IET Power Electronics

Research Article

Detection and identification of multiple IGBT ISSN 1755-4535

Received on 25th March 2018
Revised 27th November 2018
open-circuit faults in PWM inverters for AC Accepted on 23rd December 2018
E-First on 30th January 2019
machine drives doi: 10.1049/iet-pel.2018.5188
www.ietdl.org

Cheol-Ju Bae1, Dong-Choon Lee2 , Thanh Hai Nguyen3

1ElectricalEquipment Development Team, R&D Center, Hyundai Rotem Co., Republic of Korea
2Department of Electrical Engineering, Yeungnam University, Republic of Korea
3Department of Electrical and Computer Engineering, Khalifa University of Science and Technology, United Arab Emirates

E-mail: [email protected]

Abstract: This study proposes a new scheme for online detection and identification of single and multiple insulated-gate bipolar
transistor (IGBT) open-circuit faults in three-phase pulse-width modulation (PWM) inverters of AC machine drives, which is
based on Park's vector of machine phase currents. First, the switch open-circuit faults are detected by monitoring dwell times of
the vector of the actual phase currents expressed in the stationary reference frame in each sector, where the normalisation of
the current to its reference value is utilised, by which the immunity to the false alarms in transient conditions is improved
significantly. Then, the IGBT open-circuit faults are identified, where leg open-circuit conditions are in the first place determined
based on the average values of the normalised line-to-line currents. Next, the single open-circuit IGBT is located from the
polarity of the corresponding phase current. By the proposed diagnosis method, the fault detection time is at the longest about
52.3 and 54% of a fundamental period in the simulation and experimental tests, respectively, and the 27 possible cases of IGBT
open-circuit faults combined for all IGBTs in the inverter can be identified. In addition, this algorithm can be implemented without
any additional hardware and computational burden.

1 Introduction analysis, and support vector machine have been utilised for
knowledge-based methods [11–13]. Although these techniques can
Three-phase two-level voltage-source pulse-width modulation overcome the parameter dependency, it is not easy to implement
(PWM) inverters are most commonly used in industrial AC due to the complex algorithm and computation of a large amount of
machine drives. Although the design and control techniques of the historic data.
PWM inverters have been significantly improved to achieve high For signal-based approaches, several diagnosis schemes
reliability, the inverters are still subject to the unexpected faults [1]. utilising voltage signals have been proposed for the open-circuit
It has been reported in many published research results and fault detection, which give a fast detection speed and independency
practical surveys of commercial products that the reliability of on the system parameters [14–19]. However, these methods usually
power switches together with DC capacitors is remarkably require additional voltage sensors. To the contrary, it is desirable to
concerned for obtaining the stable and safe operation and avoiding use the existing current sensors available in the motor drive
malfunction of the drive systems [2, 3]. systems for the fault diagnosis. The current-signal-based method is
The short- and open-circuit faults are generally considered as relatively easy to implement since they usually require the current
the common failures of the semiconductor devices. The insulated- signals only [20–29]. However, this method may generate false
gate bipolar transistor (IGBT) short circuits are the catastrophic alarms under the transient states or different operating speeds and
failures, which immediately trip or damage the system due to loads of the motor drive system. A simple, direct current method
extremely high values of inrushing fault currents produced by the and the normalised DC current method have been proposed for
short-circuit condition. Therefore, a protection circuit is usually avoiding the false alarms in the transient conditions [20], which are
integrated into the inverter systems, which is a standard feature for based on direct component and normalisation of phase currents. A
safety. On the other hand, the protection for the open-circuit faults diagnosis technique of IGBT faults based on the current vector
is mandatory to be included in the system, since the faults do not trajectory, which is distorted in the cases of switch faults, has been
immediately cause the system to shut-down and the inverter may studied in [21, 22]. However, in spite of diagnosing capability for
still operate for a while. In this case, however, the overstress such multiple IGBT faults, these methods have some shortcomings such
as excessive electrical and thermal stresses leads to the secondary as parameter tuning, high computational burden, and inexactness at
faults in other devices. Thereafter, the system will stop eventually, low current levels. Another diagnosis scheme which utilises the
which incurs higher costs for repairing. So, the online diagnostic amplitude of the second-order harmonic current component was
techniques for open-circuit faults of the IGBTs have been introduced for the linear inductor motor drive system [23], which is
introduced for the last decades [4–29]. not influenced by the steady-state error and DC offset of the
The fault diagnosis methods can be typically categorised into currents. However, this technique also requires a large
model-based, knowledge-based, and signal-based schemes [5]. The computational effort, where the average and root-mean-square
model-based method utilises the observers such as Kalman filter, values of the fundamental and second-order harmonic current
parity relation, model reference adaptive system (MRAS), mixed components.
logical dynamic (MLD) model and so on [6–10], where the Recently, a diagnostic algorithm based on the form factors of
diagnosis procedure is performed by monitoring the residual three-phase currents has been proposed in [24, 29], where the
signals between the measured outputs of the actual system and the current form factor and the current model based on Luenberger
estimated variables of the predefined model. A demerit of these observer are used to generate the detection variable and threshold
schemes is that the diagnosis performance heavily depends on the value, respectively. However, the detection algorithm is very
accuracy of model parameters. The intelligent control theories such complex and requires a remarkable computation effort. In addition,
as the neural networks, fuzzy logic, kernel principal component new approaches for multiple IGBT fault diagnosis based on the

IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931 923

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
where ia and ib are two phase currents of the motor, which then its
phase angle is determined as

iα_ f
θ = tan−1 − +π (2)
iβ_ f

where iα_fand iβ_f are the αβ-axis components, respectively, filtered

by low-pass filters (LPF) with a cut-off frequency of 200 Hz. The
sector number, S, based on the phase angle of the current vector is
defined as

3θ
S = int +1 (3)
π

where the operator of ‘int’ returns its function into an integer value.
When the motor is operated at the normal condition at a
constant speed as well as a constant frequency of phase currents,
the dwell times of the current vector in all the sectors are the same.
However, under the fault conditions, the change of the phase angle
Fig. 1 Three-phase PWM inverters for IM drives of the current vector will not be kept constant according to the rule
in (3). Thus, the fault can be detected by monitoring the dwell time
derivative of the phase angle of Park's vector have been introduced in each sector. A variable, K, is defined for the fault detection as
[25, 26]. However, this method may lead the false alarms for the
condition of operating speeds crossing zero. K(i − 1) + 1 if S(i) = S(i − 1)
In this paper, a new online diagnosis scheme for single and K= (4)
0 if S(i) ≠ S(i − 1)
multiple IGBT open-circuit faults is proposed for the three-phase
induction motor drive systems. At first, the algorithm for fault where the subscript i indicates the ith sampling instant. It is seen
detection is performed by analysing the dwell times of the vector of that the detection variable K based on the sector dwell time is
the actual phase currents expressed in the stationary reference updated and checked every one-sixth of the fundamental cycle, so
frame in each sector, where the detection variable is calculated it provides fast detection of the fault occurrence. The detection
from the actual currents and normalised ones with the motor variable denoted as K̄ is defined as the average value of the K.
current references. This helps to achieve a reliable fault diagnosis Under the transient states or different motor speed conditions, the
without generating false alarms for abrupt variations of loads. To frequency of the phase currents may change, resulting in a
localise the faulty IGBTs, the identification variables calculated
variation of K̄ , which may cause false alarms in such cases. So, the
from the average values of the absolute normalised line-to-line
detection variable should be normalised, named as a normalised
currents are determined, by which the leg open-circuit faults are
detection variable, KN, which is calculated from
identified. Then, the single open-circuit IGBT is located from the
polarity of the corresponding phase current. A combination of the
identification variables for the leg and single IGBT open-circuit K̄
KN = (5)
faults is used to localise 27 possible cases of IGBT open-circuit K̄*
faults in the inverters. The simulation and experimental results
have verified the feasibility of the proposed diagnosis technique. where the variable K̄* is calculated in the same way with K̄ as in
(2)–(4), which are based on the αβ-axis components of the current
2 Proposed fault diagnosis technique references. For a high-performance current regulator of the IM
drive with a high control bandwidth, it can be reasonably assumed
A circuit and control block diagram for induction machine (IM) that the actual current tracks its reference closely even in the
drives is shown in Fig. 1, where field-oriented vector control is transient states such as a start-up case and a step change of loads or
employed. The proportional and integral (PI) regulators are used speed references [30, 31]. Thus, the normalised detection variable
for the speed and current control, where the space-vector PWM is is kept around the value of unity in the normal operation.
utilised to generate the gating signals for the IGBTs of the inverter. Under the IGBT open-circuit faults, the values of K̄ are
The proposed fault diagnosis technique is performed in two significantly increased since the dwell time in a particular sector is
steps. At first, the detection for the open-circuit faults of the longer, while K̄* is kept the same due to unchanged current
switches is carried out, where the sensed motor phase currents and references. Then, the normalised detection variable is also
their references expressed in the stationary reference frame are increased higher than a threshold value. From the analysis
required. Then, the localisation of faulty IGBTs is performed, mentioned above, the fault detection can be determined by
where leg open-circuit and single IGBT faults are identified based comparing the KN with the threshold value, where the IGBT
on the normalised information and the polarity of the output phase
condition is considered as open-circuit faults if the KN is higher
currents, respectively. It is seen from Fig. 1 that the fault detection
and identification scheme is performed independently from the than the threshold value th1. Fig. 2 shows the block diagram of the
controllers, so it will not have an impact on the machine control proposed fault detection method. In this work, several threshold
performance. values are used to detect the fault and identify the faulty switches,
where the threshold values will be determined in the following
2.1 Detecting the faults subsections.

The components of iα and iβ in the αβ-stationary reference frame 2.2 Fault identification
can be expressed from the three-phase currents as
When the IGBT open-circuit fault is detected, faulty IGBTs should
iα = ia be identified and replaced to prevent further damage for other
devices or malfunction of the drive system. Based on the shapes of
ia + 2ib (1) phase currents under different open-circuit faults, a leg open-circuit
iβ =
3 fault, where two IGBTs in the same leg are faulty simultaneously,
is determined in the first place. The following analysis provides a
comprehensive way to localise the open-circuit phase.

924 IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 2 Block diagram of the proposed fault detection scheme

where l, m ∈ {a, b, c}, l ≠ m ≠ n. Then, the three-phase normalised

currents become 0, ±( 3/2), and ±( 3/2) for phases n, m, and l,
respectively. Similarly, the normalised currents for other cases of
open-circuit faults in legs m and l are obtained, where the
magnitude of the current vector becomes
2 2
i
3 n
= i
3 l
if the fault in leg m occurs
iαβ = (10)
2 2
i
3 n
= i
3 m
if the fault in leg l occurs

Based on the analysis aforementioned, the identification variables

for phase open-circuit faults, Pn, are defined in (11), which are
based on the average of the absolute value of the normalised line-
to-line currents calculated from the other two phase currents as

Pn = ∥ f abs(imN − ilN ) ∥ − 1.1 (11)

Fig. 3 Current waveforms under the normal and fault conditions
(a) Three-phase currents, (b) Three-phase normalised currents, (c) Absolute where l≠m≠n and ‘fabs’ is the absolute operator. In this work, the
normalised line-to-line currents line-to-line currents are utilised to eliminate the effect of the
common-mode current component, which may exist in the phase
2.2.1 Localisation for the leg open-circuit faults: It is assumed currents of the machine drive systems [32].
that the three phase currents are sinusoidal and balanced under Fig. 3 shows the current variations under the normal and fault
normal operations and the measured motor currents do not contain conditions, where the detailed simulation condition will be
any dc-offset value. Then, it can be expressed as described in Section 3. Under the normal condition, the three-phase
currents are balanced and the magnitude of iαβ is constant as
ia = IM sin(ωet + ϕ) shown in Fig. 3a. Similarly, the normalised phase currents are also
balanced with a unity magnitude as shown in Fig. 3b. Fig. 3c
ib = IM sin(ωet − 2π/3 + ϕ) (6) shows the absolute normalised line-to-line currents, which are non-
ic = IM sin(ωet + 2π/3 + ϕ) negative and have the magnitude of 3. Next, an IGBT open-
circuit fault is investigated, where the phase with the faulty IGBT
where ωe is a current angular frequency, ϕ is an initial phase angle, just conducts the current for a half cycle, while the current is
and IM is a magnitude of the phase currents. During the normal almost zero during the other half cycle as shown in Fig. 3a. With
operation, the magnitude of the phase current is expressed by this type of fault, the waveforms of the normalised phase currents
and the absolute normalised line-to-line currents are shown in the
middle interval in Figs. 3b and c, respectively. Especially, under the
IM = iαβ = iα2 + iβ2 (7)
leg open-circuit fault, the shapes of the phase currents, their
normalised ones, and absolute normalised line-to-line currents
To avoid the effect of the load conditions, the normalised phase match with those ones derived in (9) and (10), where the
currents, inN (n = a, b, c), divided by its magnitude in (7) are normalised current of the faulty phase is around zero and the other
utilised, which are expressed as currents are close to ±( 3/2) as shown in the last interval of
Fig. 3b. So, the absolute normalised line-to-line current between
iaN = sin(ωet + ϕ) the two sound legs is always about 3, whereas those of between
in in
inN = = = ibN = sin(ωet − 2π/3 + ϕ) . (8) the faulty and sound legs are about 3/2 as shown in Fig. 3c. The
IM iαβ
icN = sin(ωet + 2π/3 + ϕ) typical waveforms of the currents in Fig. 3, associated with the
aforementioned numerical values under the leg open-circuit faults,
In the normal operation, the three-phase normalised currents are are utilised to identify the faulty legs.
balanced, whose magnitudes are all unity. By normalised currents. It is noted that the average values of the absolute normalised
line-to-line currents for a fundamental cycle under the normal
However, under IGBT open-circuit fault conditions, iαβ is not
operation or for a half cycle while the current is conducting under
constant, which varies depending on the location of faulty IGBTs. an IGBT fault is obtained as
Let consider a case of an open-circuit fault in leg n, where the
current flowing through phase n is almost zero. So, the currents in
∫
T /2
the other two phases are the same with an opposite direction each 2 π
∥ f abs(imN − ilN ) ∥ = 3cos ωet + ϕ − dt
other. From (1) and (7), the magnitude of the current vector can be T 0
3
(12)
obtained as 2 3
= ≃ 1.1
2 2
π
iαβ = i
3 m
= i
3 l
(9)
So, the average values of the absolute normalised line-to-line
currents under an IGBT open-circuit fault in leg n are calculated as

IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931 925

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3 + 1.1 2.2.3 Multiple IGBT open-circuit fault identification: Multiple
∥ f abs(imN − ilN ) ∥ ≃ = 1.416 IGBT open-circuit fault location is performed by identifying the
2
combination of the leg fault and single IGBT fault as described in
0.5 3 + 1.1 the two previous subsections, which is similar to the fault
∥ f abs(inN − ilN ) ∥ = ∥ f abs(inN − imN ) ∥ ≃ = 0.983.
2 identification method presented in [27]. However, the method in
(13) [27] requires the current references additionally for determining the
faults of single IGBT, while the proposed method utilises just the
Table 1 lists the typical identification variables for the normal measured currents. Fig. 4 shows the flowchart which implements
condition and the leg open-circuit fault compared with the other the fault identification algorithm, where the leg open-circuit faults
cases of IGBT open-circuit faults. It is noted that the identification are monitored by observing the identification variables Pn. If the
variable for one leg is determined from the currents of the other leg is detected as an open-circuit fault, the polarity of that phase
two phases. It is seen from Table 1 that the identification variable current is not required for checking the single IGBT fault.
for the open-circuit leg is 0.632, which is highest compared with Inversely, the identification variables in (15) based on the polarity
the identification variables corresponding to the other legs and zero are calculated to monitor the fault status of the single IGBTs. By
for the normal condition and 0.316 in the case of an IGBT open- combining these two identification variables, it can identify the
circuit fault. If a certain identification variable exceeds the total 27 distinct signals for all possible faulty combinations of
threshold value th2, the corresponding phase is identified as a leg open-circuit IGBTs, which are indicated in Table 2 [27]. The
open-circuit fault. variable indicating the fault signals under the leg open-circuit
conditions is defined as
2.2.2 Localisation of single IGBT open-circuit faults: From the
analysis in the previous section, Table 1 lists the typical values of F Pn ≥ th2
identification variables, Pn, for the cases of leg open-circuit fault PF_n = (16)
S Pn < th2
and single IGBT faults. However, under the open-circuit faults of
multiple IGBTs at different legs, the faulty IGBTs cannot be
where the leg-fault signal, PF_n, based on the identification
distinguished by the Pn. So, other variables are defined to localise
variable, Pn, is used to determine the open-circuit faults of leg n.
the single IGBT faults even though they occur simultaneously in
different legs. It is known that the phase current flowing the leg The value of the PF_n is ‘F’ or ‘S’, where ‘F’ indicates the leg fault
containing the faulty open-circuit IGBT is only either positive or occurrence, whereas ‘S’ is for the sound leg. Similarly, the variable
negative, which is dependent on the faulty IGBT location. So, the indicating the fault signals of the open-circuit condition for the
faulty phase current is either positive or negative value for the first- single IGBT is defined as
half cycle and almost zero for the other-half cycle as shown in
Fig. 3a. Therefore, the faulty-IGBT localisation can be determined H Ln ≤ − th4
from the information of the phase current polarity, where a polarity LF_n = 0 Ln < th4 (17)
detection variable, ln, is defined as
L Ln ≥ th4
1 in ≥ th3
where the single IGBT fault signal, LF_n, is used to identify the
ln = −1 in ≤ − th3 . (14) open-circuit condition of the upper or lower IGBT in leg n, which
0 in < th3 is determined by comparing the Ln with the threshold value ±th4. If
the Ln is less than −th4, the upper IGBT in leg n is regarded as
where th3 is the threshold value utilised to observe the current faulty, where the LF_n has the value of ‘H’. If the LF_n has a value
polarity, which is set close to zero to avoid the influence of noise of ‘L’, the lower IGBT is faulty, where the Ln is higher than the th 4.
components in the measured currents in the real system. The In other cases, no single IGBT fault occurs.
effective identification variable, Ln, corresponding to the faulty As listed in Table 2, the open-circuit faults of total six IGBTs in
IGBT in the leg n, is obtained by averaging the value in (14) as the inverter can be classified into five groups, such as single IGBT
faults, phase open-circuit fault (two IGBTs in the same leg), double
Ln = ∥ ln ∥ . (15) IGBT faults (upper and lower IGBTs in the different legs), triple
IGBT faults (two upper (or lower) IGBT faults and a lower (or
In this work, the average values are obtained by the moving- upper) IGBT fault in the different leg), and quadruple IGBT faults
window technique, where each window is composed of four (a phase open-fault and double IGBT faults). It is noted that for the
samples. cases of triple IGBT and quadruple IGBT faults the shapes of the
The Ln is always near zero under normal operations due to a phase currents are similar under two different fault conditions, one
symmetrical phase current. However, the value of Ln is close to of three or four IGBTs cannot be determined as faulty or sound.
about +0.5 when the lower IGBT of the corresponding leg is an For example, when the S1 and S3 are an open circuit, the S6 is
open circuit, where the faulty phase only conducts the current in either faulty or sound, the shapes of three-phase currents are the
the positive side. Similarly, the Ln is about −0.5 in the case of the same for these two conditions. In this case, the condition of the S6
upper IGBT fault, where the faulty phase current just flows in the should be checked offline. It is similar in the other cases as
negative side. Then, the Ln is compared with the threshold values, indicated at the bottom of Table 2.
±th4, which is around 0.5, determined empirically. As aforementioned, since the current references are needed for
the fault detection, the proposed fault detection scheme can be
applied to the current-regulated motor drives such as vector control

Table 1 Identification variables for normal and fault conditions

Conditions Variables Notes
iαβ inN imN ilN Pn Pm Pl
normal operation IM inN imN ilN 0 0 0
leg-n open-circuit fault 2 2 0 3 3 0.632 −0.234 −0.234
i
3 l
= i
3 m
± 2
± 2
single IGBT fault in leg n 2 2 0 3 3 0.316 −0.117 −0.117 first half cycle
i
3 l
= i
3 m
± 2
± 2
IM inN imN ilN second half cycle

926 IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
and some model predictive control (MPC) methods with a direct current reference. Thus, the proposed identification method can be
current control [33]. However, it is not applicable to the direct applicable to any machine drive system fed by PWM inverters.
torque control, V/f control, and other types of MPC schemes This is the advantage of the proposed scheme over the method in
without a direct current regulation, since the current references, do [27] which requires the current references additionally to identify
not exist [34]. On the other hand, for the localisation of the faulty the faulty single IGBT.
IGBTs, only the motor phase currents are needed regardless of the
3 Selection of threshold values
The proposed method requires four threshold parameters of th1,
th2, th3, and th4 for fault detection and identification. A selection of
the appropriate threshold values is especially important to achieve
a correct and effective diagnosis. Similar to the existing diagnosis
methods, each threshold value can be determined by analysing the
behaviour of corresponding diagnosis variable for the normal and
fault operations, and a trade-off between fast diagnosis and
robustness to the false alarms is normally taken into account. All
the threshold values in this work are listed in the section of
simulation results. The selection of the threshold values is
described as follows.
The threshold value, th1, is compared with the detection
variable, KN, for the fault detection, where it is analysed that the
KN is always kept around the value of unity under normal
operations and increased only at the switch open-fault condition.
Therefore, the th1 should be selected higher than unity. If the th1 is
selected near unity, the false detection alarm may occur under the
normal operations since the KN can be fluctuated in the transient
states of the variations in load level or motor speed. On the
contrary, if the th1 is much higher than 1, the false alarm cannot
occur in transient conditions. However, in this case, the fault may
not be detected or the detection time is excessively increased
compared with the case of appropriate threshold value. So, the
trade-off is needed in order to satisfy both detection reliability and
speed.
Fig. 4 Flowchart of the proposed faulty IGBT identification scheme

Table 2 Combined possibility of multiple IGBT faults with identification variables [27]
Faulty switches LF_a LF_b LF_c PF_a PF_b PF_c
S1 H 0 0 S S S
S2 L 0 0 S S S
S3 0 H 0 S S S
S4 0 L 0 S S S
S5 0 0 H S S S
S6 0 0 L S S S
S1, S2 — 0 0 F S S
S3, S4 0 — 0 S F S
S5, S6 0 0 — S S F
S1, S4 H L 0 S S S
S2, S3 L H 0 S S S
S1, S6 H 0 L S S S
S2, S5 L 0 H S S S
S3, S6 0 H L S S S
S4, S5 0 L H S S S
S1, S3, [S6] H H L S S S
S2, S4, [S5] L L H S S S
S3, S5, [S2] L H H S S S
S4, S6, [S1] H L L S S S
S1, S5, [S4] H L H S S S
S2, S6, [S3] L H L S S S
S1, S2, (S3|S6) — H L F S S
S1, S2, (S4|S5) — L H F S S
S3, S4, (S1|S6) H — L S F S
S3, S4, (S2|S5) L — H S F S
S5, S6, (S1|S4) H L — S S F
S5, S6, (S2|S3) L H — S S F
[ ] indicates that the IGBT is either faulty or sound.
(—) means that either both or at least one IGBT is faulty.

IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931 927

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Table 3 Parameters of PWM inverter and motor
PWM inverter Induction motor
power rating 3 kW speed rating 1500 rpm
DC-link voltage 310 V torque rating 19 Nm
DC-link capacitance 2350 µF no. of poles 4
switching frequency 5 kHz stator/rotor 0.53/0.93 Ω
resistances
sampling frequency 10 kHz stator/rotor 3/3 mH
inductances
current sampling 100 µs magnetising 76 mH
period inductance
speed sampling 1 ms moment of inertia 0.0071 kg·m2
period

Fig. 5 Motor operation in speed variations

(a) Motor speeds, (b) Phase currents, (c) Detection variable KN

The threshold value, th2, is used for identification variable Pn.

Aforementioned in Section 2, the Pn is about 0.632 under the leg
open-circuit fault conditions, whereas it is around zero at normal
conditions. If a single IGBT open-circuit fault occurs, the
corresponding variable of the Pn becomes about 0.316. Therefore,
the th2 should be selected in the range of 0.316–0.632.
The threshold value, th3, is utilised to distinguish the polarity of
the phase current. In practice, since the measured currents include
the switching ripple components and sensor offsets, the th3 should
be selected carefully in order to ensure the identification
performance of the faulty switches over a wide range of load
currents. By the empirical tests, the th3 can be selected as 2–5% of
the rated current.
The threshold value, th4, is used for identification variable, Ln,
Fig. 6 Motor operation for load variations
that maintains the value of zero under normal operations. In the
(a) Motor speed, (b) Phase currents, (c) Detection variable KN
case of upper switch open-circuit faults, the corresponding variable
of the Ln is close to about −0.5, whereas it is around +0.5 in the
case of lower switch faults. However, since the currents of the is obvious that the proposed scheme has a high immunity against
healthy phases are not completely symmetrical under the single false alarms in transient states.
IGBT faults, the polarity detection variables for the healthy phase Next, simulation tests under the different IGBT open-circuit
is not always zero, which may reach ±0.25. In addition, the faults are performed, where the condition of IGBT faults is
intervals of the first- and second-half cycles of the currents in the produced by removing the gating signals. It is noted that to avoid
faulty phases are not balanced, so the Ln for the faulty phase is an excessive increase of the values of the detection variable in
some cases of the faults, the KN is limited to 1.5 for ease of display.
normally lower than 0.5. The th4 should be selected in the range of
The performance of the fault detection in the case of leg open-
0.3–0.45 for ensuring the identification performance.
circuit faults is illustrated in Fig. 7, where the faults occur in the
switches S1 and S2 of phase A simultaneously at 1.682 s. In this
4 Simulation results case, the phase-A current becomes zero as shown in Fig. 7a. The
The simulation was carried out for the three-phase PWM inverter fault detection variable is shown in Fig. 7b, where the KN is higher
of AC machine drives, where the induction motor is coupled with a than the threshold th1 of 1.3 at 1.689 s and the fault is detected
mechanical load. The parameters of the inverter and induction within about 28.6% of a fundamental cycle after the fault instant. It
motor are listed in Table 3. In this work, the th1, th2, th3, and th4 is seen in Figs. 7c and d that the faulty switch S2 is firstly identified
have been empirically selected to 1.3, 0.45, 0.25, and 0.4, by the variable, La, at 1.697 s, which is about 57.7% of a
respectively. fundamental period. Through monitoring the identification variable
Figs. 5 and 6 show the responses of the induction motor drive continuously, then, the faulty status of the switches S1 and S2 of leg
for the variations of motor speed and load, from which the A are recognised by variable Pa at 1.703 s, which is about 83.6%
robustness of the proposed diagnosis scheme to false alarms can be
of a fundamental cycle.
evaluated in transient conditions. Fig. 5a shows the motor speed
Fig. 8 shows the simulation test results for triple IGBT faults,
and its reference at the load of 0.3 p.u., where the speed reference
where the multiple IGBT open-circuit faults occur on the switches
is changed from 1200 to −1200 rpm and back to 1200 rpm. Fig. 5b
S1, S3, and S6 at 1.687 s. Due to the fault occurrence on the two
shows the motor phase currents, which are balanced with a
frequency variation during the transient condition. Fig. 5c shows upper IGBTs of the different legs and a lower IGBT of the other
the detection variable, KN, which is unity in the steady state but leg, their corresponding phases just conduct their phase currents in
one direction, which are only either positive or negative as shown
fluctuated in the transient state. It is seen that the KN is always
in Fig. 8a. It is seen in Fig. 8b that the faults are detected by KN at
lower than the threshold value of th1 even at abrupt speed changes.
1.701 s (52.3% of a fundamental period). Fig. 8c illustrates that no
For the torque step change test, the load is changed from 20 to phase open-circuit fault takes place. Then, the single IGBT fault is
80% and back to 20%, while the motor is operating at 800 rpm as examined in Fig. 8d, where the faulty IGBT S6 is firstly identified
shown in Fig. 6a. Fig. 6b shows the three-phase motor currents,
by the variable Lc at 1.704 s (65.8% of a fundamental cycle) and
which are regulated well. As shown in Fig. 6c, the KN does not
the values of the La and Lb indicate the faulty status of the IGBTs
exceed the th1, resulting in no false alarms. From these two tests, it
S1 and S3 at 1.706 s (73.3% of a fundamental period) and 1.710 s

928 IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 9 Operation under motor speed change
(a) Motor speed, (b) Phase currents, (c) Detection variable KN

the effects of the faults cannot be observed quickly if the fault

occurs during the negative-half cycle. In this scenario, the fault can
be detected at least in the next half cycle. It should be noticed that
the fault detection time is dependent on both the fundamental
period of machine currents and the instants of fault occurrence.

Fig. 7 Diagnosis for the open-circuit fault of phase A 5 Experimental results

(a) Three-phase currents, (b) Detection variable KN, (c) Identification variables Pn, (d)
To verify the validity of the proposed fault diagnosis method, the
Identification variables Ln, (e) Flags for faulty switches
experimental tests were also carried out for the 3-kW induction
motor drive system at the laboratory. A diode rectifier was utilised
to produce a DC voltage source of about 310 V feeding the
inverter. The SEMIKRON IGBT modules of SKM75GB12T4 were
used to build the PWM inverter and the main control board of a 32-
bit DSP chip (TMS320F28335) was used. The switching frequency
is 5 kHz with a dead time of 2 μs. The system parameters and the
detection and identification of threshold values are the same as
those of the simulation conditions.
First of all, the robustness of the proposed diagnosis technique
against false alarms in transient states is investigated, which is
shown in Fig. 9 in the case of reversal of motor speed. Fig. 9a
shows the motor speed, which follows well its reference changed
from 1200 to −1200 rpm. Fig. 9b shows the three-phase currents,
where their magnitude and frequency are fluctuated. The detection
variable, KN, is also fluctuated during the transient condition.
However, it is still lower than the threshold value, th1, with a
maximum value of about 1.075 as shown in Fig. 9c. So, a false
alarm does not occur for a step change of motor speed.
Next, the fault detection and identification under different faulty
IGBT conditions are investigated, where the faults of the IGBT
open circuits in the PWM inverter are produced by disconnecting
the gating signals. Also, the KN is limited to 1.5 for ease of display.
A performance of fault detection and identification in the case
of a phase open-circuit fault in the leg A is shown in Fig. 10. The
three-phase currents, shown in Fig. 10a, are balanced before the
fault, but the phase-A current becomes zero after the fault
occurrence. It is seen in Fig. 10b that the fault is detected by the KN
in 6 ms after the fault, which is about 24% of a fundamental cycle
period of the phase current. Next, in Figs. 10c and d, the faulty
Fig. 8 Fault of multiple IGBT (S1, S3, S6) IGBT S2 is firstly confirmed by variable La in 14 ms after the fault,
(a) Currents, (b) Detection variables KN, (c) Identification variables Pn, (d) which is about 56% of a fundamental period. Then, the phase-A
Identification variables Ln, (e) Flags for faulty switches fault including S1 and S2 is identified as faulty by variable Pa in 18
ms after the fault, which is about 72% of a fundamental period.
(86.7% of a fundamental period), respectively. The fault flags for Fig. 11 shows the fault detection and identification performance
the triple IGBT faults are illustrated in Fig. 8e. for triple IGBT faults, where the open-circuit faults occur in the
It is noted that the fault occurrence instant impacts the detection switches, S1, S3, and S6. In this case, since the faults occur in the
and identification times. If the open-circuit fault of the upper IGBT two upper switches of the different legs and one lower switch of
occurs during the positive-half cycle, the corresponding phase the third leg, their corresponding phase currents have either
current is immediately forced to zero. Then, the fault can be positive- or negative-half cycle components only, respectively,
observed quickly by the variations in the current shapes. Inversely, depending on the location of faulty switches as shown in Fig. 11a.

IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931 929

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
the fundamental cycle and guarantees the remarkable immunity to
false alarms even under the harsh transient conditions such as
abrupt variations of motor speed and load torque. Meanwhile, the
identification of faulty IGBTs is based on the normalised line-to-
line currents and the polarity of the phase currents for the leg open-
circuit and single IGBT faults, respectively. It allows to find out the
single IGBT fault, phase-open faults, and multiple IGBT faults,
which produce 27 possible combinations of IGBT faults. It has
been confirmed that the proposed algorithm is independent of
system parameters and operating conditions. Furthermore, the
proposed method is relatively simple to implement and extra
hardware is not required. The experimental results, which show a
good match with simulation, have verified the effectiveness of the
proposed method.

7 Acknowledgments
This work was supported by the National Research Foundation of
Korea, funded by the South Korea government, under Grant
2014R1A2A1A11052748.

8 References
[1] Melo, V.F.M.B., Jacobina, C.B., Rocha, N.: ‘Fault tolerant performance of
dual-inverter-based six-phase drive system under single-, two-, and three-
Fig. 10 Test for phase-A open-circuit fault phase open-circuit fault operation’, IET Power Electron., 2018, 11, (1), pp.
(a) Currents, (b) Detection variables KN, (c) Identification variables Pn, (d) 212–220
[2] Gautam, S.P., Gupta, S., Kumar, L.: ‘Reliability improvement of transistor
Identification variables Ln clamped H-bridge-based cascaded multilevel inverter’, IET Power Electron.,
2017, 10, (7), pp. 770–781
[3] Mohsenzadeh, S., Zarghani, M., Kaboli, S.: ‘A series stacked IGBT switch
with robustness against short circuit fault for pulse power applications’, IEEE
Trans. Power Electron., 2018, 33, (5), pp. 3779–3790
[4] Lu, B., Sharma, S.K.: ‘A literature review of IGBT fault diagnostic and
protection methods for power inverters’, IEEE Trans. Ind. Appl., 2009, 45,
(5), pp. 1770–1777
[5] Gao, Z., Cecata, C., Ding, S.X.: ‘A survey of fault diagnosis and fault-tolerant
techniques—part I: fault diagnosis with model-based and signal-based
approaches’, IEEE Trans. Ind. Electron., 2015, 62, (6), pp. 3757–3767
[6] Campos-Delgado, D.U., Espinoza-Trejo, D.R.: ‘An observer-based diagnosis
scheme for single and simultaneous open-switch faults in induction motor
drives’, IEEE Trans. Ind. Electron., 2011, 58, (2), pp. 671–679
[7] Salehifa, M., Idris, N.R.N., Aziz, M.J.A.: ‘FPGA based robust open transistor
fault diagnosis and fault tolerant sliding mode control of five-phase PM motor
drives’, J. Power Electron., 2015, 15, (1), pp. 131–145
[8] Jung, S.M., Park, J.S., Kim, H.W., et al.: ‘An MRAS based diagnosis of open-
circuit fault in PWM voltage source inverters for PM synchronous motor
drive systems’, IEEE Trans. Power Electron., 2013, 28, (5), pp. 2514–2526
[9] An, Q.T., Sun, L., Sun, L.Z.: ‘Current residual vector-based open-switch fault
diagnosis of inverters in PMSM drive systems’, IEEE Trans. Power Electron.,
2015, 30, (5), pp. 2814–2827
[10] Gou, B., Ge, X., Wang, S., et al.: ‘An open-switch fault diagnosis method for
single-phase PWM rectifier using a model-based approach in high-speed
railway electrical traction drive system’, IEEE Trans. Power Electron., 2016,
31, (5), pp. 3816–3826
[11] Charfi, F., Sellami, F., Al-Haddad, K.: ‘Fault diagnosis in power system using
wavelet transforms and neural networks’. Proc. of IEEE Int. Symp. on
Industrial Electronics, Montreal, Canada, July 2006, pp. 1143–1148
[12] Zidani, F., Diallo, D., Benbouzid, M.E.H., et al.: ‘A fuzzy based approach for
the diagnosis of fault modes in a voltage-fed PWM inverter induction motor
drive’, IEEE Trans. Ind. Electron., 2008, 55, (2), pp. 586–593
Fig. 11 Test for multiple IGBT S1, S3, and S6 faults [13] Lin, M., Li, Y.-H., Qu, L., et al.: ‘Fault detection of a proposed three-level
inverter based on a weighted kernel principal component analysis’, J. Power
(a) Currents, (b) Detection variables KN, (c) Identification variables Pn, (d)
Electron., 2016, 16, (1), pp. 182–189
Identification variables Ln [14] Kim, D.E., Lee, D.C.: ‘Fault diagnosis of three-phase PWM inverters using
wavelet and SVM’, J. Power Electron., 2009, 9, (3), pp. 377–385
[15] Ribeiro, R.L.A., Jacobina, C.B., Silva, E.R.C., et al.: ‘Fault detection of open-
Fig. 11b shows the detection variable, which indicates that the switch damage in voltage-fed PWM motor drive systems’, IEEE Trans.
faults are detected by the KN in 18 ms after fault, which is 54% of a Power Electron., 2003, 18, (2), pp. 587–593
fundamental cycle. Next, from Fig. 11c, the faulty IGBT, S6, is [16] An, Q.-T., Sun, L.-Z., Zhao, K., et al.: ‘Switching function model based fast-
diagnostic method of open-switch faults in inverters without sensors’, IEEE
firstly confirmed by the variable Lc in 23 ms after fault (69% of a Trans. Power Electron., 2011, 26, (1), pp. 119–126
fundamental period), and then the IGBTs S1 and S3 are identified [17] Choi, C., Lee, W.: ‘Design and evaluation of voltage measurement-based
sectoral diagnosis method for inverter open switch faults of permanent
by the La and Lb in 27 ms (81% of a fundamental cycle) and 32 ms magnet synchronous motor drives’, IET Electr. Power Appl., 2012, 6, (8), pp.
after fault (96% of a fundamental cycle), respectively. 526–532
[18] Wang, Y., Li, Z., Xu, M., et al.: ‘A comparative study of two diagnostic
methods based on the switching voltage pattern for IGBT open-circuit faults
6 Conclusions in voltage-source inverters’, J. Power Electron., 2016, 16, (3), pp. 1087–1096
[19] Freire, N.M.A., Estima, J.O., Cardoso, A.J.M.: ‘A voltage-based approach
This paper has proposed a novel diagnosis technique for multiple without extra hardware for open-circuit fault diagnosis in closed-loop PWM
IGBT open-circuit faults in three-phase PWM inverters for the IM AC regenerative drives’, IEEE Trans. Ind. Electron., 2014, 61, (9), pp. 4960–
drives. The fault detection method is performed based on the dwell 4970
[20] Rothehagen, K., Fuchs, F.W.: ‘Performance of diagnosis methods for IGBT
times of the actual current vector, which provides a relatively fast open circuit faults in three phase voltage source inverters for AC variable
detection speed as the detection time of at the longest about 54% of

930 IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931

© The Institution of Engineering and Technology 2019
 17554543, 2019, 4, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-pel.2018.5188 by Algeria Hinari NPL, Wiley Online Library on [26/06/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
speed drives’. Proc. of European Conf. on Power Electronics and [28] Bae, C.-J., Lee, S.-M., Lee, D.-C.: ‘Diagnosis of multiple IGBT open-circuit
Applications, Dresden, Germany, September 2005, pp. 1–10 faults for three-phase PWM inverters’. Proc. of Int. Power Electronics and
[21] Trabelsi, M., Boussak, M., Gossa, M.: ‘Multiple IGBTs open circuit faults Motion Control Conf., Hefei, China, May 2016, pp. 2504–2509
diagnosis in voltage source inverter fed induction motor using modified slope [29] Jlassi, I., Estima, J.O., El Khil, S.K., et al.: ‘A robust observer-based method
method’. Proc. of Int. Conf. on Electrical Machines, Rome, Italy, September for IGBTs and current sensors fault diagnosis in voltage-source inverters of
2010, pp. 1–6 PMSM drives’, IEEE Trans. Ind. Appl., 2017, 53, (3), pp. 2894–2905
[22] Trabelsi, M., Boussak, M., Chaari, A.: ‘High performance single and multiple [30] Telford, D., Dunnigan, M.W., Williams, B.W.: ‘Adaptive high bandwidth
faults diagnosis in voltage source inverter fed induction motor drives’. Proc. current control for induction machines’, IEEE Trans. Power Electron., 2003,
of Int. Conf. on Electrical Machines, Marseille, France, September 2012, pp. 18, (2), pp. 527–538
1717–1723 [31] Brooks, L.A.A., Castro, J.L., Castro, E.L., et al.: ‘Induction motor control:
[23] Choi, J.H., Kim, S., Yoo, D.S., et al.: ‘A diagnostic method of simultaneous multivariable analysis and effective decentralized control of stator currents for
open-switch faults in inverter-fed linear induction motor drive for reliability high-performance applications’, IEEE Trans. Ind. Electron., 2015, 62, (11),
enhancement’, IEEE Trans. Ind. Electron., 2015, 62, (7), pp. 4065–4077 pp. 6818–6832
[24] Jlassi, I., Estima, J.O., Khojet El Khil, S., et al.: ‘Multiple open-circuit faults [32] Mutoh, N., Kanesaki, M., Nakashima, J., et al.: ‘A new method to control
diagnosis in back-to-back converters of PMSG drives for wind turbine common mode currents focusing on common mode current paths produced in
systems’, IEEE Trans. Power Electron., 2015, 30, (5), pp. 2689–2702 motor drive systems’. Proc. of 38th IAS Annual Meeting, Salt Lake City, UT,
[25] Freire, N.M.A., Estima, J.O., Cardoso, A.J.M.: ‘Open-circuit fault diagnosis USA, 2003, pp. 459–466
in PMSG drives for wind turbine applications’, IEEE Trans. Ind. Electron., [33] Afmed, A.A., Koh, B.K., Lee, Y.I.: ‘A comparison of finite control set and
2013, 60, (9), pp. 3957–3967 continuous control set model predictive control schemes for speed control of
[26] Estima, J.O., Cardoso, A.J.M.: ‘A new approach for real-time multiple open- induction motors’, IEEE Trans. Ind. Inf., 2018, 14, (4), pp. 1334–1346
circuit fault diagnosis in voltage source inverters’, IEEE Trans. Ind. Appl., [34] Zhang, Y., Yang, H.: ‘Model-predictive flux control of induction motor drives
2011, 47, (6), pp. 2487–2494 with switching instant optimization’, IEEE Trans. Energy Convers., 2015, 30,
[27] Estima, J.O., Cardoso, A.J.M.: ‘A new algorithm for real-time multiple open- (3), pp. 1113–1122
circuit fault diagnosis in voltage-fed PWM motor drives by the reference
current errors’, IEEE Trans. Ind. Electron., 2013, 60, (8), pp. 3496–3505

IET Power Electron., 2019, Vol. 12 Iss. 4, pp. 923-931 931

© The Institution of Engineering and Technology 2019