UDC 621.314 https://doi.org/10.20998/2074-272X.2023.1.

05

E. Parimalasundar, R. Senthil Kumar, V.S. Chandrika, K. Suresh

Fault diagnosis in a five-level multilevel inverter using an artificial neural network approach
Introduction. Cascaded H-bridge multilevel inverters (CHB-MLI) are becoming increasingly used in applications such as distribution
systems, electrical traction systems, high voltage direct conversion systems, and many others. Despite the fact that multilevel inverters
contain a large number of control switches, detecting a malfunction takes a significant amount of time. In the fault switch configurations
diode included for freewheeling operation during open-fault condition. During short circuit fault conditions are carried out by the fuse,
which can reveal the freewheeling current direction. The fault category can be identified independently and also failure of power switches
harmed by the functioning and reliability of CHB-MLI. This paper investigates the effects and performance of open and short switching
faults of multilevel inverters. Output voltage characteristics of 5 level MLI are frequently determined from distinctive switch faults with
modulation index value of 0.85 is used during simulation analysis. In the simulation experiment for the modulation index value of 0.85, one
second open and short circuit faults are created for the place of faulty switch. Fault is identified automatically by means of artificial neural
network (ANN) technique using sinusoidal pulse width modulation based on distorted total harmonic distortion (THD) and managed by its
own. The novelty of the proposed work consists of a fast Fourier transform (FFT) and ANN to identify faulty switch. Purpose. The proposed
architecture is to identify faulty switch during open and short failures, which has to be reduced THD and make the system in reliable
operation. Methods. The proposed topology is to be design and evaluate using MATLAB/Simulink platform. Results. Using the FFT and
ANN approaches, the normal and faulty conditions of the MLI are explored, and the faulty switch is detected based on voltage changing
patterns in the output. Practical value. The proposed topology has been very supportive for implementing non-conventional energy sources
based multilevel inverter, which is connected to large demand in grid. References 22, tables 2, figures 17.
Key words: artificial neural network, fast Fourier transform, multilevel inverter, sinusoidal pulse width modulation, total
harmonic distortion.

Вступ. Каскадні багаторівневі інвертори H-bridge все частіше використовуються в таких пристроях, як розподільні системи,
електричні тягові системи, системи прямого перетворення високої напруги та багато інших. Незважаючи на те, що
багаторівневі інвертори містять велику кількість перемикачів, що управляють, виявлення несправності займає значний час. У
конфігурації аварійного вимикача увімкнено діод для роботи в режимі вільного ходу в умовах обриву несправності. При
короткому замиканні аварійні стани виконуються запобіжником, який може визначити напрямок струму вільного ходу.
Категорія несправності може бути визначена самостійно, а також відмова силових вимикачів, що порушує функціонування та
надійність каскадних багаторівневих інверторів H-bridge. У цій статті досліджуються наслідки та характеристики обривів та
коротких замикань багаторівневих інверторів. Характеристики вихідної напруги 5-рівневого інвертору часто визначаються
характерними несправностями перемикача, при цьому при аналізі моделювання використовується значення індексу модуляції
0,85. В імітаційному експерименті значення індексу модуляції 0,85 в місці несправного перемикача створюються односекундні
обриви і коротке замикання. Несправність ідентифікується автоматично за допомогою методу штучної нейронної мережі з
використанням синусоїдальної широтно-імпульсної модуляції на основі спотвореного повного гармонійного спотворення та
керується самостійно. Новизна запропонованої роботи полягає у застосуванні швидкого перетворення Фур’є та штучної
нейронної мережі для ідентифікації несправного перемикача. Мета. Пропонована архітектура призначена для виявлення
несправного комутатора при розмиканні та короткочасних відмовах, що має знизити повне гармонійне спотворення та
забезпечити надійну роботу системи. методи. Запропонована топологія має бути спроектована та оцінена з використанням
платформи MATLAB/Simulink. Результати. Використовуючи підходи швидкого перетворення Фур’є та штучної нейронної
мережі, досліджуються нормальні та несправні стани багаторівневих інверторів, і несправний перемикач виявляється на основі
моделей зміни напруги на виході. Практична цінність. Запропонована топологія дуже сприятлива для реалізації нетрадиційних
джерел енергії на основі багаторівневого інвертора, пов'язаного з великим попитом у мережі. Бібл. 22, табл. 2, рис. 17.
Ключові слова: штучна нейронна мережа, швидке перетворення Фур’є, багаторівневий інвертор, широтно-імпульсна
модуляція, повні гармонічні спотворення.

1. Introduction. Multilevel inverters (MLIs) have In [1] had investigated in both moderate and high-
aroused huge attention in the examination of established power applications of multilevel converters which play a
manufacturing electric drive organizations in recent days, significant role. MLIs come in three typical
with the intention of reaching their power quality as well configurations: diode clamped, flying capacitor, and
as demands. The key benefits of MLIs are the elimination cascaded H-bridged. The modular design of cascaded H-
of harmonic distortion in the output voltage waveform by bridged multilevel inverter (CHB-MLI) characteristics
increasing level capacity, and even the portability of and performance are used to achieve medium voltage and
battery packs or fuel including in intervals. Despite the high-performance characteristics. Short and open circuit
fact that MLIs are an established technology that may be faults are two types of failures which can occur in power
used in engineering applications, the failure of power switching devices in CHB-MLIs. Short circuit (SC)
electronic switches and fault analysis is a new research problems mostly damage, so protection from SC is
issue for researchers. It's used in engineering to check the
required. Artificial neural network (ANN) approaches for
condition of power switches in inverters. The number of
SC protection by using high potency fuses and de-
levels in the inverter varies, as does the quantity of
additional switching devices, increasing the risk of any saturation method.
one of the switches collapsing; hence, any such problem In [2] had analyzed open-circuit faults in power
should be addressed at the outset so that the drive and switches the device shutting down, and they can go
motor processes are not accepted during abnormal undetected for a long time. This could cause secondary
situations. To improve system reliability, an effective defects in the inverter or other drive components,
problem diagnosis system must be implemented. © E. Parimalasundar, R. Senthil Kumar, V.S. Chandrika, K. Suresh

Electrical Engineering & Electromechanics, 2023, no. 1 31

culminating in the entire system being shut down and techniques. The fast Fourier transform (FFT) technique
expensive repairs. and ANN approach are used to evaluate the spectrum
In [3] had investigated short-circuit faults in power properties of output voltage wave forms produced using
electronic switches, on the other hand, are extremely both modeling and experimental investigations at various
damaging and necessitate special precautions to fault situations. By using FFT technique, extract salient
automatically shut down the entire drive. These types of features such as total harmonic distortion (THD) and
failures must be identified and repaired in a microsecond in harmonic contents of output voltage signal at different
order to safeguard analogous semiconductor devices from fault cases. The performance characteristics of the FFT-
damage in the converter leg. On the other perspective, ANN model-based fault detection approach for MLIs can
extended open circuit fault behaviour of the power be compared to develop an effective fault diagnostic
converters might cause the entire system. Expertise in fault system. These concerns include the identification of
behaviours, fault prediction, and fault diagnostics will be switch faults and the monitoring of tolerances because
necessary to keep the MLI system functioning smoothly. parameters contribute to the reliability of the power
The two aspects of power electronic device fault diagnosis converter systems. The validation of proposed model is
are as follows: fault information acquisition, which entails implemented with the help of open and SC fault voltages
gathering data whenever a failure occurs using a specific and total harmonic distortion.
fault detection approach; and fault identification and 2. Literature Review. A fault-tolerant method for a
characterization, which rely upon that specific of failure CHB inverter was proposed in [11]. Additional versions
modes to recognize the category. The position of faults is provide the converter’s trustworthy and efficient
identified by an algorithm which was developed in [4]. In operation in the event of a failure. The recommended
[5] the open circuit fault can be caused by a number of method utilises an additional cross-coupled cascaded H-
factors, including a damaged inner wire, a transient short- bridged unit in addition to existing CHB components to
circuits, or a gate driver failure. Over-voltage, over-current, preserve output voltage and ensure continuation of
safety component failure, and improper gating signal are function in the event of an open/short-circuit fault.
some of the causes of SC failure. In [6] authors explained In [12] was developed an innovative technique of
the recurrent neural network-based voltage stability for grid fault diagnostics based on the minimum squares support
connected solar photovoltaic systems using static vector machine using back propagation algorithm.
synchronous compensator with recurrent neural network. Authors [13] established a digital circuit-based approach
Authors in [7, 8] investigated a number of recent articles on for identifying SC problems. The suggested approach
problems such as the creation of inverted pulse width detects two types of SC faults: hard switch fault and fault
modulation (PWM) method in CHB-MLI systems. In [9] under load; it can be utilised through any switch,
has analyzed fault analysis in inverter and also faults an independent of its characteristics; and it does not employ
inverter device is used continuously under abnormal artificial intelligence strategies and procedures when
settings, further issues will arise, resulting in severe inverter function is in progress. Rough sets theory (RST)
consequences. Furthermore, the MLI is composed of is used to create the digital diagnostic circuit, which
several switching devices and the entire system is complex optimizes and specifies a minimal set of variables
in structure, and there are numerous nonlinear impacts. As required to identify problems. When the variables are
a result, MLIs need some novel diagnostic strategies which subjected to RST, a sequence of diagnostic rules is
could not deal with nonlinear detection issues but also generated. These criteria are implemented using simple
diagnose and locate faults easily. The device voltage and logic operations, resulting in a digital diagnostics system.
current of a multilayer inverter might vary based on the part In [14] had developed a fault diagnostic method for
and location of the faults. Some research concentrates on photovoltaic (PV) inverters that allows for various open-
the device output current or voltage to assess fault form and circuit fault analyses. In [15] recommended that a fuzzy-
position more quickly and easily, and then used the sample based fault detection technique be used to analyse a
to expand a number of fault diagnosis techniques. Owing to voltage source inverter supplied three-phase permanent
the dangerous effects of SC faults on converter circuits, this magnet synchronous motor driving. The average current
type of fault must be detected as soon as possible. It is Park’s vector approach, which uses phase current
necessary to remember that certain circuit drivers are information, is used to calculate the fault symptom
already in a position to detect defective switches. Hence variables. A fuzzy logic approach is used to process fault
considering the value of MV drives on the industry, robust symptom variables and recover faulty information from
detection mechanisms need to be discussed. In [10] had power switches. The suggested fault detection technique
investigated electrical drives and devices require complex can identify and find not only two or more distinct open-
electrical converters to conform to high power circuit problems in switching devices, and moreover
periodic failures in power switches, which may enhance
requirements. MLI methods have also been tested as an
the motor drive system’s dependability. In a three-phase
approach to high and low voltage systems. Compared to
quasi-Z-source inverter, authors [16] proposed a method
traditional two-level inverters, MLIs produce major output
for detecting open-circuit failures. The proposed method
voltage and low harmonic output current distortion.
is confined to Z-source inverters and is based on assessing
The fundamental objectives of the proposed research
the impact of shoot-through durations on state variables
effort is to develop a high-performance fault detection
during switching periods. Defect site identification and
methodology for evaluating open and SC faults in MLI open-circuit monitoring are the two steps of the proposed
using enhanced signal processing and soft computing approach. After both steps of the open circuit fault

32 Electrical Engineering & Electromechanics, 2023, no. 1

detection method have been completed, a redundant leg is may be used on any level CHB inverter without any
activated and utilised in lieu of the failed limb. The further control modifications.
recommended technique is validated by test results from a To resolve the insulated gate bipolar transistor
low-voltage q-ZSI device. (IGBT) open-circuit failure problem of the propulsion
Authors [17] proposed a signal processing technique inverter in a transmission line power supply system,
for detecting switch open-circuit faults and identifying the subsequently, using the IGBT inverter open-circuit fault
individual problematic switch, which will help with identification, a simulation model which is based on
servicing and fault-tolerant functioning. The faulty phase propulsion inverter structure is constructed, and different
and defective switching combination, consisting of the switching fault signal waveforms are evaluated. Secondly,
defective component and the switching on its diagonally, the bus voltage magnitude data is poorly represented and
was identified using the stator phase currents statistical turned into a fault signal using direct detection
characteristics. methodology.
A fault diagnostic method comprised of multistate The fault-tolerant five-level inverter technique
data processing through subsection variation analysis described in [22] for open-end induction motor driving
lock, and ANN which is created using feature extraction applications uses a single DC connection. One end of the
for the output currents of balanced and unbalanced stages. drive is fed by a main inverter, while the other is fed by a
A novel generalized open-switch fault-diagnostic supplemental inverter. The proposed approach, in
approach was proposed in [18]. In this detection method, compared to other current inverter systems, allows for
defect prediction features are implemented as half-cycle five-level inverter operation with little interruption.
mean bridge voltages for an N-level CHB-MLI, which are Furthermore, the design is fault-tolerant in the case that
computed separately for positive and negative half cycles. the H-bridge switching devices and the extra two-level
Precedent half-cycle averages are utilised to anticipate inverter fail.
these averages during open-circuit fault circumstances, Since these MLIs feature a large number of power
which are then compared to observed values to determine semiconductors, the chance of failure is much higher. As
the open-circuit fault. This fast detection approach may a result, identifying potential faults and operating under
detect the faulty switch in cascaded inverters with many faulty situations are critical. The identification of a failure
voltage levels by looking at one structural characteristic might be complex in concept due to the large number of
of the output voltage. This technique may successfully components. MLIs have interesting advantages due to
identify the faulty switch of the cascaded inverter using their faulty structure, such as the ability to operate in
various level-shifted pulse width modulation methods, medium, high voltage, and high-power applications,
variable loading conditions, modulation indexes, and providing a better voltage waveform with low total
switching frequencies. harmonic distortion for electric machines applications,
Authors [19] presented a novel fault-tolerant control output filter elimination, dv/dt transient reduction during
technique for CHB-MLI to increase the greatest commutation, low electromagnetic interference emissions
achievable voltage under faulty conditions. Instead of from over voltages, and reduced power loss.
bypassing the damaged cell when a semiconductor breaks, 3. Proposed method of fault diagnosis system in MLI.
this method uses it to generate voltage. Due to the lack of Figure 1 illustrates literally the entire fault detection
one level of defective cell voltage level, the voltage is system set up for identifying defective power
decreased to half. In comparison to traditional fault- semiconductor switches in MLIs, which includes both
tolerant techniques, the maximum possible output voltage MLI and fault diagnosis system. The characteristics of
could be increased for the majority of problematic output voltage sequences were extracted using the FFT
circumstances. In addition, the defective cell switching approach. Frequency domain analyses of the terminal
mechanism is changed to enhance the output voltage voltage patterns are required to construct a significant
quality. A grid-connected single to three-phase multilevel application assessment method. The FFT approach was
converter with a fault-tolerant design was proposed in used to retrieve distinct attributes from the output voltage
[20]. The induction motor drive can operate even if any of signal. Despite the fact that a skilled feature extractor
the power semiconductor switches have open circuit should supply critical data facts more about ANN in the
faults. A control mechanism on a single active front end selected area, it was the highest degree of consistency
converter was used to accomplish this. A separate control reached within the adaptive intelligence network.
method is used to provide the power converter’s on-grid
fault tolerance capabilities. The control approach Basic Model of MLI
employed on a single grid-connected converter enhances
the input power factor, resulting in a unity power factor at DC Supply Multilevel Inverter IM
the source. The voltage control loop adjusts the DC-link
voltage to get the command voltage.
Fault Diagnosis System
The redundancy of the triangular carrier signals is a
criterion for expanding sinusoidal pulse width modulation FFT ANN offline Faulty
(SPWM) to numerous output voltage levels per phase-leg, Analysis/Feature Trained Pattern Switch
using MATLAB Identification
according to [21]. The recommended control technique Extraction

creates suitable modulation patterns for the CHB inverter

by modifying a sinusoidal modulating pattern to fit within Fig. 1. Proposed fault diagnosis system of MLI
a single triangle carrier signal range. These frameworks

Electrical Engineering & Electromechanics, 2023, no. 1 33

 Figure 2 depicts the signal separation of output where Vn is the root means square (RMS) value of the
voltage in terms of harmonics and RMS voltage output. voltage of the nth harmonic; n = 1 is the frequency of the
The FFT provides a frequency response representation for signal. It determines the degree of distortion in a voltage
any periodic or non-periodic signal. Figure 3 depicts one output.
of the 11th order sample harmonics, as well as THD and A. Fault classification using ANN techniques. In a
Vrms values, which are all utilized to extract characteristics multilayer feed forward network, Fig. 4 depicts the error
from the output signal employing FFT analysis. back propagation. The back propagation technique is used
to compute the necessary modifications once the
network’s weights are picked at random. The back
propagation algorithm may be separated into four stages
in general. The four steps of the back propagation
algorithm are feed-forward processing, back propagation
to the output layer, back propagation to the hidden layer,
Fig. 2. FFT harmonic analysis of output voltage signal and weight changes. The algorithm is disrupted whenever
the value of the error function has become reasonably
low, i.e., when the error between the actual and planned
output is less than a given number (convergence criteria).
In this network, the deviation is back-propagated, and the
weights and biases are essentially reconfigured using an
approach to reduce the mean square error (MSE), which is
the mean of all the errors for all sets of inputs and outputs,
a b and is determined as:
1 n
MSE   Pm  Qm  .
n m
(4)

c d

Fig. 4. Back propagation of error in multilayer feed forward

network
e f
Fig. 3. Representation of FFT investigation during: In this case, Pm and Qm are the desired and measured
a) 11th harmonic with open circuit fault; outputs for the mth input set, where n is the total number
b) 11th harmonic with short circuit fault; of input sets. The failure detection of MLIs was
c) THD analysis for open circuit fault; automated using an ANN in this investigation. The ANN
d) THD analysis for short circuit fault; was used to solve the difficulty of detecting the faulty
e) Vrms analysis for open circuit fault; switch in a cascaded MLI. Due to its simple methodology
f) Vrms analysis for short circuit fault and excellent predictive potential, the multilayer feed
The FFT provides frequency domain representation forward network with back propagation learning
of any periodic or non-periodic signal. In general, the technique has been regarded one of the numerous ANN
Fourier transform (FT) is a generalization of the Fourier designs accessible in the literature.
series. Instead of sines and cosines, as in a Fourier series, Using the 9 output neurons, the fault is categorized as
the FT uses exponentials and complex numbers. For a no fault, S1A to S4A fault, and S1B to S4B fault. The output
signal or function f(t), the FT is defined as: layer neurons are set up to perform multiple binary training
 patterns in response to different degradation scenarios.
Network topology, size, and learning rate, number of
F     f t  e
 jt
dt . (1)
training sets, convergence criterion, and number of

iterations are all essential parameters that influence the
The inverse FT is defined as: neural network’s convergence and learning time. The

1 learning rate is known to damp out oscillations to some
f t    F   e
 jt
d . (2) extent, during the training phase. Higher values of
2
 learning rate may result in fast convergence, but it may
The expression below is used to calculate THD for result in oscillation. The training time of the neural
an output voltage pattern as: network will rise as the number of training sets and
training cycle increases. For improved classification
V2 2  V32  V4 2  ....  Vn 2 results, an appropriate neural network structure must be
THD  , (3)
V1 found. As a result, in order to arrive at an ideal topology,

34 Electrical Engineering & Electromechanics, 2023, no. 1

the effectiveness of the neural network for varying at each point with a different DC voltage configuration.
parameters of the learning rate, training sets, convergence Figure 7 shows that there is a traditional cascaded H-
criterion, iterations, and number of neurons in the hidden bridged inverter topology with three H-bridged cells
layer must be examined in depth and assessed. attached to the three-phase inductive motor load at each
The majority of processing facilities in the hidden point. The degree of waveforms in the output signal
layer and the number of observations is two significant ranges may be calculated using 2N+1, where N is
parameters that influence the neural network’s typically the size of H-bridged cells in the network.
functionality. The neural network fails to meet the
convergence requirements when the number of hidden
layer neurons is fewer than 10. The neural network takes
longer to train and meet the convergence requirements
when the quantity of hidden layer neurons rises over 24.
The current network achieves convergence criterion
at 3800 iterations throughout the training phase. It shows
that 3800 iterations are enough for the optimal neural
network to be successfully trained. Then 150 more data Fig. 7. Structure of traditional three phase cascaded H-bridged
points were employed for identification testing with inverter topology comprising induction motor
modulation indices ranging from 0.8 to 0.95, which is as
shown in Fig. 5. However a three-phase MLI strategy is typically used
across drive systems, a single-phase MLI may be used at this
point, while a three-phase based fault detection system could
be expanded.
Figure 8 shows the schematic single-stage, five-stage
voltage output of the CHB-MLI used in existing positions
correlated with induction motor load. Induction motor
characteristics such as unbalanced stator currents and
voltages, torque oscillations, efficiency and torque
decreases, overheating, and excessive vibration are all
affected if the MLI power semiconductor switches fail.
Furthermore, some harmonic components of currents and
voltages can be amplified by these motor problems. To
Fig. 5. Identification of best training performance of MSE achieve good performance, the defective switch must be
identified and replaced as quickly as possible. Each IGBT
As the number of hidden layer neurons exceeds 18, switch is categorized as S1A, S1B, etc. by its cell position.
the neural network takes more time to train and adhere to The simulation experiments have been performed using the
the convergence requirements. The network’s mean MATLAB/Simulink tool and Table 1 summarizes the
square error was computed by keeping the step size at 0.1 parameters used for the simulation analysis.
with 18 hidden layer neurons in order to arrive at an
optimal value for the number of epochs. The mean square
error values derived from different amounts of hidden
layer neurons are shown in Fig. 6 depicts the proposed
fault diagnostic system’s detection rates for various
numbers of hidden layer neurons. As compared to other
instances, the device performs better with 18 secret layer
neurons. In this context, the overall detection performance
for all fault conditions is 100 %, and the device can
correctly find the fault in nearly any situation.

Fig. 8. Single-phase cascaded H-bridged 5-level MLI coupled to

the drive system
Table 1
Simulation parameters of five level MLI
Parameters Values
DC input voltage 115 V
Number of H-bridges 2
Fig. 6. Investigation of the ANN-MSE at various statistics of No. of switches 4
hidden layer neurons for 18
Output levels 5
B. Fault classification using ANN techniques. Modulation Index 0.85
Cascaded H-bridged inverter topology consists of a series Carrier frequency 3 kHz
of H-bridge inverter cells are interconnected to each cell Load Single phase IM, 0.5 HP, 50 Hz, 230 V

Electrical Engineering & Electromechanics, 2023, no. 1 35

 C. Fault classification using ANN techniques. The Mode 3: Five level cascaded H-Bridge inverter
SPWM switching technique is frequently utilised to create switches, all the bridge switches are zero, the output
the requisite IGBT transition pulses. The pulse of the voltage obtained across the load is zero.
triangular carrier is the same as the reference sinusoidal Mode 4: Five level cascaded H-Bridge inverter
signal when SPWM is resent. As illustrated in Fig. 9; the switches S3 & S4 – A is turned on in this mode of operation.
switching signals are processed using a sinusoidal and +VDC is the output voltage obtained across the load.
triangular signal with a modulation index of 0.85. The Mode 5: Five level cascaded H-Bridge inverter
technique of SPWM addresses sinusoidal waveform switches S3 & S4 – A and S3 & S4 – B are turned on in
creation by contrasting reference to carrier waves or this mode of operation. +2VDC is the output voltage
filtering the pulse output waveform by altering the widths obtained across the load. The same operation is shown in
of triangular waveforms. The basic pulse generation tabular form above in Table 2.
circuit for the sine pulse width modulation technique is 4. Results and discussion.
shown in Fig. 9. Low frequency reference sinusoidal A. Open circuit fault analysis and discussion. The
waveforms are compared with high frequency triangular primary open circuit failure occurs around 1 s with the
waves, often known as carrier waves. The switching bridged cell A switch in S1A to determine the voltage level
phase is changed when the sine and carrier waves cross is output before and after the start of the open circuit failure
shown in Fig. 10. of the MLI. Figure 11 reflects the typical output voltage,
amplified view and output current at both the failure of
such a single stage cascaded H-bridge inverter topology
related to the inductive motor. Related voltages and pulses
are sampled at 20 kHz, which is shown in Fig. 11.

Fig. 9. The sinusoidal reference signal and the triangular carrier

signal in PWM are used for modulation index of 0.85 carrier
frequency of 3 kHz values

Fig. 11. Voltage and load current pattern of cascaded H-bridged

inverter topology with inductive motor before and after S1A
switch open circuit fault initialization

Figure 10 displays the typical output waveforms

Fig. 10. SPWM based cell A switching pulses are generated by given by the normal and independent open-switch voltage
3 kHz carrier frequency and 0.85 modulation index fluctuations of cells A and B according to the basic
requirements of the 0.85 modulation index. In the case of
Table 2 illustrates the switching patterns of five level open switch fault waveforms, there is a fifty percent
cascaded MLI during healthy condition. SPWM based cell A clipping at the voltage magnitude level in the positive or
switching pulses are generated by 3 kHz carrier frequency negative half of the time as opposed to no fault condition.
and 0.85 modulation index is shown in Fig. 4. Figure 11 illustrates the output voltage and current
Table 2 waveforms when an open circuit fault occurs in cell A's
Switching table of five level cascaded MLI S2A switch. Visual examination of voltage and current
Switching sequences waveforms reveals that the output voltage amplitude
Bridge – A Bridge – B Voltage levels pattern varies significantly for each open circuit fault;
S1 S4 S2 S3 S1 S4 S2 S3 however, classifying the sort of fault solely on the load
1 1 0 0 1 1 0 0 +2VDC current waveform is difficult. The equivalent load current
1 1 0 0 0 0 0 0 +VDC waveform appears to follow a similar pattern following
0 0 0 0 0 0 0 0 0 the onset of an open circuit fault in switches S1A and
S2A, making it difficult to differentiate the problematic
0 0 1 1 0 0 0 0 –VDC
switch. Figure 12 shows the output voltage and load
0 0 1 1 0 0 1 1 –2VDC current waveforms with an induction motor load
following an S2A open circuit fault. Figure 13 depicts the
Mode 1: Five level cascaded H-Bridge inverter
typical output voltage waveforms of cells A and B at 0.85
switches S1 & S2 – A and S1 & S2 – B are turned on in modulation index values under normal, open-switch fault
this mode of operation. +2VDC is the output voltage circumstances, respectively. When comparing open
obtained across the load. switch fault voltage waveforms to no fault conditions,
Mode 2: Five level cascaded H-Bridge inverter there is a fifty percent clipping in the voltage magnitude
switches S1 & S2 – A is turned on in this mode of operation. level in the positive or negative half cycle. When
+VDC is the output voltage obtained across the load. comparing the output voltage waveforms under open-

36 Electrical Engineering & Electromechanics, 2023, no. 1

switch fault scenarios to the normal state, there is a conditions of cells A and B at the 0.85 modulation index
significant variation in all output voltage patterns. These value respectively. As a consequence, there is a small
waveform images clearly depict the fluctuations at each decrease in the output voltage and in the positive or
problem instance, which can help to create a fault negative half loop in the case of short switch fault
detection system that is more efficient. waveform generation. As a result of a detailed inspection
of the output voltage waveforms, it has been a significant
change in all voltage output patterns in both the open-
switch and short-switch voltage fluctuations as opposed to
normal conditions. These waveform diagrams explicitly
display the variations for each fault situation, which
would help encourage the development of a successful
fault detecting system.

Fig. 12. Voltage and current after S2A open circuit fault

Fig. 14. Output voltage and load current waveform of MLI

during short circuit condition

Fig. 15. Magnified view of fault analysis before and after the
initiation of S1A short circuit fault with induction motor load
m = 0.85. Fault initiated at 1 s

Fig. 13. Output voltage waveforms at normal condition and after

open switch fault of Cell A and B Fig. 16. Output current and voltage waveform of S1B SC fault
with induction motor load for the modulation index of 0.85
B. Short circuit fault analysis and discussion. The
initial short-circuit fault is triggered by one second (1 s) Figure 15 illustrates visual examination of
of the H-bridge A S1A switch to consider the pre and post waveforms reveals that the output voltage pattern varies
current and voltage series starting the short-circuit fault of somewhat for each SC fault; however, classifying the
the inverter topology. The output displays the normal nature of the fault solely on the load current is difficult.
output voltage and current waveform of the H-bridged The load current pattern of a SC fault in S1A and S1B, for
multilevel cascaded single-phase inverter. The short example, is identical, making it difficult to distinguish
circuit fault of S1A is generated at one second (1 s) and between the S1 switch faults in H-bridge A and H-bridge
Fig. 14 shows the embellished current and voltage B. Furthermore, the type of the load and its changes affect
waveform and also shows the start of the fault. Figure 15 the load current waveform. As a result, it’s possible that a
illustrates a magnified view of the output voltage and load switch issue will be misdiagnosed. The output voltage
current waveforms with an induction motor load waveform is a crucial parameter to create the fault
following an S1B short circuit failure. Figure 16 shows diagnosis system since it is irrespective of load and has
the output voltage and load current waveforms with an unique patterns for each switch problem. When a SC fault
induction motor load following an S1B short circuit fault. develops in cell B’s S1B switch, Fig. 16 depicts the
Figure 16 displays the standard output waveforms output voltage and current waveforms. Visual
obtained under the normal and distinct short switch fault examination of voltage and current waveforms indicates

Electrical Engineering & Electromechanics, 2023, no. 1 37

that the output voltage magnitude waveform for each SC regarding the malfunctioning switch of the MLI. For high
fault varies greatly, however diagnosing the fault type power applications, the number of levels of MLIs is rising
purely based on the load current waveform is challenging. day by day; the suggested system may be tested for 7 level
Following the commencement of a SC defect in switches and 9 level inverter system.
S1B and S2B, the equivalent load current waveform 5. Conclusions. This paper examines the
seems to follow a similar pattern, making it difficult to malfunctioning transition fault diagnosis of a single-phase
differentiate the faulty switch. H-bridged cascaded 5-level inverter topology attached to
Figure 17 depicts the typical output voltage the inductive motor. Specified output voltage waveform
waveforms of cells A and B at 0.85 modulation index properties are explored through simulation research in
values under normal, short-switch fault circumstances, various open-switch and short-switch fault conditions.
respectively. When comparing short switch fault voltage This approach focuses on the similarity of the output
waveforms to no fault conditions, there is a 50 % clipping voltage and indeed the modulated voltage signal, as well
in the voltage magnitude level in the positive or negative as on the effect of the modification of the multilevel
half cycle. When comparing the output voltage inverters on the discrepancy. Digital filters may be
waveforms under short-switch fault scenarios to the created and implemented in the future for real-time
normal state, there is a significant variation in all output applications while recording the output voltage signal to
voltage patterns. These waveform images clearly depict remove high-frequency noise. The suggested topology is
the fluctuations at each problem instance, which can help restricted to a five-level multilevel inverter, but it may be
to create a fault detection system that is more efficient. expanded to seven, nine, eleven, and more levels. As the
number of switches increases, so would the need for
identification to locate a faulty switch. Also, the transients
detected in the current signal during the open circuit or
short circuit faults of the multilevel inverter may be
studied and a diagnostic procedure created using this
information for 7-level and 9-level inverters.
Conflict of interest. The authors declare that they
have no conflicts of interest.

REFERENCES
1. Jahan H.K., Panahandeh F., Abapour M., Tohidi S.
Reconfigurable Multilevel Inverter With Fault-Tolerant Ability.
IEEE Transactions on Power Electronics, 2018, vol. 33, no. 9, pp.
7880-7893. doi: https://doi.org/10.1109/TPEL.2017.2773611.
2. Haji-Esmaeili M.M., Naseri M., Khoun-Jahan H., Abapour
M. Fault-Tolerant and Reliable Structure for a Cascaded Quasi-
Z-Source DC–DC Converter. IEEE Transactions on Power
Electronics, 2017, vol. 32, no. 8, pp. 6455-6467. doi:
https://doi.org/10.1109/TPEL.2016.2621411.
3. Jalhotra M., Gautam S.P., Kumar L., Gupta S., Chander
A.H. Fault Tolerance and Energy Sharing Analysis of a Single
Phase Multilevel Inverter Topology. IECON 2018 - 44th Annual
Conference of the IEEE Industrial Electronics Society, 2018, pp.
1209-1213. doi: https://doi.org/10.1109/IECON.2018.8591853.
4. Amini J., Moallem M. A Fault-Diagnosis and Fault-Tolerant
Control Scheme for Flying Capacitor Multilevel Inverters. IEEE
Transactions on Industrial Electronics, 2017, vol. 64, no. 3, pp.
1818-1826. doi: https://doi.org/10.1109/TIE.2016.2624722.
5. Babaei E., Laali S., Bayat Z. A Single-Phase Cascaded
Multilevel Inverter Based on a New Basic Unit With Reduced
Fig. 17. Voltage output patterns in normal state as well as short
Number of Power Switches. IEEE Transactions on Industrial
circuit fault states of cells A and B
Electronics, 2015, vol. 62, no. 2, pp. 922-929. doi:
The key properties of the load voltage output waveform https://doi.org/10.1109/TIE.2014.2336601.
are explored utilising simulation under various open-switch 6. Praveen Kumar T., Ganapathy S., Manikandan M.
Improvement of voltage stability for grid connected solar
and short-switch faulty conditions. Furthermore, major photovoltaic systems using static synchronous compensator with
elements such as THD analysis using the FFT technique, as recurrent neural network. Electrical Engineering &
well as an assessment of the MSE at various numbers of Electromechanics, 2022, no. 2, pp. 69-77. doi:
hidden layer neurons, are defined to be fed to that back https://doi.org/10.20998/2074-272X.2022.2.10.
propagation trained ANN. In a cascaded MLI, this suggested 7. Suresh K., Parimalasundar E. A Modified Multi Level
fault detection approach could identify each individual fault Inverter with Inverted SPWM Control. IEEE Canadian Journal
switch. When compared to the load current pattern, the of Electrical and Computer Engineering, 2022, vol. 45, no. 2,
pp. 99-104. doi: https://doi.org/10.1109/ICJECE.2022.3150367.
output voltage waveform at different fault circumstances 8. Suresh K., Parimalasundar E. A novel dual-leg DC-DC converter
exhibits distinct patterns, making it easier to discern the for wide range DC-AC conversion. Automatika, 2022, vol. 63, no. 3,
nature of the defect via visual observation. THD and pp. 572-579. doi: https://doi.org/10.1080/00051144.2022.2056809.
harmonic/fundamental ratios up to 11th order harmonics 9. Suresh K., Parimalasundar E. Design and Implementation of
retrieved from the FFT analysis give crucial information Universal Converter. IEEE Canadian Journal of Electrical and

38 Electrical Engineering & Electromechanics, 2023, no. 1

Computer Engineering, 2022, vol. 45, no. 3, pp. 272-278. doi: Inverter to Increase Maximum Output Voltage. 2019 IEEE
https://doi.org/10.1109/ICJECE.2022.3166240. Energy Conversion Congress and Exposition (ECCE), 2019, pp.
10. Ezhilvannan P., Krishnan S. An Efficient Asymmetric 2922-2927. doi: https://doi.org/10.1109/ECCE.2019.8912282.
Direct Current (DC) Source Configured Switched Capacitor 20. Parimalasundar E., Kumar N.M.G., Geetha P., Suresh K.
Multi-level Inverter. Journal Européen Des Systèmes Performance investigation of modular multilevel inverter topologies
Automatisés, 2020, vol. 53, no. 6, pp. 853-859. doi: for photovoltaic applications with minimal switches. Electrical
https://doi.org/10.18280/jesa.530611. Engineering & Electromechanics, 2022, no. 6, pp. 28-34. doi:
11. Mhiesan H., Wei Y., Siwakoti Y.P., Mantooth H.A. A Fault- https://doi.org/10.20998/2074-272X.2022.6.05.
Tolerant Hybrid Cascaded H-Bridge Multilevel Inverter. IEEE 21. Odeh C.I., Lewicki A., Morawiec M. A Single-Carrier-Based
Transactions on Power Electronics, 2020, vol. 35, no. 12, pp. Pulse-Width Modulation Template for Cascaded H-Bridge
12702-12715. doi: https://doi.org/10.1109/TPEL.2020.2996097. Multilevel Inverters. IEEE Access, 2021, vol. 9, pp. 42182-42191.
12. Cheng Y., Dong W., Gao F., Xin G. Open-circuit fault doi: https://doi.org/10.1109/ACCESS.2021.3065743.
diagnosis of traction inverter based on compressed sensing theory. 22. Ghosh Majumder M., Rakesh R., Gopakumar K., Umanand
Chinese Journal of Electrical Engineering, 2020, vol. 6, no. 1, pp. L., Al-Haddad K., Jarzyna W. A Fault-Tolerant Five-Level
52-60. doi: https://doi.org/10.23919/CJEE.2020.000004. Inverter Topology with Reduced Component Count for OEIM
13. De Mello Oliveira A.B., Moreno R.L., Ribeiro E.R. Short- Drives. IEEE Journal of Emerging and Selected Topics in
Circuit Fault Diagnosis Based on Rough Sets Theory for a Power Electronics, 2021, vol. 9, no. 1, pp. 961-969. doi:
Single-Phase Inverter. IEEE Transactions on Power Electronics, https://doi.org/10.1109/JESTPE.2020.2972056.
2019, vol. 34, no. 5, pp. 4747-4764. doi:
https://doi.org/10.1109/TPEL.2018.2861564. Received 13.05.2022
14. Huang Z., Wang Z., Zhang H. Multiple Open-Circuit Fault Accepted 20.08.2022
Diagnosis Based on Multistate Data Processing and Subsection Published 06.01.2023
Fluctuation Analysis for Photovoltaic Inverter. IEEE Transactions
on Instrumentation and Measurement, 2018, vol. 67, no. 3, pp. 516- Ezhilvannan Parimalasundar 1, Associate Professor,
526. doi: https://doi.org/10.1109/TIM.2017.2785078. Ramanathan Senthil Kumar 2, Assistant Professor,
15. Yan H., Xu Y., Cai F., Zhang H., Zhao W., Gerada C. Vanitha Selvaraj Chandrika 3, Professor,
PWM-VSI Fault Diagnosis for a PMSM Drive Based on the Krishnan Suresh 4, Associate Professor,
1
Fuzzy Logic Approach. IEEE Transactions on Power Department of Electrical & Electronics Engineering,
Electronics, 2019, vol. 34, no. 1, pp. 759-768. doi: Sree Vidyanikethan Engineering College,
https://doi.org/10.1109/TPEL.2018.2814615. Tirupati, AP – 517102, India,
16. Yaghoubi M., Moghani J.S., Noroozi N., Zolghadri M.R. IGBT e-mail: parimalasundar.e@vidyanikethan.edu (Corresponding
Open-Circuit Fault Diagnosis in a Quasi-Z-Source Inverter. IEEE Author);
2
Transactions on Industrial Electronics, 2019, vol. 66, no. 4, pp. Department of Electrical and Electronics Engineering,
2847-2856. doi: https://doi.org/10.1109/TIE.2018.2847709. SRM Institute of Science and Technology,
17. Yang S., Sun X., Ma M., Zhang X., Chang L. Fault Chennai, TN – 603203, India,
Detection and Identification Scheme for Dual-Inverter Fed e-mail: rskrren@gmail.com
3
OEWIM Drive. IEEE Transactions on Industrial Electronics, Department of Electrical and Electronics Engineering,
2020, vol. 67, no. 7, pp. 6112-6123. doi: KPR Institute of Engineering and Technology,
https://doi.org/10.1109/TIE.2019.2922924. Coimbatore, TN – 641407, India,
18. Anand A., Akhil Vinayak B., Raj N., Jagadanand G., George S. e-mail: mailchandrika@gmail.com
4
A Generalized Switch Fault Diagnosis for Cascaded H-Bridge Department of Electrical and Electronics Engineering,
Multilevel Inverters Using Mean Voltage Prediction. IEEE Christ (Deemed to be University), Bangalore, India,
Transactions on Industry Applications, 2020, vol. 56, no. 2, pp. e-mail: sureshk340@gmail.com
1563-1574. doi: https://doi.org/10.1109/TIA.2019.2959540.
19. Ouni S., Narimani M., Zargari N.R., Cheng Z. A New Fault-
Tolerant Control Method for Cascaded H-Bridge Multilevel

How to cite this article:

Parimalasundar E., Senthil Kumar R., Chandrika V.S., Suresh K. Fault diagnosis in a five-level multilevel inverter using an artificial
neural network approach. Electrical Engineering & Electromechanics, 2023, no. 1, pp. 31-39. doi: https://doi.org/10.20998/2074-
272X.2023.1.05

Electrical Engineering & Electromechanics, 2023, no. 1 39