a combination of R1 –R3 is required to effectively
detect ITSC faults. The R1 –R3 residuals obtained in the previous section, are designed
based on abc frame voltage equations e1 –e3 in (2), and an ITSC fault in any phase creates
unbalance in the residual output. Before designing the statistical detector and to form a
better index that obtains a nonzero dc value in case of an ITSC fault, the residuals in the
abc frame are taken into an αβ frame using the power invariant Clarke transformation
as follows:
Sensors 2022, 22, 3407 14 of 22
r " # R
1
− −√21 1
Rα 2 1 √2
= 3 R2 (14)
Rβ 3 0 2 − 23 R
3
Rr = | Rα + jR β | (15)
Figure 11 shows the absolute value of the resultant residual in an αβ frame where
ITSC faults in all phases are more obvious compared to abc residuals R1 –R3 .
In implementing a structural analysis, the goal was to form residuals that have a zero
value in a healthy scenario and a nonzero value in a faulty scenario. However, derivatives,
integrals, and even uncertainties in the dynamic model affect the calculation of unknown
variables and cause the variable output signal to be a little bit distorted. In addition,
phenomena such as environmental noise and switching noise affect the signals. These lead
to a residual output signal that fluctuates around zero instead of having a perfect signal
that holds the absolute zero value in a healthy scenario. Even in a faulty scenario, the
residual signal fluctuates around a nonzero value as seen in Figure 11. Therefore, extra
signal processing is required to deal with model uncertainties and environmental noise and
to be able to distinguish and isolate the indicator signal from noise. Here, a generalized
likelihood ratio test (GLRT) is proposed to deal with such model uncertainties and also
to provide the ground for calculating and setting thresholds based on the probabilities of
detection and false alarms in a formulated and scientific manner.
p( x; θˆ1 , H1 )
LG ( x) = >γ (16)
p( x; θˆ0 , H0 )
where θˆ1 is the MLE of θ1 assuming H1 is true, θˆ0 is the MLE of θ0 assuming H0 is true, and
γ is the threshold.
Sensors 2022, 22, 3407 15 of 22
30
Residual
25
WGN
20
PDF (x)
15
10
0
-0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06
x
Figure 12. Comparison of PDF of residual and WGN.
To design a realistic detector, it is assumed that the arrival time of the fault is completely
unknown. Furthermore, the PDF is not completely known, meaning that the parameters
mean µ and variance σ2 are to be estimated using MLE. The noise in the resultant residual
during operation in a healthy condition is modeled as WGN. Since the resultant residual
(Rr ) obtains a nonzero dc value when ITSC faults appear, the data are considered as only
noise under nonfaulty hypothesis H0 , and an added dc level value to the noise under faulty
hypothesis H1 . Thus, the detection problem becomes as follows:
H0 : x [ n ] = w [ n ] n = 0, 1, ..., N − 1 (17)
H1 : x [ n ] = A + w [ n ] n = 0, 1, ..., N − 1
where A is unknown amplitude with −∞ < A < ∞, and w[n] is WGN with unknown
positive variance 0 < σ2 < ∞. The GLRT decides H1 if:
p( x; Â, σˆ1 2 , H1 )
LG ( x) = >γ (18)
p( x; σˆ0 2 , H0 )
where  and σˆ1 2 are the MLE of parameters A and σ12 under H1 , and σˆ0 2 is the MLE of the
parameter σ02 under H0 . By maximizing p( x; A, σ2 , H1 ), parameters  and σˆ1 2 are obtained
as follows [43]:
N −1
1 1
p( x; A, σ2 , H1 ) = N
exp[−
2σ2 ∑ ( x [ n ] − A )2 ]
(2πσ2 ) 2 N =0
∂p( x; A, σ2 , H1 )
= 0 ⇒ Â = x̄ (19)
∂A
N −1
∂p( x; A, σ2 , H1 ) 1
∂σ12
= 0 ⇒ σˆ1 2 =
N ∑ ( x [ n ] − A )2
N =0
Sensors 2022, 22, 3407 16 of 22
N −1
1 1
p( x; σ2 , H0 ) = N
exp(−
2σ2 ∑ x2 [n])
(2πσ2 ) 2 N =0
N −1
∂p( x; σ2 , H 0) 1
∂σ02
= 0 ⇒ σˆ0 2 =
N ∑ x2 [n] (21)
N =0
σˆ0 2 N
LG ( x) = ( )2 (23)
σˆ1 2
σˆ0 2
2lnLG ( x ) = Nln (24)
σˆ1 2
N −1 N −1
1 1
σˆ1 2 =
N ∑ ( x [ n ] − A )2 =
N ∑ (x[n] − x̄)2
N =0 N =0
N −1 N −1
1 1
=
N ∑ (x[n]2 − 2x[n]x̄ + x̄2 ) = N ∑ x [n]2 − x̄2
N =0 N =0
2 2
= σˆ0 − x̄ (25)
which yields:
x̄2
2lnLG ( x ) = Nln(1 + ) (26)
σˆ1 2
x̄2 x̄2
Since ln(1 + ) is monotonically increasing with respect to , an equivalent and
σˆ1 2 σˆ1 2
normalized test statistic can be obtained as follows:
x̄2
T (x) = > γ0 (27)
σˆ1 2
The GLRT has normalized the statistic by σˆ1 2 which allows the threshold to be de-
termined. Since the PDF of T ( x ) under null hypothesis H0 does not depend on σ2 , the
threshold is independent of the value σ2 [42].
Sensors 2022, 22, 3407 17 of 22
and therefore:
(
x̄ N (0, 1) under H0
∼ (29)
σ N ( Aσ , 1) under H1
Squaring the normalized statistic in (29) will lead to the modified test statistic T ( x )
in (27) which produces a central chi-squared distribution under H0 and a noncentral
chi-squared distribution under H0 , with one degree of freedom:
(
x̄2 X12 under H0
T (x) = 2 ∼ 0 2 (30)
σ X 1 (λ) under H1
A2 x̄2
λ= 2
= 2 (31)
σ σ
It was shown in (30) that T ( x ) has a noncentral chi-squared distribution with one
√ freedom, and it is equal to the square of random variable x in (29), therefore
degree of
x ∼ N ( λ, 1). Thus, the probability of a false alarm (PFA ) can be obtained as:
PFA = Pr { T ( x ) > γ0 ; H0 }
p p
= Pr { x > γ0 ; H0 } + Pr { x < − γ0 ; H0 }
p
= 2Q( γ0 ) (32)
where Q( x ) is the right-tail probability of random variable x. Thus, the threshold can be
obtained as follows:
PFA 2
γ 0 = [ Q −1 ( )] (33)
2
Similarly, the probability of detection PD can be obtained as follows:
PD = Pr { T ( x ) > γ0 ; H1 }
p p
= Pr { x > γ0 ; H1 } + Pr { x < − γ0 ; H1 }
p √ p √
= Q( γ0 − λ) + Q( γ0 + λ)
P √ P √
= Q( Q−1 ( FA ) − λ) + Q( Q−1 ( FA ) + λ) (34)
2 2
have lower PFA values. Using PFA = 2%, the threshold is obtained as γ0 = 5.41, and this
results in PD = 60.93% for ITSC in phase a. Furthermore, the probability of detection for
ITSC faults in phases b and c and the encoder fault are calculated PD = 98.13%, PD = 100%,
and PD = 99.65%, respectively.
The test statistics were implemented on the resultant residual as shown in Figure 14.
The values x̄2 and σˆ1 2 were calculated using a moving window (FIFO register) with the
length of N = 10, 000, which runs through the resultant residual over time. Figure 14a
shows the output of test statistic on resultant residual along with the threshold of γ0 = 5.41
while Figure 14b shows the output of the test statistic on R4 . The test statistic’s output value
is compared with the threshold value over time, and if it exceeds the threshold, the fault
alarm is tripped accordingly. Figure 15 shows the detector’s logical output value which
attains a low value in a healthy condition and a high value during a faulty case. This proves
that the detector has successfully detected all the faults that are fairly close to expected
values of PD , while experiencing no false alarm.
20
15
Threshold
10
0
0 0.01 0.02 0.03 0.04 0.05
P FA
ITSC a
PD
0.5
ITSC b
ITSC
c
Encoder
0
0 0.01 0.02 0.03 0.04 0.05
P FA
60
50
30
20
10
0
0 5 10 15 20
time (s)
(a)
50
Encoder Fault Index
40
30
20
10
0
0 5 10 15 20
time (s)
(b)
Figure 14. Test statistic for ITSC fault and encoder faults.
1
Denc
Diagnostic Decision
0.8
DITSC
0.6 f enc
f ITSC
a
0.4
f ITSC
b
0.2 f ITSC
c
0
0 5 10 15 20
time (s)
Figure 15. Timing of actual faults ( f enc , f ITSCa , f ITSCb , f ITSCc ) versus diagnostic system’s decision
(Denc , D ITSC ).
7. Discussion
Some remarks can be withdrawn regarding the presented methodology and the ob-
tained results. First, structural analysis for detecting ITSC and encoder faults was success-
fully implemented on the in-house setup including the PMSM and the drive system, and
the residuals were formed based on ARRs. Second, a GLRT-based detector was designed to
effectively detect the changes in the residuals even with unknown noise parameters. Third,
Sensors 2022, 22, 3407 20 of 22
a scientific threshold was calculated based on the probability of a false alarm (PFA ) and the
probability of detection (PD ). The suggested combination method is very effective for the
fault detection since it can detect the lowest level of ITSC fault, i.e., one single shorted turn
(<1%) in the stator winding. On the other hand, using a Clarke transformation disabled
the diagnostic system to isolate the ITSC faults in different phases, and using a moving
window with the length of N = 10, 000 over the test statistics causes a delay in detection of
the faults. These small demerits were found when testing the diagnostic method under the
smallest ITSC fault.
In previous studies, a GLRT-based detector has been implemented for stator imbal-
ancefault detection in induction motors [44]. The noise parameters were also considered
unknown, and therefore, they have been replaced with their MLEs. Moreover, a threshold
was calculated based on PFA = 0.1% and PD , which makes the diagnostic system experi-
ence fewer false alarms. However, the first fault level that the system can detect is 25%
of stator-phase resistance, which is a quite high level of fault severity. As a result, the
system would go into severe imbalance from the time that the fault appears until the time
the diagnostic system detects it. In our case, even if the PFA was chosen as 0.1%, the PD
for ITSC in phase a would be 24.61%, the PD for ITSC in phase b would be 86.8%, the PD
for ITSC in phase c would be 99.99%, and the PD for the encoder fault would be 95.86%.
Thus, the diagnostic system still detects the smallest fault, even with PFA = 0.1%. However,
knowing that a slightly higher probability of a false alarm is not that irritating (PFA = 2%),
a better probability of detection is achieved (PD = 60.93%) in our study based on setting
a lower threshold. Other studies with different methods have also chosen a higher level
of fault as the starting point. A Kalman filter for detection of ITSC in PM synchronous
generators has been implemented in [45], which can successfully detect fault levels as low
as 8%. In addition, a combination of an extended Park’s vector approach with spectral
frequency analysis was introduced in [46] which could successfully detect three shorted
turns in synchronous and induction motors.
8. Conclusions
This paper presents a novel method for real-time and effective detection of incipient
ITSC and encoder faults in the PMSM. Structural analysis was employed to form the
structural model of the PMSM. The Dulmage–Mendelsohn decomposition tool was used
to evaluate the analytical redundancy of the structural model. The proposed diagnostic
model was implemented on industrial PMSM, ITSC, and encoder faults were applied to the
system in different time intervals, and residuals responses were obtained. Subsequently, a
GLRT-based detector was designed and implemented based on the behavior of the residuals
during healthy (only noise) and faulty (noise + signal) conditions. To make the GLRT-based
detector effective to deal with such a realistic problem, the parameters such as mean µ
and variance σ2 in the probability density function of the noise signal were considered
to be unknown. By replacing these unknown parameters by their maximum likelihood
estimates, a test statistic was achieved for the GLRT-based ITSC and encoder fault detector.
Following this step, a threshold was obtained based on choosing the probability of a false
alarm PFA and the probability of detection PD for each detector based on which decision
was made to indicate the presence of the fault. The experimental results show that the
designed GLRT-based detector is able to efficiently detect even small ITSC and encoder
faults in the presence of noise, proving the effectiveness of this diagnostic approach.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author. The data are not publicly available due to ongoing research within Ph.D. program.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Hang, J.; Wu, H.; Zhang, J.; Ding, S.; Huang, Y.; Hua, W. Cost function-based open-phase fault diagnosis for PMSM drive system
with model predictive current control. IEEE Trans. Power Electron. 2020, 36, 2574–2583. [CrossRef]
2. Wang, X.; Wang, Z.; Xu, Z.; He, J.; Zhao, W. Diagnosis and tolerance of common electrical faults in T-type three-level inverters fed
dual three-phase PMSM drives. IEEE Trans. Power Electron. 2019, 35, 1753–1769. [CrossRef]
3. Zeng, C.; Huang, S.; Yang, Y.; Wu, D. Inter-turn fault diagnosis of permanent magnet synchronous machine based on tooth
magnetic flux analysis. IET Electr. Power Appl. 2018, 12, 837–844. [CrossRef]
4. Ebrahimi, B.M.; Roshtkhari, M.J.; Faiz, J.; Khatami, S.V. Advanced eccentricity fault recognition in permanent magnet synchronous
motors using stator current signature analysis. IEEE Trans. Ind. Electron. 2013, 61, 2041–2052. [CrossRef]
5. Awadallah, M.A.; Morcos, M.M.; Gopalakrishnan, S.; Nehl, T.W. Detection of stator short circuits in VSI-fed brushless DC motors
using wavelet transform. IEEE Trans. Energy Convers. 2006, 21, 1–8. [CrossRef]
6. Cruz, S.M.; Cardoso, A.M. Multiple reference frames theory: A new method for the diagnosis of stator faults in three-phase
induction motors. IEEE Trans. Energy Convers. 2005, 20, 611–619. [CrossRef]
7. Vaseghi, B.; Nahid-Mobarakh, B.; Takorabet, N.; Meibody-Tabar, F. Inductance identification and study of PM motor with
winding turn short circuit fault. IEEE Trans. Magn. 2011, 47, 978–981. [CrossRef]
8. Choi, S.; Haque, M.S.; Tarek, M.T.B.; Mulpuri, V.; Duan, Y.; Das, S.; Garg, V.; Ionel, D.M.; Masrur, M.A.; Mirafzal, B.; et al. Fault
diagnosis techniques for permanent magnet ac machine and drives—A review of current state of the art. IEEE Trans. Transp.
Electrif. 2018, 4, 444–463. [CrossRef]
9. Liang, H.; Chen, Y.; Liang, S.; Wang, C. Fault detection of stator inter-turn short-circuit in PMSM on stator current and vibration
signal. Appl. Sci. 2018, 8, 1677. [CrossRef]
10. Seshadrinath, J.; Singh, B.; Panigrahi, B.K. Vibration analysis based interturn fault diagnosis in induction machines. IEEE Trans.
Ind. Inform. 2013, 10, 340–350. [CrossRef]
11. Hang, J.; Zhang, J.; Cheng, M.; Huang, J. Online interturn fault diagnosis of permanent magnet synchronous machine using
zero-sequence components. IEEE Trans. Power Electron. 2015, 30, 6731–6741. [CrossRef]
12. Zanardelli, W.G.; Strangas, E.G.; Aviyente, S. Identification of intermittent electrical and mechanical faults in permanent-magnet
AC drives based on time—Frequency analysis. IEEE Trans. Ind. Appl. 2007, 43, 971–980. [CrossRef]
13. Akin, B.; Choi, S.; Orguner, U.; Toliyat, H.A. A simple real-time fault signature monitoring tool for motor-drive-embedded fault
diagnosis systems. IEEE Trans. Ind. Electron. 2010, 58, 1990–2001. [CrossRef]
14. Espinosa, A.G.; Rosero, J.A.; Cusido, J.; Romeral, L.; Ortega, J.A. Fault detection by means of Hilbert–Huang transform of the
stator current in a PMSM with demagnetization. IEEE Trans. Energy Convers. 2010, 25, 312–318. [CrossRef]
15. Obeid, N.H.; Battiston, A.; Boileau, T.; Nahid-Mobarakeh, B. Early intermittent interturn fault detection and localization for a
permanent magnet synchronous motor of electrical vehicles using wavelet transform. IEEE Trans. Transp. Electrif. 2017, 3, 694–702.
[CrossRef]
16. Rosero, J.A.; Romeral, L.; Ortega, J.A.; Rosero, E. Short-circuit detection by means of empirical mode decomposition and
Wigner–Ville distribution for PMSM running under dynamic condition. IEEE Trans. Ind. Electron. 2009, 56, 4534–4547. [CrossRef]
17. Liu, X.Q.; Zhang, H.Y.; Liu, J.; Yang, J. Fault detection and diagnosis of permanent-magnet DC motor based on parameter
estimation and neural network. IEEE Trans. Ind. Electron. 2000, 47, 1021–1030.
18. Awadallah, M.; Morcos, M. Diagnosis of stator short circuits in brushless dc motors by monitoring phase voltages. IEEE Trans.
Energy Convers. 2005, 20, 246–247. [CrossRef]
19. Nyanteh, Y.; Edrington, C.; Srivastava, S.; Cartes, D. Application of artificial intelligence to real-time fault detection in permanent-
magnet synchronous machines. IEEE Trans. Ind. Appl. 2013, 49, 1205–1214. [CrossRef]
20. Mazzoletti, M.A.; Bossio, G.R.; De Angelo, C.H.; Espinoza-Trejo, D.R. A model-based strategy for interturn short-circuit fault
diagnosis in PMSM. IEEE Trans. Ind. Electron. 2017, 64, 7218–7228. [CrossRef]
21. Forstner, G.; Kugi, A.; Kemmetmüller, W. A magnetic equivalent circuit based modeling framework for electric motors applied to
a pmsm with winding short circuit. IEEE Trans. Power Electron. 2020, 35, 12285–12295. [CrossRef]
22. Fitouri, M.; Bensalem, Y.; Abdelkrim, M.N. Modeling and detection of the short-circuit fault in PMSM using Finite Element
Analysis. IFAC-PapersOnLine 2016, 49, 1418–1423. [CrossRef]
23. Krysander, M. Design and Analysis of Diagnosis Systems Using Structural Methods. Ph.D. Thesis, Department of Electrical
Engineering, Linköping University, Linköping, Sweden, 2006.
24. Krysander, M.; Åslund, J.; Nyberg, M. An efficient algorithm for finding minimal overconstrained subsystems for model-based
diagnosis. IEEE Trans. Syst. Man Cybern.-Part A Syst. Hum. 2007, 38, 197–206. [CrossRef]
25. Krysander, M.; Frisk, E. Sensor placement for fault diagnosis. IEEE Trans. Syst. Man Cybern.-Part A Syst. Hum. 2008, 38, 1398–1410.
[CrossRef]
Sensors 2022, 22, 3407 22 of 22
26. Svard, C.; Nyberg, M. Residual generators for fault diagnosis using computation sequences with mixed causality applied to
automotive systems. IEEE Trans. Syst. Man Cybern.-Part A Syst. Hum. 2010, 40, 1310–1328. [CrossRef]
27. Svärd, C.; Nyberg, M.; Frisk, E. Realizability constrained selection of residual generators for fault diagnosis with an automotive
engine application. IEEE Trans. Syst. Man Cybern. Syst. 2013, 43, 1354–1369. [CrossRef]
28. Svärd, C.; Nyberg, M.; Frisk, E.; Krysander, M. Automotive engine FDI by application of an automated model-based and
data-driven design methodology. Control. Eng. Pract. 2013, 21, 455–472. [CrossRef]
29. Sundström, C.; Frisk, E.; Nielsen, L. Selecting and utilizing sequential residual generators in FDI applied to hybrid vehicles. IEEE
Trans. Syst. Man Cybern. Syst. 2013, 44, 172–185. [CrossRef]
30. Liu, Z.; Ahmed, Q.; Rizzoni, G.; He, H. Fault detection and isolation for lithium-ion battery system using structural analysis and
sequential residual generation. In Proceedings of the ASME 2014 Dynamic Systems and Control Conference, American Society of
Mechanical Engineers Digital Collection, San Antonio, TX, USA, 22–24 October 2014.
31. Zhang, J.; Yao, H.; Rizzoni, G. Fault diagnosis for electric drive systems of electrified vehicles based on structural analysis. IEEE
Trans. Veh. Technol. 2016, 66, 1027–1039. [CrossRef]
32. Ebrahimi, S.H.; Choux, M.; Huynh, V.K. Diagnosis of Sensor Faults in PMSM and Drive System Based on Structural Analysis. In
Proceedings of the 2021 IEEE International Conference on Mechatronics (ICM), Kashiwa, Japan, 7–9 March 2021; pp. 1–6.
33. Ebrahimi, S.H.; Choux, M.; Huynh, V.K. Detection and Discrimination of Inter-Turn Short Circuit and Demagnetization Faults in
PMSMs Based on Structural Analysis. In Proceedings of the 2021 22nd IEEE International Conference on Industrial Technology
(ICIT), Valencia, Spain, 10–12 March 2021; Volume 1, pp. 184–189.
34. Wang, B.; Wang, J.; Griffo, A.; Sen, B. Stator turn fault detection by second harmonic in instantaneous power for a triple-redundant
fault-tolerant PM drive. IEEE Trans. Ind. Electron. 2018, 65, 7279–7289. [CrossRef]
35. Ebrahimi, S.H.; Choux, M.; Huynh, V.K. Modeling Stator Winding Inter-Turn Short Circuit Faults in PMSMs including Cross
Effects. In Proceedings of the 2020 International Conference on Electrical Machines (ICEM), Gothenburg, Sweden, 23–26 August
2020; Volume 1, pp. 1397–1403.
36. Bouzid, M.B.K.; Champenois, G. New expressions of symmetrical components of the induction motor under stator faults. IEEE
Trans. Ind. Electron. 2012, 60, 4093–4102. [CrossRef]
37. Blanke, M.; Kinnaert, M.; Lunze, J.; Staroswiecki, M. Diagnosis and Fault Tolerant Control; Springer: Berlin, Germany, 2006.
38. Krysander, M.; Åslund, J.; Frisk, E. A structural algorithm for finding testable sub-models and multiple fault isolability analysis.
In Proceedings of the 21st International Workshop on Principles of Diagnosis (DX-10), Portland, OR, USA, 13–16 October 2010;
pp. 17–18.
39. Dulmage, A.L.; Mendelsohn, N.S. Coverings of bipartite graphs. Can. J. Math. 1958, 10, 517–534. [CrossRef]
40. Zhang, J.; Rizzoni, G. Selection of residual generators in structural analysis for fault diagnosis using a diagnosability index. In
Proceedings of the 2017 IEEE Conference on Control Technology and Applications (CCTA), Mauna Lani Resort, HI, USA, 27–30
August 2017; pp. 1438–1445.
41. Hamming, R.W. Error detecting and error correcting codes. Bell Syst. Tech. J. 1950, 29, 147–160. [CrossRef]
42. Kay, S.M. Fundamentals of Statistical Signal Processing, Volume II: Detection Theory; Prentice Hall PTR: Hoboken, NJ, USA, 1998.
43. Kay, S.M. Fundamentals of Statistical Signal Processing, Volume I: Estimation Theory; Prentice Hall PTR: Hoboken, NJ, USA, 1993.
44. Elbouchikhi, E.; Amirat, Y.; Feld, G.; Benbouzid, M. Generalized likelihood ratio test based approach for stator-fault detection in
a pwm inverter-fed induction motor drive. IEEE Trans. Ind. Electron. 2018, 66, 6343–6353. [CrossRef]
45. Aubert, B.; Regnier, J.; Caux, S.; Alejo, D. Kalman-filter-based indicator for online interturn short circuits detection in permanent-
magnet synchronous generators. IEEE Trans. Ind. Electron. 2014, 62, 1921–1930. [CrossRef]
46. Cruz, S.M.; Cardoso, A.M. Stator winding fault diagnosis in three-phase synchronous and asynchronous motors, by the extended
Park’s vector approach. IEEE Trans. Ind. Appl. 2001, 37, 1227–1233. [CrossRef]