Вы находитесь на странице: 1из 11

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

An Intelligent Based Fault Tolerant System for


Solar Fed Cascaded Multilevel Inverters
Albert Alexander Stonier and Brad Lehman, Senior Member, IEEE

unattended operation. Hence, it is difficult to monitor and


Abstract-- The paper presents an intelligent based fault maintain the overall system which has a wide distribution [5].
tolerant system for a solar Photovoltaic (PV) inverter. Artificial The various faults in solar PV array are surveyed in [6,7], but
Neural Network (ANN) based controller is used to monitor, the research does not consider the impact of other faults in
detect and diagnose the faults in solar PV panels, battery, solar PV energy conversion system, such as battery damage,
semiconductor switches and inverters. The Cascaded Multilevel inverter failure, or fault due to AC isolation.
Inverter (CMI) is connected across the combination of solar PV
panel and battery for DC-AC conversion. The major advantage
To overcome the faults in the entire solar PV system
of the proposed topology is that it can deliver power from source (PV array, battery and inverter), this research proposes an
to the load even under fault and partial shaded conditions. The efficient power electronic interface whose switching action is
paper also reviews on various faults in the solar PV energy stimulated by the intelligent controller. Unlike a conventional
conversion process and provides suitable solutions for each inverter, an improved Cascaded Multilevel Inverter (CMI) is
circumstance. Simulations are undertaken in MATLAB proposed to act as an interface whose operation is manifested
2016a/Simulink and experimental investigation is carried out for by the intelligent controller to shift the conduction of switches
a 3kWp solar PV system. The results have proven that the system based on fault and normal conditions. On considering the
is capable to deliver power in spite of faulty environments. The various advantages and applications of CMI in [8], it is found
comparison and discussion of the results were made to show the
effectiveness of the proposed system.
that CMI can significantly increase reliability and are able to
Index Terms-- Artificial Neural Network, Cascaded Multilevel operate even with some faulty components. CMI comprises of
Inverter, Fault diagnosis, Maximum Power Point Tracking, Solar many stages. Hence, we will demonstrate that the fault in any
Photovoltaic, Pulse Width Modulation one of the solar PV modules, batteries, or inverter stages can
be compensated by the other auxiliary switches whose action
I. INTRODUCTION is controlled by an intelligent system.
In the literature a great deal of work is reported under the
N order to meet the growing demand for electrical energy,
I renewable energy sources, especially solar and wind have
become widely accepted alternatives to conventional energy
areas of multilevel inverters with various control topologies,
but only fewer works deal with fault tolerant capability for the
entire solar PV system. There is a need to develop methods for
generation systems. It has been forecasted that the energy
PV systems to provide the continuous power to the load
produced by the non-conventional energy sources will satisfy
without any interruption during the unique types of faults PV
50% of the total power needs in 2050 [1].
systems experience.
A common problem with the solar Photovoltaic (PV) is the
A single phase five level inverter configuration is proposed
reduction of its delivered power, which can be caused by
in [9] that works as a three level inverter during fault and
certain abnormal conditions like faults, partial shading, MPPT
requires output end transformer to maintain the actual output
failure and inappropriate choice of converters. In addition, PV
for balancing the reduced power. It also requires separate truth
system for commercial and industrial applications must have
table for normal, severe fault and high fault conditions. Only
availability, reliability [2], power quality [3] and fault
the open switch fault is considered using T type three level
tolerance. Among these, fault tolerant capability plays a
inverter in [10] and used Schmitt trigger for diagnosis. In [11],
predominant role in providing high quality of power to the
inductors are replaced with open end transformers to limit the
load without any interruption. Fault tolerant capability is the
circulating current.
ability to continue service even under faulty conditions.
It has been concluded that multilevel inverters can operate
Fault is a serious problem which causes production loss
in a balanced condition at reduced power while the fault
[4]. Faults may subsequently lead to unexpected safety
occurs [12]. Although the multilevel inverters can
hazards, reduced system efficiency and deteriorate reliability.
continuously operate in a balanced condition, it will not be
A fault which remains even for a few seconds can cause
able to operate at its rated power. The fault diagnostic system
undesirable effects to the loads connected to the solar PV
proposed in [13] requires about six cycles to clear an open
conversion system. Normally the solar PV system runs under
circuit or short circuit fault. The desired output voltage is
This work was supported in part by the University Grants Commission achieved by varying the modulation index to maintain the
(Government of India) under Grant 5-179/2016 (IC). operation under balanced load condition. The fault detection
Albert Alexander Stonier is with the Electrical and Electronics given in [14] requires monitoring devices for the individual
Engineering Department, Kongu Engineering College, Perundurai, India panels, which makes the system more complex and expensive.
638060 (e-mail: ootyalex@gmail.com).
Brad Lehman is with the Electrical and Computer Engineering It is postulated that the fault diagnosis and fault localization
Department, Northeastern University, Boston, MA 02120 USA (e-mail: require data on array behavior under each fault conditions.
lehman@ece.neu.edu)

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

A fault switching device is detected without using Step 5: Develop an intelligent control algorithm to connect
additional sensors but by monitoring the distorted shape of the auxiliary inverter during the fault.
output current waveform in [15]. A seven level inverter which Step 6: Implement ANN and auxiliary inverter control in
produces a symmetrical six level waveform during unbalanced 3kWp solar PV system.
condition and the reconfiguration of gating signals according Step 7: Test the experimental PV system for 20 different types
to the failure modes on occurrence of failure in power devices of faults to verify the fault detection and mitigation.
is discussed in [16]. Utilization of bidirectional switches, Often the faults are associated with reduction/increase in
transformers with different turns ratio, PWM of high the voltage/current. This can result in either lower voltage or
switching frequency makes the fault tolerant system much asymmetry in any individual H bridge or CMI. The novelty of
more complex [17]. the paper resides in overcoming these issues using an
It is observed that there remain a few short-comings in the intelligent based controller.
research on performance of the fault tolerant capability for the
B. Design Layout
entire solar PV system, i.e. it cannot meet the need of real
world applications. Much of the works are reported on inverter Fig.1 shows the proposed circuit topology for fault tolerant
without PV array [9-15].The little work that includes PV only capability system. Although we explain our results for our
reports fault in switches [3]. None of the works have dealt specific simulation and experimental test bed the methods and
with fault detection on PV with respect to batteries or inverter steps are applicable to other size PV arrays with different
and also not addressed fault tolerant operation for the entire levels of CMI. The figure comprises of seven PV arrays (each
solar PV system. Thus, the investigation of the fault detection array is a series connection of 4 modules) coupled with seven
and tolerant systems is an important issue. inverters connected in series which depicts the 15 level
The major contribution of the proposed research work is configuration. An additional inverter is connected in series
summarized as below: with this configuration termed as ‘auxiliary inverter’ whose
• Analyzed the causes and effects of faults in solar PV, action (ON/OFF) is manifested by the control of two switches
battery and inverter SA1 and SA2. The auxiliary inverter is also powered by the
• Performed extensive simulation and extensive experimental solar PV array. Hence the circuit has the capability to generate
studies for various fault conditions in PV systems 17 level output voltage as it illustrates an 8 stage CMI. The
• Implemented an intelligent controller to ensure the power control action for the inverter switches is performed with
flow even during fault and partial shaded conditions PDPWM (Phase Disposition Pulse Width Modulation) by
making the corresponding analysis on other PWM approaches
• The salient results obtained thus proved that the proposed
in [18].
inverter topology performs well in various fault conditions
The switching pulses are given to all the eight inverter
The paper is organized as follows: Section II presents the
stages without any variations to avoid the high switching
problem statement and methodology and Section III exhibits
frequency patterns, which are found as a drawback in most of
the problem formulation with ANN. Section IV illustrates the the literature. The switching frequency is fixed at 1 kHz and
experiemntal results, discussions and final conclusions. the inverter output frequency is 50Hz which is determined by
the carrier and reference signals respectively.
II. PROBLEM STATEMENT AND METHODOLOGY The individual solar PV module has the specification of
A. General Problem Statement open circuit voltage, Voc=21.2V, short circuit current,
Isc=7.4A, maximum peak voltage, Vmp=16.5V, maximum peak
The major objective of the proposed work is divided into
current, Imp=6.95A and maximum peak power, Pmp=115Wp.
fault identification and fault mitigation in the entire solar PV
Four such modules are connected in series to form a solar PV
energy conversion system. The fault identification process
array. Hence the overall solar PV rating of the system is
comprises of the following steps:
3220Wp. The solar PV output is connected to the battery of
Step 1: Model and simulate 20 types of different faults (8 solar
rating 48V, 100Ah (4 batteries in series with individual rating
PV faults: ground fault, line to line fault, open circuit fault,
of 12V, 100Ah) which is depicted in Fig.1 as B1 to B8. With
floating PV systems, double ground fault, blind spot,
the State of Charge (SoC) of 50% and greater, the battery
undervoltage and hot spot heating, 9 inverter faults:
provides 48V. The reason behind the 48Vp input to the
overcurrent, overheating, switch not driven, switch open, both
inverter stage is to provide the 230Vrms across the load
diode and switch open, only diode open, short circuit, gate
terminals. In the proposed design, Vdc=336V which is greater
drive failure and blackout, and 3 battery faults: overvoltage,
than the square root of the grid voltage (325.26V), makes the
undervoltage and abnornal charging-discharging cycles).
system suitable for both standalone and grid connected PV
Step 2: Experimentally create the 20 types of faults in a 3kWp
configurations [19].
solar PV test bed
Hence, the faults developed in the battery such as
Step 3: Measure and record H Bridge output voltage for fault
overvoltage, under voltage, abnormal charging and discharge
to create N datasets. In this research we have collected 1500
cycles also been considered. The individual battery (12V,
data sets (685 simulation and 815 experimental data sets).
100Ah) discharges to 10.8V at 270C. When the SoC >50%, the
Step 4: Use 70% of data to train Artificial Neural Networks
battery attains the maximum voltage leading to overcharging
(ANN) using advanced Back Propagation (BPN) algorithm to
condition. Similarly the voltage reduces while the SoC is less
identify faults.
than 50%. The cut off voltage of the battery is found to be
In continuation, the fault mitigation process comprises of
36V and the fully charged voltage is 55.8714V.
the following steps:

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

Fig. 2. Fifteen level output voltage waveform

Fig. 3. Output voltage waveform during fault in inverter 1 (Vbattery drop= 48V)
In general, a fault may cause a change of output voltage by
∆Vo across the output terminals of a fifteen level inverter. The
reduction ∆Vo may become equal to either ∆VPVi or ∆Vbatteryi
or an inverter failure, where i represent the ith stage of the
multi-level inverter. The reduction in PV voltage is expressed
as,
∆VPVi = VPV reference - VPV peak measured (1)
Similarly, the reduction in battery voltage is expressed as,
∆Vbatteryi = Vbattery reference - Vbattery peak measured (2)
The change in the output voltage is given as,
∆Voi = Vo reference – Vo peak measured (3)
where ∆VPVi , ∆Vbatteryi and ∆Voi represent the change in PV,
battery and output voltages of a particular stage in Fig.1,
i=1,2,…7. The variations given as ∆VPVi , ∆Vbatteryi and ∆Voi
are used to monitor the deviations incurred in the source and
load voltages. The reference value is fixed at 48V (±2V
tolerance) at all the three quantities.
When one of the inverter stages is in OFF, short or open
circuit, then Vo peak becomes less than the desired/reference
Fig. 1. Power circuit of the proposed topology
voltage by ∆Vo. The peak value of the output voltage is
decreased when the inverter is under fault. This low input
C. Simulation analysis voltage supply is due to the fault in the solar PV or battery,
Simulations are carried out in MATLAB 2016a/Simulink. which is unable to provide the voltage required for the
The output voltage of a CMI while making all the eight stages individual inverter stages. The low peak value of CMI causes
of the inverter including auxiliary inverter to operate produces the adverse effects to the loads connected in the system which
17 levels. The normal operation of CMI with 15 levels is is to be avoided. The purpose of this section is to provide the
given in Fig.2. Here the auxiliary inverter is turned OFF by the effects of faults in the source or converter. It shows that when
switching condition SA1=1 and SA2=0. there is a fault in solar PV cell or module it exhibits a reduced
If any one of the inverter stages is completely turned OFF voltage required to charge the battery.
due to the occurrence of solar PV or battery fault, then the In addition, if there is a fault in the battery, it provides the
overall output voltage of the system will be the summation of reduced voltage required for the converter (CMI). Also, the
output voltages in the other inverter stages. Suppose inverter 1 semiconductor devices failure provides the reduced and
whose input voltage of 48V is under fault. Then, the overall asymmetrical voltage waveform. On considering the faults
output voltage is 48Vx7stages-48V=288V as shown in Fig.3. from the source to the load, the effect is reduced or achieves
The output voltage of CMI during the occurrence of fault in low voltage at the output end.
two inverter stages is 48Vx7stages-48V-48V= 240V. When Hence in the proposed work, the reference value of the
any one of the inverters provides the output voltage of 0V, due output voltage in an individual inverter stage is fixed. Any
the cascaded connection at the output end, the other inverter change in the output voltage is considered as the uncertainity
stages tends to operate in providing the reduced output voltage and imprecision (even the condition is not primarly due to
without any interruption. fault). With a small variation over the tolerance limit in the

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

output voltage may cause a unstable event. Hence the


mechanism is adopted in the work to compensate this loss and
also to assure the continuous and constant voltage output even
in the fault environment.
The possible detection and identification of fault at either
solar PV or battery or inverter is carried out based on the
analysis made on the reduction in the output voltage.
Common faults result in the reduction in the output voltage or
current or both. Some may also results in the phenomenal
increase in voltage or current due to overheating. Hence the
rate of change of these quantities is taken into consideration
for the fault identification and diagnosition. Fig. 4. Switching sequences for the eight stage inverter
It is noticed that during open circuit fault the waveform solar PV array, overcharging and discharging modes of a
obtained is asymmetrical and peak voltage reduction in battery, complete inverter stage shutdown, semiconductor
positive half cycle, whereas the normal peak voltage appears switches failure, solar cell damage, connection cable fault and
in the negative half cycle. When considering the open-circuit so on. During the abnormal conditions it is found that the
faults in more than a switch in an inverter, noise signals are inverter output voltage seems to be either 0V or at reduced
produced. The noise signals are produced due to the adverse voltage.
aymmetrical nature of the output voltage waveform. The open It is therefore required to set the switching patterns for the
circuit analysis is made by including and excluding the control switches SA1 and SA2. Setting the switching patterns
intervention of a diode incorporated in the switch which acts results in the operation of auxiliary inverter and makes the
as a current path while the switch is open. output voltage compensated in lieu of the loss incurred due to
The output voltage of CMI during short circuit fault is a the fault.Nearly 1500 datasets (by both simulation and
vice-versa situation as for open circuit condition in terms of experiment) are collected in which 1050 are used for training
reduction of output voltage in negative half cycle. It is clear and remaining 450 is used for testing. The testing data is very
that the multilevel inverter continues to operate even under the important to analyse whether the network is properly trained
fault condition but with reduced voltage. This reduced power and can provide the desired output. As the specification of the
creates a serious drawback to the loads connected in the system is same in both simulation and experiment, the dataset
system. obtained provides the similar values. Out of 1500 data set, 685
Usage of asymmetrical output signal deteriorates the data set are obtained through simulation and the remaining
performance of the entire system. The open circuit and short 815 data set by experimental setup.
circuit condition in a solar PV module or open circuit in The dataset refers to the value of current, voltage,
battery results in the much reduced voltage required for the frequency, temperature, time, switching angle obtained during
CMI. Majority of the faults associated with solar PV or battery both normal and abnormal conditions through simulation and
provides either reduced power or asymmetrical/distorted experiment in the entire solar PV conversion process. For an
waveform in the output end. example, 28 different cases refers to 28 dataset obtained in a
Fig.4 shows the switching signals required for the eight solar PV module. Such 28 datasets are obtained in various
stage CMI. For the switching signal generation, eight carrier conditions (temperature, irradiance, different loading
signals are used and the eighth carrier signal operation will be conditions such as inductive and 1.5HP motor, inducing faults
realized only during the fault and partial shading conditions. and so on). Hence the maximum number of dataset has been
The eight carrier signals above and below the zero reference obtained in both simulation and experiment.
will be referred to as fault mode operating signals. START
The initialization of this signal will be made by the intelligent
controller through auxiliary inverter. When the fault occurs,
Measure Voi
the control signal directed from the intelligent controller (i=1,2,..7)
through pulse generator makes the upper carrier signals to ANN
function which turn ON the auxiliary inverter. Fault
No Calculate the change in output
D. Algorithmic Design Is
V oi = Voi ref?
voltage
(∆Voi =Voi ref -Voi peak)
Fig.5 shows the flowchart of the proposed system. As the Normal Yes
output voltage of CMI is independent of load and figuratively Identify the type of fault in
solar PV conversion system
represents the fault type and location, it is taken as the input
dataset for ANN represented as Voi (where i=1,2,3…7). The Reconstruct the Reconstruct the switching
dataset is collected for both normal (no fault and no partial switching table to
[1,0]
table to [0,1] for the type of
fault identified
shading) and abnormal (faults listed in appendices at various
irradiance and temperature levels etc.).
The data set under abnormal condition is obtained in both STOP Control the auxiliary inverter
as per the fault duration
simulation and experimental setup under various
circumstances such as varying the irradiance from minimum
to maximum based on the geographical area, partial shading of Fig. 5. Flowchart of the proposed system

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

The red color dotted line marked in the flow chart in In addition, the proposed system considers the fault in the
Fig.5 specifies the action performed by ANN (explained in entire standalone solar PV conversion process (solar PV,
section III). Initially, the output voltages of the individual battery, inverter, accessories) and also maintains the output
inverter stages are measured. If the voltages are equal to the voltage due to partial shading conditions. The auxiliary
desired 48V each, then the controller performs the normal switches are active not only during the fault, but also during if
mode (marked in green color dotted line). If the output any one of the solar PV array (numbered as 1 to 7) is subjected
voltages are not equal to the desired reference, then the fault to partial shading condition.
mode operation will be performed. In this mode, the controller The architecture of ANN used for the system is given in
will calculate the change of output voltage (∆Vo). During this Fig.6 which depicts the input layer with seven neurons (Vo1 to
process, the controller initiates the auxiliary inverter to operate Vo7), output layer with two neurons ([0,1] or [1,0]) and two
on compensating the loss ‘x’ and restores the continuity of the hidden layers with 20 and 10 neurons in each layer
supply without any interruption. Meanwhile, the location and respectively. The choice of hidden neurons is based on the
type of the fault is identified based on the value of voltage complexity of the problem. The hidden neurons can also be
dip//rise, shape of the waveform and measurement of voltage, calculated based on the formula given in the equation (4) [5].
current at the input/output terminals of battery. Here K represents the number of neurons in hidden, I is input
The output of ANN is set as [1,0] for normal and [0,1] for neurons, O is output neurons and θ is the positive semi definite
abnormal conditions. The [1,0] and [0,1] logic is used to turn constant between 0 to 10.
ON/OFF the switches SA1 and SA2 which then operates K = I +O + θ (4)
auxiliary inverter. During the normal operation, [1,0] logic
will provide the 15 level output voltage waveform. Similarly The activation functions used are ‘tansig’, ‘purelin’ and
the logic [0,1] will provide the 15 level output voltage ‘purelin’ between input and output through two hidden layers.
waveform if any fault exists in the system on considering the Levenberg Marquart algorithm (LMA) is used for the weight
faulted inverter stage output is equal to 0V. Hence the updation and ‘traingdm’ for training. LMA is the combination
proposed topology will provide 15 levels at all time without of error back propagation (gradient descent) and Gauss
reduction in the output voltage inspite of fault and partial Newton algorithm and blends between these two methods
shading conditions. Before the dataset is used for training, it is during training for stability and speed.
normalized. If the normalization is not made, the large LMA is the fastest back propagation algorithm (BPA) and
variations in the output voltages would monopoly the learning designed to approach second order training speed without
phase and would be difficult to react for any small variations. computing the Hessian matrix. The benefits of LMA include
The input dataset is normalized with the bipolar functions for fast, stable convergence, and it does not require more memory
initial scaling than other algorithms [3]. The performance of ANN after
training is shown in Fig.14.
III. ARTIFICIAL NEURAL NETWORK If any fault occurs in the system, ANN can detect the
change of output voltage (∆Vo) and provides subsequent
Artificial Neural Network (ANN) symbolically mimics the switching patterns to SAI and SA2 for the operation of
human brain and is widely used in various decision making auxiliary inverter which in turn provides 15 level output
applications. The success of ANN in any application resides in voltage waveform. It is also considered the condition of short
the way of its learning and adaptation with reduced duration fault which is due to the sudden change in climatic
computational complexity. Partially trained systems may not condition or resuming from any fault operation to normal as
be able to achieve the desired output. The three basic steps shown in the Fig.7 and Fig.8.
involved in ANN are dataset collection, learning/training and Fig.7 shows that when there is a fault (any one inverter
execution. The dataset collection can be made by either output voltage is 0V) from 2ms to 6ms (2 cycles) there is a
fundamental formulas [3] or with simulation and experimental reduction in peak voltage until the fault resumes at 6ms. If the
studies. Once the dataset is collected, it has to be preprocessed fault does not resume, even after few cycles, the reduction of
for feature extraction and then proceeded towards learning. voltage will be continued.
The process of finding weights is called learning.
The learning phase is structurally similar to the tuning of
synapses in the natural neuron structure. The weights are to be
adjusted or tuned to achieve the desired output, which will be
supported by various feedforward and feedback algorithms.
Once the network is trained it should be tested before it moves
for the final execution phase.
In the earlier works on fault diagnosis [12,13], ANN is used
to calculate the switching pattern of CMI, which alters the
carrier signals of PWM resulting in distorted output voltage
waveform. It also makes the system more complex. Only one
fault (either open circuit or short circuit) is considered [12]. In
this proposed work, the control to two switches (SA1, SA2) is
undertaken whose switching action is decided by Artificial
Neural Networks.
Fig. 6. ANN architecture

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

stage inverter. The comparison of these carrier signals with


sinusoidal reference generate the desired pulses for the
inverter switches. The detailed architectural analysis of POD
based pulse generation using DSP processor is dealt in [19].
The control of two switches SA1, SA2 is made through the
controller equipped with ANN trained program using
DSPACE 1104 as given in Fig.9.
Initially, the dataset collected (both normal and abnormal)
is stored in the workstation. The voltage sensors are provided
in the individual output end of the H bridge. At every instant
the voltage values are given as the input to the controller. The
offline trained neural network receives these voltage values
Fig. 7. Fault diagnostic system without the intervention of ANN and compares it with the dataset if there is a rate of change of
voltage.
If it finds any deviation in the rate of change of voltage
then the decision will be arrived on account of either normal
or fault condition. If the condition is normal the modulating
signal arrived from the ANN will direct the pulse generation
by shifting the eighth carrier signal to turn OFF the auxiliary
inverter. If fault exists, the eighth carrier signal is compared
with the control signal from ANN to turn ON the auxiliary
inverter.
The output voltage of each H Bridges are sensed and given
as the input to ANN which in turn provides the switching
pattern for the switches SA1 and SA2. The overall
Fig. 8. Fault diagnostic system with the intervention of ANN
The intervention of ANN for this condition is shown in experimental setup consists of eight DC inputs for the inverter
stages and seven output nodes of output voltages. Due to the
Fig.8 in which the auxiliary inverter operates from 2ms to 6ms
control of SA1 and SA2 instead of entire CMI, the fault
without any interruption for the output voltage. In this
detection and diagnosis is faster and does not consume any
condition, ANN functions appropriately in altering the
additional cycles for fault diagnostics. If there is a time delay,
switching pattern from [1,0] to [0,1] at 2ms and again to [1,0]
at 6ms. the power delivered from the source to load will not cut-off,
In certain works reported earlier, the controllers take a few rather it produces a reduced voltage until the fault diagnostic
system operates which is not more than 2 cycles.
cycles to diagnose or tolerate the fault without mitigation [12].
Fig.10 shows the fault detection system to understand the
The reason is due to the fact that the controller significantly
location and type of the fault caused in the solar PV energy
varies the switching pattern required for the CMI, which
conversion process. The detection and diagnostic system are
consumes more time and drastic increase in switching
frequency. To avoid this delay and also to perform the control incorporated in a single hardware arrangement for providing
action at a faster rate, the CMI is coupled to the auxiliary multiple purposes.
inverter.
The switching action is intended for only two states
(ON/OFF). Based on the rate of reduction in Vo1 to Vo7, the
location of the fault can also be viewed based on the fixation
of reference value in battery (0 -12V) and inverter (0-48V) for
the individual stages.

IV. EXPERIMENTAL RESULTS


The experimental investigation is carried out in a 3kWp
solar PV system. The rating of the individual solar PV module
and the battery is the same as that of given in the section II. A Fig. 9. Experimental set up of eight stage inverter depicting output voltage
seven stage inverter is powered by a solar PV system through
the series of battery rated 48V, 100Ah. The auxiliary inverter
is additionally coupled at the output end of seventh stage
inverter. Hence, the eight stage inverter (including auxiliary)
is provided with switching pulses through a TMS320F240
Digital Signal Processing (DSP) controller.
The carrier signals are generated and shifted above and
below the zero reference pertaining to the Phase Opposition
Disposition (POD) based switching technique which requires
eight carriers above and below the zero reference for an eight Fig. 10. Fault detection system to detect its type and location

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

The fault detection process is implemented by the In both the cases (a) and (b), the CMI provides the reduced
necessary preconditions (in terms of voltage, current and output voltage until or otherwise the fault is cleared by any
power) determined for the individual set of faults in binary bit external intervention. Similar faults are also experimentally
pattern. Based on the reduction or increase in the output tested for the different cases of solar PV and battery which
voltages it can locate the faults in the subsequent inverters. For reduces the CMI output levels to even of three level with 48V.
an example, for detecting the inverter fault, the output voltage It is noted that the fault in the system provides the reduced
will be depecited as 1110111 which mean that the H bridge 4 output voltage or even 0V if and only if the fault is considered
is under fault. Similar criteria hold for other fault conditions. in all the sources which may not be practically feasible.
Fig.11 and Fig.12 depicts the output voltage waveforms The power quality analyzer WT3000 is also used to
obtained during short circuit and open circuit fault conditions. visualize the output voltages given above as it can monitor the
The open circuit and short circuit conditions are waveforms for short interval of times under fault conditions.
experimentally made by rearranging the connections made for The output voltage obtained by the intervention of ANN is
the individual inverter stages. The waveforms obtained are given in Fig.14 which shows the fifteen levels during normal
asymmetrical with the Vbattery drop of 48V at either positive or condition on inducing the various faults to the system.
negative half cycles respectively. Fig.13 shows output voltage The analysis and discussions on the results was made on
waveform obtained during the fault operation of individual comparison with the other fault detection methods given in the
inverter stage. literature. The comparison given in Table I shows the
The waveform in Fig.13 is realized by considering the advantages of proposed system such as increased levels for
faults such as: a) battery fault (deep discharge, electrolyte power quality improvement, simple structure, fast detection
failure, discontinuation of series connection, combination of for diagnostic process and considering the fault for the entire
normal and abnormal battery in a series node) and b) solar PV solar PV conversion process.
fault (discontinuation of PV module, complete shading in one Based on the Table I it is found that majority of the works
or more modules, cable failure from solar PV to battery, solar reported are for switch faults on considering either open
cell damage and abnormal series connection). circuit and/or short circuit faults. This is due to the estimation
that about 38% of all the failures are found in the power
converter [20,21] and the most of faults are occurred in the
power switches [22,23].
Hence the proposed work extended to find a solution of
faults which occur in the entire system using an intelligent
controller. The most important parameters in a fault-tolerant
system design such as component quality, level of
redundancy, redundancy type and implementation,
maintenance, etc. are analyzed and studied which is well
depicted in the results obtained.

Fig. 11. Output voltage waveform during short circuit fault

Fig. 14. Output voltage waveform waveform with the intervention of ANN
Fig. 12. Output voltage waveform during open circuit fault (normal)
TABLE I
COMPARISON AND ANALYSIS
Ref. Method adopted Levels Faults considered
[5] ANN (5-9-8 structure) - Switch faults
[9] Bidirectional switches 5 Switch faults
[10] T type inverter 3 Open switch fault only
[12] ANN (40-2-1 structure) 5 Open and short circuit
[13] ANN (4-4-3 structure) 5 Switch faults
[15] Level shifted PWM 7 Switch faults
[16] Carrier modulation 5 Switch faults
[24] Vector machine approach 5 Open switch fault only
[25] ANN 5 Switch fault only
Fig.13. Output voltage waveform during fault in one inverter stage Proposed work 15 & 17 20 different faults

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

V. CONCLUSION [16] Mingyao Ma, Lei Hu, Alian Chen, and Xiangning He, “Reconfiguration
of carrier-based modulation strategy for fault tolerant multilevel
The paper presented the fault tolerant system which inverters”, IEEE Transactions on Power Electronics, Vol. 22, No. 5,
performed the three tasks such as detection (to classify normal pp.2050-2060, Sep.2007.
or abnormal), identification (to identify the faults as listed in [17] Muhammet Biberoglu, Kayhan Ince and Filiz Karaosmanoglu, “Design
of fault-tolerant cascaded h-bridge multilevel inverter with output-side
Appendices) and remedial action (with the aid of intelligent transformers using bidirectional switches”, 4th International Conference
controller). A detailed study has been undertaken to analyze on Renewable Energy Research and Applications, Palermo, Italy, 2015,
the faults which occurs in the entire solar PV energy pp.98-103.
conversion process which was never dealt in any literature. It [18] Behrooz Mirafzal, “Survey of fault-tolerance techniques for three-phase
voltage source inverters”, IEEE Transactions on Industrial Electronics,
has been found that the improved features of fault tolerant Vol. 61, No. 10, pp.5192- 5202, Oct. 2014.
system will certainly provide the potential for better reliability, [19] Albert Alexander, “Development of solar photovoltaic inverter with
availability and permits the use of circuit elements with a wide reduced harmonic distortions suitable for indian subcontinent”, Journal
range of quality. of Renewable and sustainable Energy Reviews, Vol.56, pp.694-704,
Apr. 2016.
[20] Nasrudin A. Rahim, Krismadinata Chaniago and Jeyraj Selvaraj, “Single
VI. REFERENCES phase seven level grid connected inverter for photovoltaic system”,
[1] Carlo Cecati, Fabrizio Ciancetta and Pierluigi Siano, “A multilevel IEEE Transactions on Industrial Electronics, Vol. 58, No. 6, pp. 2435-
inverter for photovoltaic systems with fuzzy logic control”, IEEE 2443, June 2011.
Transactions on Industrial Electronics, Vol. 57, No. 12, pp.4115-4125, [21] Ye Zhao “Fault analysis fo solar photovoltaic arrays”, M.S. Thesis,
Dec. 2010. Dept. Electrical Eng., Northeastern University, Boston, USA, 2010.
[2] Albert Alexander and Manigandan Thathan, “Modelling and analysis of [22] O. V. Thorsen and M. Dalva, “A survey of the reliability with an
modular multilevel converter for solar photovoltaic applications to analysis of faults on the variable frequency drives in industry”, Proc.
improve power quality”, IET Renewable Power Generation, Vol. 9, Conf. Rec. 6th Eur. Power Electron. Appl. Conf., Spain, 1995, pp. 1033–
Issue. 1, pp. 78–88, Jan. 2015. 1038.
[3] Albert Alexander and Manigandan Thathan, “Power quality [23] H. Schwab, A. Klonne, S. Reck, and I. Ramesohl, “Reliability
improvement in solar photovoltaic system to reduce harmonic evaluation of a permanent magnet synchronous motor drive for an
distortions using intelligent techniques”, Journal of Renewable and automotive application,” Proc. Conf. Rec. 10th Eur. Power Electron.
Sustainable Energy, Vol.6, Issue. 4, pp. 043127 (1)-(19), June 2014. Appl. Conf., Toulouse, France, 2003.
[4] J. Rodríguez, S. Bernet, B. Wu, J. O. Pontt, and S. Kouro, “Multilevel [24] Pablo Lezana, Josep Pou, Thierry A. Meynard, Jose Rodriguez, Salvador
voltage-source-converter topologies for industrial medium-voltage Ceballos, and Frédéric Richardeau, “Survey on fault operation on
drives”, IEEE Transactions on Industrial Electronics, Vol. 54, No. 6, multilevel inverters”, IEEE Transactions on Industrial Electronics,
pp. 2930–2945, Dec. 2007. Vol. 57, No. 7, pp.2207-2218, July 2010.
[5] Yuchuan Wu, Qinli Lan and Yaqin Sun, “Application of BP neural [25] Albert Alexander.S, Manigandan.T & Saranya.E, “Fault Detection and
network fault diagnosis in solar photovoltaic system”, IEEE Diagnosition of Cascaded multilevel inverter using ANN”, Proceedings
International Conference on Mechatronics and Automation, Changchun, of Second International Conference on Science and Technology, VIT
China, 2009, pp.2581-2585. University, 21-22 April 2011
[6] Mohammed Khorshed Alam, Faisal Khan, Jay Johnson, and Jack
Flicker, “Comprehensive review of catastrophic faults in PV arrays: VII. BIOGRAPHIES
Types, detection, and mitigation techniques”, IEEE Journal of Albert Alexander Stonier received the B.E. degree in Electrical and
Photovoltaics, Vol. 5, No. 3, pp.982-997, May 2015. Electronics Engineering from Bharathiar University,
[7] Ye Zhao, Jean-François de Palma, Jerry Mosesian, Robert Lyons, Jr., India, the M.E. degree in Power Electronics and
and Brad Lehman, “Line–Line Fault Analysis and Protection Challenges Drives and Ph.D in Renewable energy from Anna
in Solar Photovoltaic Arrays”, IEEE Transactions on Industrial University, India in 2004, 2007 and 2014,
Electronics, Vol. 60, No. 9, pp.3784-3795, Sep. 2013. respectively. He joined the Department of Electrical
[8] Mariusz Malinowski, K. Gopakumar, Jose Rodriguez and Marcelo A. and Electronics Engineering, Kongu Engineering
Pérez, “A survey on cascaded multilevel inverters”, IEEE Transactions College in 2007, and is currently a Associate
on Industrial Electronics, Vol. 57, No. 7, pp.2197-2206, July 2010. Professor. He served as the Principal Investigator for
[9] Madhukar Rao. A, and K. Sivakumar, “A fault-tolerant single phase five research projects sponsored by the Government of
level inverter for grid independent PV systems”, IEEE Transactions on India related to power quality, converter
Industrial Electronics, Vol. 62, No. 12, pp.7569-7577, Dec. 2015. configuration, controller design and renewable
[10] V. Fernão Pires, D. Foito and Tito G. Amaral, “Fault detection and energy. He also received Raman research fellowship from Government of
diagnosis in a PV grid connected T-type three level inverter”, 4th India to pursue his post-doctoral studies in Northeastern University, USA.
International Conference on Renewable Energy Research and He has received 15 awards for the excellence in research from various
Applications, Palermo, Italy, 2015, pp.933-937. national and international organizations such as ISTE, IE (I), SPRERI, IET
[11] Hamed Nademi, Anandarup Das, Rolando Burgos, and Lars E. Norum, etc., and his special fields of interest include performance improvement in
“A new circuit performance of modular multilevel inverter suitable for solar PV systems, microgrid and smart grid.
photovoltaic conversion plants”, IEEE Journal of Emerging and
Selected Topics in Power Electronics, Vol. 04, No. 2, pp.393-404, June Brad Lehman (M’92–SM’08) is presently a Professor in the Department of
2016. Electrical and Computer Engineering at
[12] Surin Khomfoi, and Leon M. Tolbert, “Fault diagnostic system for a Northeastern University, Boston, MA. He performs
multilevel inverter using a neural network”, IEEE Transactions on research in solar energy, fault detection, and high
Power Electronics, Vol. 22, No. 3, pp.1062-1069, May 2007. density power converters with applications in solar
[13] Surin Khomfoi, and Leon M. Tolbert, “Fault diagnosis and energy, LED lighting, battery chargers, and
reconfiguration for multilevel inverter drive using AI based techniques”, telecommunication power supplies. Dr. Lehman
IEEE Transactions on Industrial Electronics, Vol. 54, No. 6, pp.2954- previously chaired the IEEE LED lighting standards
2968, Dec.2007. group (PAR1789). He is Editor-in-Chief of the IEEE
[14] Sunil Rao et.al., “An 18 kW solar array research facility for fault Transactions on Power Electronics and was recently
detection experiments”, Proceedings of the 18th Mediterranean Electro highlighted in the inaugural edition of the book The
Technical Conference, Limassol, Cyprus, 2016. 300 Best Professors, Princeton Review, 2012. He is
[15] Seok-Min Kim, June-Seok Lee and Kyo-Beum Lee, “A modified level- the recipient of the 2015 IEEE Power Electronic Society Modeling and
shifted PWM strategy for fault tolerant cascaded multilevel inverters Control Technical Achievement Award. Prior to his career as a professor,
with improved power distribution”, IEEE Transactions on Industrial Professor Lehman served as the head coach of the nationally ranked Georgia
Electronics, Vol.63, No.11, pp.7264-7274, Mar. 2016. Institute of Technology varsity swimming and diving team.

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

APPENDIX 1
Table A1 provides the list of various solar PV faults along with its causes, effects and mitigation methods. Table A2 shows
the various inverter faults with its causes and consequences.

TABLE A1. SOLAR PV FAULTS AND ITS MITIGATION METHODS ( SPECIFIES THE FAULTS CONSIDERED IN THE WORK)

S.No. Fault type Causes Effects Mitigation methods


1 Ground fault Unexpected or accidental connection Establishes low impedance and Ground fault detection and
between the current carrying conductors to significant current flows to the interruption fuse, residual current
the earth/ground. Also caused by the ground circuit. It results in the monitoring device, insulation
insulation damage in the cables and permanent reduction of output resistance measurement.
misfortunate short circuit in the PV voltage. Fire hazard, if not detected.
combiner box.
2 Line to line fault Short-circuit between current carrying Establishes low impedance between Usage of blocking diodes and over
conductors with different voltage two points in a solar PV array and current protection devices.
variations. reverses the current through the
faulty strings. Results in fire hazard.
3 Arc fault Discontinuity (series) or insulation Results in fire hazard. Arc fault circuit interrupter,
breakdown (parallel) in the current frequency spectrum analysis.
carrying conductors.
4 Open circuit fault Melting of over current protection devices Solar junction failure and cracking Inspection of the affected strings.
caused due to line to line fault under full in solar cells.
illumination.
5 Floating PV systems No connection between current carrying Shock hazard in the form of Resistive grounding, residual
conductors and the ground. discharge of capacitance. current monitoring device,
insulation resistance measurement.
6 Double ground fault If the conductors are not capable for Allows large amount of current to Fault detection within blind spot
carrying high current. flow through the grounding wire. range.
7 Blind spot Undetected fault and hence the fuse does Entire array current flows through Inspection on ground fault
not trip. the grounding wire and causes detection fuse.
severe damage to the array.
8 Under voltage Damage of solar cells or poor light. Reduced output power. Replacing the damaged solar cell.
9 Hot spot heating Non uniform irradiance and mismatch of Reduced output power. Bypass diodes can be used.
solar cell characteristics.

TABLE A2. INVERTER FAULTS WITH ITS CAUSES AND CONSEQUENCES

S.No. Fault type Causes Consequences


1 Overcurrent in semiconductor switch Damage in internal resistances Switch in fault mode and damage
2 Semiconductor switch cannot be driven Under-voltage of driving signal Inspection of control circuit required
3 Overheating of semiconductor switch Overvoltage or overcurrent Internal rupture of connection
4 Switch open circuit with healthy diode Non availability of gating pulses Unbalanced voltage and current with distorted
waveforms and reduced output power along with
increased harmonic content
5 Both switch and diode open Unavailability of gating pulses and over High voltage spike in the module due to the
reverse recovery voltage interruption of current in one direction

6 Only diode open (uncommon) Discontinuity in the connection between load High voltage spike in the module due to the
and switch or internal rupture of connectors interruption of current in one direction
7 Short circuit fault Continuous gating pulse, internal fault due to Causes immediate damage to the system and DC
overheating and freewheeling diode failure bus will get short circuited
8 Gate drive failure Non operation of control circuit, Inappropriate Unbalanced voltage and current with distorted
reference and carrier signals, non-uniform waveforms and reduced power
modulation regions
9 Total black out Defect in inverter or defect in control devices Complete shut down

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

APPENDIX II
Table A3 gives the complete analysis of entire system faults and the possibility in achieving the faults to obtain dataset
required for the intelligent controller.

TABLE A3. FAULT ANALYSIS


S.No. Fault Solar PV Battery Inverter
1 Ground fault Any one of the 4 PV panel outputs in Not tested Output of the individual H bridge is
a module is connected to the ground conneected to the ground (7 cases)
(28 cases)
2 Open circuit Any one of the 4 PV modules is kept i) One of the individual cells in a i) The terminals of the semiconductor switch
open (28 cases) battery is kept open (28 cases) excluding the diode is kept open (28 cases)
ii) One of 4 individual battery is kept ii) The terminals of the semiconductor switch
open (28 cases) including the diode is kept open (28 cases)
iii) The terminals of the diode excluding the
semiconductor switch is kept open (28 cases)
3 Short circuit Any one of the 4 PV modules is i) One of the individual cells in a i)The individual terminals of the
shorted with an external current battery is shorted (28 cases) semiconductor switches are shorted
carrying conductor (28 cases) ii) One of 4 individual battery is (28 cases)
shorted (28 cases) ii) The output of the individual H bridges is
shorted with an external current carrying
conductor (7 cases)
4 Line to line fault Connected the conductor acros the Not tested Connected the conductor across the
PV modules within the individual individual H bridge to the other H bridges
stages and also to other stages (21 cases)
(56 cases)
5 Floating PV The terminal connection is removed Not tested Not tested
from the individual PV module
(28 cases)
6 Double ground The conductor of less current (2A) Not tested Not tested
beyond the rated of 7A is used
across the PV modules (28 cases)
7 Blind spot The fuse rating is replaced with 12A Not tested Not tested
with the exiting rating of 10A and
induced a external overcurrent fault
with 10.5A in the individual PV
module (28 cases)
8 Undervoltage Degraded the individual solar cell by The charging phase is bypassed once One of the individual H bridge is left open
removing it from the module the battery discharged completely (7 cases)
(28 cases) (7 cases)
9 Overvoltage Not tested Not tested Connected the two sent of battery ouput
terminals to the individual inverter stage
(7 cases)
10 Partial shading Allowed the individual PV module Not tested Not tested
with complete, medium and partial
shading (28 cases)
11 Overcurrent Not tested Not tested Input resistance of the individual
semiconductor switch is varied (28 cases)
12 Gate terminal fault Not tested Not tested i) Gate pulse loosing: Controlled the PWM
output port to reach the switch gate terminal
(28 cases)
ii) Gate drive failure: The output ports
directed towards PWM pulses is terminated
(28 cases)
13 Overheating Not tested Not tested The heat sink for the individual
semiconductor switch is removed and the
circuit is subjected to an inductive load
(28 cases)
14 Total black out Not tested Not tested The complete inverter power circuit is turned
OFF (7 cases)
15 Overcharging Not tested The battery is allowed to charged Not tested
than its limit (SoC calculated)
16 Overdischarding Not tested The battery is allowed to discharge Not tested
than its limit (DoD calculated)

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TEC.2017.2786299, IEEE
Transactions on Energy Conversion

APPENDIX III

Figure A3. Discharge charateristics of 48V, 100Ah battery

Based on the characteristics it is found that when the SoC is greater than 50%, the battery attains the maximum voltage
leading to overcharging condition. Similarly the voltage reduces while the SoC is less than 50%.

0885-8969 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Вам также может понравиться