Measurement 183 (2021) 109771

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

A Tuned classification approach for efficient heterogeneous fault diagnosis

in IoT-enabled WSN applications
S. Lavanya a, 1, A. Prasanth b, 2, *, S. Jayachitra c, 2, A. Shenbagarajan d, 1
a
Muthayammal Engineering College, Namakkal, India
b
Sri Venkateswara College of Engineering, Sriperumpudur, India
c
Karpagam Academy of Higher Education, Coimbatore, India
d
AAA College of Engineering and Technology, Sivakasi, India

A R T I C L E I N F O A B S T R A C T

Keywords: The advancement of the Internet of Things (IoT) technologies will play a significant role in the growth of smart
Fault diagnosis cities and industrial applications. Wireless Sensor Network (WSN) is one of the emerging technology utilized for
Heterogeneous Faults sensing and data transferring processes in IoT-based applications. However, heterogeneous faults like hardware,
Internet of Things
software, and time-based faults are the major determinants that affect the network stability of IoT based WSN
Network Stability
Wireless Sensor Network
(IWSN) model. In this paper, a novel Energy-Efficient Heterogeneous Fault Management scheme has been pro­
posed to manage these heterogeneous faults in IWSN. Efficient heterogeneous fault detection in the proposed
scheme can be achieved by using three novel diagnosis algorithms. The new Tuned Support Vector Machine
classifier facilitates to classify the heterogeneous faults where the tuning parameters of the proposed classifier
will be optimized through Hierarchy based Grasshopper Optimization Algorithm. Finally, the performance re­
sults evident that the diagnosis accuracy of the proposed scheme acquires 99% and the false alarm rate sustains
below 1.5% during a higher fault probability rate. The diagnosis accuracy rate is enhanced up to 17% as
compared with existing techniques.

1. Introduction
Internet of Things has been forecasted as one of the most prevalent
contemporary technologies that encompasses several smart objects
gathered to provide reliable information. These smart objects are
distributed over the targeted area to monitor physical circumstances [1].
Typically, the IoT architecture can be categorized into three layers:
sensor, communication, and application [2]. The significant role of the
sensor layer has to gather the physical observations. The communication
layer is responsible for the transmission of data acquired by the IoT
sensors. Finally, the application layer accords with human–machine
interaction and evaluating the information. Here, the data is supervised
by several network systems and engrossed through different physical
devices [3].
The IoT paradigm brings smart electronic objects to interact with the
physical environment. WSN is one of the most conventional technologies
to track the periodic status of these smart objects [4–6]. The low-
powered, tiny-sized, and low-cost sensor make it more competent to
integrate with IoT. It is the major component of IoT to sense the precise
physical observations and transmit the information to the central sink
for more appropriate action. Nowadays, IoT-based WSN (IWSN) inte­
gration becomes smarter for several applications such as smart agri­
culture, smart grid, remote health monitoring, and disaster detection
[7]. These smart applications necessitate real-time analytics for sys­
tematic decision making and consistent network process. IWSN offers
unrestricted storage and computation capability for efficient data
analysis in smart applications [8].
The sensor nodes are usually operated with limited battery energy in
the IWSN environment. The replacement of the battery becomes
complicated once they deplete their entire battery resources [9–11].
IWSN are vulnerable to faults due to their deployment in unmonitored
and harmful circumstances. Furthermore, the interruption of faults

* Corresponding author at: Department of Electronics and Communication Engineering, Sri Venkateswara College of Engineering, Sriperumpudur, Tamilnadu,
India.
E-mail addresses: lavanyasharvesh@gmail.com (S. Lavanya), draprasanthdgl@gmail.com (A. Prasanth).
1
Associate Professor, Department of Computer Science and Engineering,
2
Assistant Professor, Department of Electronics and Communication Engineering,

https://doi.org/10.1016/j.measurement.2021.109771
Received 19 March 2021; Received in revised form 3 June 2021; Accepted 14 June 2021
Available online 27 June 2021
0263-2241/© 2021 Elsevier Ltd. All rights reserved.
S. Lavanya et al.
during normal network function prompts hazardous consequences such
as environmental, commercial, and human life losses. To alleviate these
circumstances, the faults are classified into different classes. The key
motive is to identify and classify the faults in an initial period. The
appropriate measures can be established to resolve them or to select an
alternate solution to avoid any harmful circumstances [12]. Hence, it
plays a vital role to identify the failure node in the sensing field to
construct the accurate decision for IWSN based smart applications.
In IWSN, faults can be categorized into three classes: hardware,
software, and time-based failures [13]. Hardware faults commonly arise
owing to the damage of hardware modules of the deployed sensor nodes.
Some of the hardware faults are transceiver unit, microcontroller unit,
battery unit, and sensor unit faults. These faults cause the function of the
faulty sensor node to entirely collapse so that readings of sensed data
cannot be attained for the particular node. Most of the existing fault
diagnosing schemes are not able to detect different hardware malfunc­
tion of sensor nodes. However, they only examine the software and time-
based failures that increase the number of dead nodes in the network. In
this scenario, the fault-free nodes are failed to reach the sink node and
thus it drives the network partitioning problem. As a consequence, the
whole network is exposed to premature death due to the separation of
the failed nodes.
In case of software failure, the nodes can generate and convey
erroneous data which feasibly reduces the quality of service [14]. Some
of the software failures are as follows, (a) Gain fault: This fault occurs
due to the variation of the observed data that is different from the ex­
pected reading data. (b) Offset fault: It states that the improper value
being added to the observed data. This variety of faults arises owing to
the poor calibration of the sensing unit. (c) Stuck-at fault: The data series
undergoes zero or almost zero deviation in the observed data gathered
by the sensor node is termed as a stuck-at fault.
In general, time-based failures are classified as permanent, inter­
mittent, and transient faults. The permanent fault category produces
fallacious results constantly in deployed sensor nodes. Owing to tem­
porary changes in geographical conditions, transient failures arise in the
network. The stability is not constant since it occurs only for a short
interval and rectifies automatically. In contrast, intermittent failures
progress for a prolonged duration when compared with transient fail­
ures. Such kinds of faults are not difficult to detect and classify in the
IWSN environment. The various heterogeneous faults of IWSN are
Fig. 1. The Structure of Heterogeneous Fault Management in IWSN.
2
Measurement 183 (2021) 109771
depicted in Fig. 1.
In case, the heterogeneous faults cannot be identified properly on a
timely basis, then it will impact the network stability and provoke
imprecise sensed data in the sensing field. The network stability is a
significant metric that illustrates how fault diagnosis schemes manage
the faults and effectively transmit the sensed data to the sink node. The
network stability has a direct impact on the performance of the entire
network where it can be assessed based on operating status and esti­
mating detection time. Subsequently, the battery energy of the sensor
nodes is continuously consumed owing to its larger number of inaccu­
rate data are produced by faulty nodes [15,16]. Henceforth, the
assessment of the network stability and energy consumption are the
essential performance metrics in the fault diagnosis scheme of IWSN.
Besides, the faulty node affects the operation and decisions of the whole
network. The identification of faulty data can be exploited as a reference
for the network update. Consequently, the implementation of an effec­
tive heterogeneous fault detection scheme is most significant to enhance
the Quality of service and network stability of the IWSN.
In this paper, the Energy-Efficient Heterogeneous Fault Management
(E2HFM) scheme has been proposed to detect and classify the hetero­
geneous faults of the scattered nodes in IWSN. The E2HFM scheme in­
corporates two methods: heterogeneous fault diagnosis and
classification. Primarily, the heterogeneous fault diagnosis method
employs three novel diagnosis algorithms to detect the different het­
erogeneous faults includes hardware, software, and time-based faults.
Thereafter, these three diagnosis algorithms are formulated into a
sequential process of the proposed E2HFM scheme. Finally, the hetero­
geneous fault classification stage of the proposed scheme utilizes the
Tuned Support Vector Machine (TSVM) to classify the faults according
to the fault types. The proposed Hierarchy based Grasshopper Optimi­
zation Algorithm (HGOA) is the incorporation of a hierarchy approach
with traditional Grasshopper Optimization Algorithm (GOA) in such a
way that the acquired algorithm aims at optimizing the tuned parame­
ters in the TSVM classifier. The accurate diagnosis and classification in
the proposed E2HFM scheme comprehend in improving the network
stability of IWSN.
The various novel contributions of this article can be summarized as
follows:
S. Lavanya et al. Measurement 183 (2021) 109771

• The proposed E2HFM scheme considered all the heterogeneous faults
like hardware, software, and time-based faults as compared with the
existing fault management schemes. This simultaneous assimilation
of heterogeneous faults can potentially mitigate the premature death
of the IWSN.
• The proposed E2HFM scheme implemented three new diagnosis al­
gorithms to detect multiple heterogeneous faults in IWSN. It de­
creases the communication overhead and robust to variation of
topology during the fault detection stage. Thus, the proposed scheme
plays a key rationale to find out the type of any heterogeneous faults
within the sensing field.
• The novel Hybrid TSVM-HGOA (TH) classifier has been implemented
in the proposed E2HFM scheme to classify the heterogeneous faults in
IWSN. The objective of the proposed HGOA is to optimize the TSVM
parameters and establish the finest features subset simultaneously.
However, the performance of traditional GOA extremely depends on
accomplishing a balance between exploitation and exploration.
Hence, the proposed HGOA addresses this problem by incorporating
the hierarchy of a grasshopper in the update strategy.
• The simulation environment has been assessed in two different sce­
narios: IWSN#1 and IWSN#2 where a new metric (Network Stabil­
ity) is hosted with essential metrics to analyze the performance of
proposed as well as existing research works.
The rest of the article is structured into several sections as follows.
Section 2 deliberates the summary of the related work. Section 3 in­
troduces the proposed E2HFM scheme along with its underlying
frameworks. The performance evaluation is depicted in Section 4. Sec­
tion 5 summarizes the simulation results and discussions. Finally, the
conclusion and future work are described in Section 6.
2. Related work
The following section outlines the previous state-of-the-art ap­
proaches related to the heterogeneous fault diagnosis and classification
issues of IWSNs. A comparison of several fault management approaches
is summarized in Table 1. In [17], a new Improved Negative Selection
Algorithm (INSA) has been introduced in which a bounded time period
mechanism and majority voting strategies are preferred to detect the
time-based faulty nodes in IWSN. The INSA comprises two stages,
namely, detector generation and matching stage. Primarily, a detector
set is formed by comparing individual information with randomly pro­
duced data. If observed data matched with the detector set, then the
node is recognized as faulty. The INSA scheme uses the Euclidean dis­
tance matching to compute the matching rule. Thereupon, the time-
based faults are classified through the traditional SVM technique. The
performance results depict that the INSA scheme offers poor latency and
better accuracy than the existing schemes. Here, the damaged nodes are
compared with the neighbor nodes set in the voting-based methodology
of INSA. While fault rate increases in the sensing field, it degrades the
performance of the INSA scheme.
The sensor nodes provide unpredicted responses owing to the
extensive range of real-time applications and resource restrictions. This
makes a variety of node and link failures in IWSN. A Node-Link Failure
Fault Tolerance (NLFFT) scheme is presented to manage the faults
caused either by sensor node or link failure during data communication
[18]. The NLFFT scheme comprises two mechanisms: improved-handoff
and quadratic minimum spanning tree mechanism. The improved-
handoff mechanism establishes a way to identify the failure node
caused by poor battery power. Then, the faulty node is replaced with
appropriate neighbor nodes. The second mechanism deals with link
failure and replaces the damaged edge with an alternative edge in the
communication link. These two mechanisms facilitate the NLFFT
scheme to achieve better accuracy and throughput. However, the se­
lection of alternate edge develops the additional communication over­
head and affects the network stability of IWSN.
A Fuzzy Logic-based Self-Diagnosis (FLSD) has been proposed to
address the various faults in IWSNs [19]. This scheme employed a
spatial correlation-based methodology to identify the failure nodes in
the sensing area. The fuzzy logic controller assigns the weights to every
node based on its sensed values. In case, many neighboring nodes have
different weighted values, then it can be suspected as faulty nodes. The
FLSD scheme attains lesser energy consumption and higher network
lifetime. Despite that, the detection accuracy of the FLSD scheme
entirely depends upon the consistency of neighboring sensor nodes.
To accurately identify the failure nodes, a new fault detection
approach based on Fireworks Algorithm with Convolutional Neural
Network (FWA-CNNFD) scheme is introduced in IWSN [20]. The initial
weights and biases of the CNN are computed and the observed values are
exploited as the initial population of the FWA. Thereafter, the optimum
weight and bias are acquired by using an improved mutation operator,
mapping rule, and selection scheme. These optimum values are assigned
as the initial weight and bias values of the CNN to formulate the effective
fault detection model. Nevertheless, the training process of the FWA-
CNNFD scheme is quickly dropped with local optima. This leads to
attained lower accuracy during the higher fault probability rate in
IWSN.
A Rule-based Soft Fault Diagnosis (RSFD) scheme has been estab­
lished to address the software faults in IWSN [21]. This scheme exploits
a type-2 fuzzy-model in which every node gets fault status from its
neighbor nodes with respect to deterministic interconnection topology.
A reference residual mechanism is a prerequisite in the RSFD scheme to
analyze the behavior of the residual battery energy of sensor nodes.
Subsequently, a fuzzy filter can be framed out to determine the final
fault level of every node. The main constraint of the RSFD scheme is that
it necessitates a heavier message exchange during the fault diagnosis
stage which significantly decreases the stability of the network. More­
over, while the number of nodes with fault possibility increases, the

Table 1
Comparison of existing schemes over different fault tolerance parameters.
Schemes Heterogeneous Faults Addressed Classifier Network Stability Energy Efficiency Detection Accuracy
Hardware Software Time-based

INSA [17] SVM Moderate Low Moderate

NLFFT [18] Not Addressed Very Low Moderate High
FLSD [19] Fuzzy Logic Moderate High High
FWA-CNNFD [20] CNN Moderate Very Low Very Low
RSFD [21] Not Addressed Low Moderate Moderate
SAFT [22] Not Addressed Low Low Moderate
FDRFC [23] Random Forest Very Low Low Low
CFTR [24] Not Addressed Moderate Very High High
NNFD [25] Neural Network Low Very Low Moderate
HFD[26] Neural Network Moderate High High
NFOM [27] ANFIS Very Low Very High Moderate
DSFD [28] SVR Low Very Low Moderate
A-NSGA[29] Genetic Algorithm Low Moderate Very Low

3
S. Lavanya et al.
computational rate of the RSFD scheme also increases.
In [22], a novel Self-Adaptive and Fault-Tolerant (SAFT) routing
scheme is developed to address the fault diagnosis problems in edge
computing IWSN. The SAFT primarily built the clusters in which the
header node regulates their communication range with respect to
transmission load and network circumstances. It mainly focused on the
member and header node failures in IWSN. Afterward, the constructed
routing process has been accomplished to recover from the faulty phase,
and thus the particle swarm optimization aids to initiate the routing
process. The sensor nodes in the SAFT scheme identifies the different
time-based faults autonomously and quickly. But, the SAFT scheme fails
to focus on hardware and software faults in IWSN.
Owing to the limited battery resources of sensor nodes and diverse
deployment environments, fault diagnosis is a major challenging
endeavor in IWSN. To mitigate this, the Fault Detection through the
Random Forest Classifier (FDRFC) is proposed for IWSN [23]. The
FDRFC scheme focused to identify the software faults of the deployed
nodes. The appearance of the software faults during ordinary operation
can outcomes in harsh environmental losses. To alleviate these calami­
tous circumstances, software faults are classified with the assist of
Random Forest Classifier in the FDRFC scheme. The first stage of FDRFC
is to generate an RF decision tree where n number of random attributes
are chosen from the total attributes set. Subsequently, the faulty data
can be classified according to the generated RF decision tree. The fault
detection results of the FDRFC scheme entirely based on the training
dataset. If the training dataset is valuable, it is likely to acquire superior
accuracy in the diagnosis results.
In [24], a Cluster-based Fault-Tolerant Routing (CFTR) scheme has
been presented to address the header node failures in clustering-based
IWSN environment. In CFTR, a low-overhead approach can be carried
out to tolerate the sudden damage of the header node. This approach is
mainly categorized into two stages, namely: fault diagnosis and fault
recovery. These two stages function parallel with the regular network
process. The member nodes detect the faulty header node if it does not
receive any acknowledgment frame from its header node for the trans­
mitted data frames. Consequently, if a header node is damaged, its
corresponding member nodes connect with another header node. The
CFTR scheme has the ability to ensure the normal operations of the
IWSN despite of header node failure. However, the reroute process
frequently occurs between header node and member nodes in CFTR that
cause superior packet overhead problem in IWSN.
A Neural-Network based Fault Diagnosis (NNFD) scheme has been
introduced to examine the time-based fault circumstances for IWSN
[25]. The NNFD scheme executes in four stages: clustering, communi­
cation, fault tolerance, and isolation stage. Specifically, the gradient
descent and evolutionary methods promote to diagnose as well as isolate
the damaged nodes in IWSN. The evolutionary method offers better
accuracy than the gradient descent method because the solution space of
the evolutionary method enlarges while increasing the generalization
ability in the network. Like [23], this NNFD scheme entirely depends
upon the training dataset which degrades their performance in the case
of a dense environment.
In [26], the Heterogeneous Fault Diagnosis (HFD) scheme is devel­
oped to address the hardware and time-based faults for IWSNs. This
scheme employs a variance model analysis and time-out strategy to
identify the faulty sensor nodes in the sensing field. Further, these faults
have been classified with the aid of the probabilistic neural network
method. The performance of the HFD is assessed through simulation as
well as testbed environment by varying the fault rate from low to high.
The HFD scheme obtains a higher fault detection accuracy and classi­
fication rate. In case, the leader node pretended to be faulty, then the
fault detection of a remaining node is defective. As a result, the HFD
suffers from single-point failure that reduces the network stability of the
entire network.
A Neuro-Fuzzy Optimization Model (NFOM) has been proposed for
fault management in IWSNs [27]. The NFOM scheme uses an Adaptive
4
Measurement 183 (2021) 109771
Neuro-Fuzzy Inference System (ANFIS) classifier for detecting different
software and time-based faults. The classifier adopts the fuzzy inputs
from the node metrics and yields fuzzy output through the imple­
mentation of fuzzy rules. Eventually, the obtained fuzzy output is
applied to a defuzzifier that delivers an output about the fault status of
the deployed nodes. It exploits a least-square approach to train the
membership function and modeling the observed data. The faulty node
can be isolated and not permitted to contribute in the consecutive stage
that avoids the network partitioning issues in the sensing field. At the
same time, the faulty node is recovered by its single-hop neighbor
having the greatest residual energy and thus it affects the network sta­
bility of IWSN.
In [28], A Distributed SVR Fault Detection (DSFD) scheme has been
established to address the complexity and accuracy of the fault diagnosis
model. Initially, a typical fault prediction is constructed through a
support vector regression strategy. It uses redundant data gathered by a
sensor node that attains residual sequences. Thereupon, the mutual
testing model can be carried out to detect the node status. This testing
model enlarges the detection accuracy of DSFD and preserves the packet
overhead produced by regular interactions among multiple nodes.
Nonetheless, when the number of neighbor nodes is fewer, the failure
ratio of the sensor node is greater in the network. As a result, the fault
diagnosis accuracy of the DSFD model reduces rapidly during a higher
fault probability rate.
Fault management in IWSN circumstances is a major daunting task as
compared with the wired network. A new framework called Adapted
Non-dominated Sorting-based Genetic Algorithm (A-NSGA) has been
introduced to minimize the fault occurrence in the network [29]. The
basic elements of the A-NSGA scheme consist of crossover, chromosome,
and mutation operators. The fault tolerance can be exhibited as link
attribute and density of neighbor nodes in IWSN circumstances. This A-
NSGA scheme needs a higher number of packet exchanges during the
diagnosis process that potentially reduces the stability of the network. In
[30–32], a neural network-based fault diagnosis scheme has been
emerged to address the fault tolerance in IWSN. This scheme employs
different strategies for fault detection, however, the scheme has not
investigated various hardware faults of the scattered nodes. Further­
more, it consumes more battery energy for fault tolerance that impacts
the network stability of IWSN.
In [33], a Genetic Algorithm based Fault Tolerance (GAFT) scheme is
presented to investigate the failure nodes in clustering-based IWSN. The
sensor nodes are arranged and the header node is elected for every group
through a distance algorithm. The elected header node supervises the
corresponding member nodes to recognize that whether a member node
is faulty or non-faulty. Afterward, the genetic algorithm has been
implemented to select the finest node as backup nodes where coverage
and battery energy metrics are considered during backup node selection.
Once the faulty header node is recognized in the sensing field, the
selected backup nodes will replace the faulty nodes. This backup node
methodology decreases the energy depletion and faulty environment in
IWSN. At the same time, the single point failure arises in GAFT since
backup nodes act as an essential activity in header nodes as well as
member nodes.
According to the aforesaid fault management analysis, many of the
approaches failed to address different varieties of heterogeneous fault at
a time in IWSNs. In addition, network stability is not properly sustained
in most of the existing work. On contrary, the current state-of-art directs
the research towards the detection and classification of different het­
erogeneous faults simultaneously in the sensing field. The proposed
work will examine all heterogeneous faults such as hardware, software,
and time-based faults that prolong the network consistency of IWSN.
The classification of these faults, as addressed in this work, brings up a
new concern.
The poor fault management can affect the network stability as it has
a direct impact on the entire network. Therefore, the proposed scheme
completely analysis the heterogeneous faults in different preceding
S. Lavanya et al. Measurement 183 (2021) 109771

stages to detect the faulty and non-faulty node, and further it classifies
them into correspondent fault types. It provides reliable accuracy as well
as network stability for a higher fault probability in IWSN. Alternatively,
existing fault management techniques consume additional battery en­
ergy of the node to detect and classify the heterogeneous faults in the
network. Hence, the proposed work figures out the fault information and
pays more exploration to battery energy of the sensor node in the fault
detection process.

3. Proposed Energy-Efficient heterogeneous fault management

2
The proposed E HFM scheme involves two stages: heterogeneous
fault diagnosis and heterogeneous fault classification. A detailed
description of these two stages is demonstrated in the following sections.
The framework of the proposed method is exposed in Fig. 2. The
framework incorporates the multiple heterogeneous fault diagnosis ap­
proaches into a sequential process of the proposed E2HFM method in the
IWSNs.
3.1. Heterogeneous fault diagnosis
The proposed heterogeneous fault diagnosis approach detects the
hardware failure includes transceiver unit, microcontroller unit, battery
unit, and sensor unit faults of the deployed nodes. Primarily, the sink
node forwards a Hardware Fault Diagnosis Information (InfoHFD ) to the 1
hop neighbors nodes set within the radius of 2R and initiates the timer
(∂). In this work, the battery of the sink is assumed to be more powerful
compared to the normal nodes. Fig. 3 shows the InfoHFD message format
in which the first field holds the frame identification ID and the next two
fields denote source and destination IDs. The fourth and fifth fields
indicate the health status and the last two fields designate the
geographical location of the sensor node.
If a single-hop node set gets the InfoHFD frame from the sink, every
Fig. 3. The InfoHFD message format.
sensor node formulates a health status and then sends back its status to
the sink. If the sink node receives the InfoHFD from a node (ni ) within ∂
time, then it takes a decision that the microcontroller and transceiver
units of the correspondingni are operating normally. Next, the battery
unit fault of ni is identified on the basis of the current NBR value. If the
residual battery (RBi ) of ni is lesser than a threshold (T2 ) value, then it
deduces that the battery unit of ni is faulty. Finally, the detection of a
sensor unit fault of a node can be achieved based on SCR of the sensor
unit. If an InfoHFD frame contains the NCR value of ni , then it takes a de­
cision that the sensor unit of a node ni is non-faulty otherwise it is
recognized as faulty. Similarly, all sensor nodes broadcast the InfoHFD
message to complete the mechanism of hardware fault detection which
is illustrated in Algorithm 1.
Algorithm 1 – Hardware Fault Diagnosis Algorithm
BeginN = {n1,n2,….,nk} /*set of deployed nodes*/Nhf = {nhf1, nhf2,…., nhfq} /*set of
hardware faulty nodes*/1 Initially, sink broadcasts a InfoHFD in the range 2R /* Fault
diagnosis process*/2 Initiate ∂ timer3 if (∂ ≤ T1 ) then /* Hardware fault diagnosis*/4
Receive InfoHFD frame for nodeni ;5 Microcontroller and Transceiver units of ni is
non-faulty;6 if (RBi ≤ T2 ) then7 Battery unit fault has appeared inni ;8 Nodeni is
inserted intoNhf ;9 end if10 else11 Microcontroller and Transceiver units of ni is
faulty;12 Nodeni is inserted intoNhf ;13 end if14 if InfoHFD content NCR then15
(continued on next page)

Fig. 2. The framework of proposed E2HFM method.

5
S. Lavanya et al. Measurement 183 (2021) 109771

(continued ) identification is realized through examining the different types of data

Algorithm 1 – Hardware Fault Diagnosis Algorithm gathered by the sensor node. The execution process can be defined as
Sensor unit of ni is non-faulty;16 else17 Sensor unit of ni is faulty;18 Nodeni is
follows:
inserted intoNhf ;19 end if20 end ( )
Ojk (t) = c3 Zij (t), γ (6)

where γ refers to the set of parameters involved in the conversion

In case of software failures, the proposed approach detects the gain, function and the conversion of the sensor node data to the optimal
offset, and stuck-at faults of the deployed nodes. The key objective of feature is represented by c3 (.). Consequently, the optimal feature of the
Algorithm 2 is to assess the sensor reading variations as a software fault. determination of the software fault type process can be evaluated as:
In this work, an optimal threshold value has been computed to diagnose ( ( ) )
x(t) = c2 c3 Zij (t), γ , β (7)
the software fault as exposed in Eq. (1).
⃒ (
⃒1 ∑
)⃒
⃒ According to the aforesaid analysis, the characteristic features of the
2 ∑3
T3 = ⃒
⃒
NRR (t − 3) −
⃒
ni (NCR (t) ) ⃒ (1) node will change when software faults have appeared. This change of
⃒3 t=0 t=1
⃒ software fault type will promote the conversion of data characteristics.
Finally, the sink node determines the software fault type by mapping a
where NRR (t − 3) denotes the previous three consecutive sensors
set of attributes to the set of software fault types.
original reading and ni (NCR (t) )signifies the next three consecutive sen­
Algorithm 2 – Software Fault Diagnosis Algorithm
sors current reading at a specified time t.
According to the T3 value, the software faulty node has been detected BeginN = {n1,n2,….,nk} /*set of deployed nodes*/ Nsf = {nsf1, nsf2,…., nsfq} /*set of
during subsequent rounds. If |NRR − ni (NCR ) | is less than or equal to the software faulty nodes*/1 if (∂ ≤ T1 ) then2 Receive InfoHFD frame for nodeni ;3
Compute T3 using Eq. (1);4 if (|NRR − ni (NCR ) | ≤ T3 ) then5 Software unit of ni is
T3 value, then it decides that the software unit of ni is non-faulty, non-faulty;6 else7 Software unit of ni is faulty;8 Nodeni is inserted intoNsf ;9
otherwise, it is recognized as software faulty. Once the software fault Determine software fault type using Eq.;10 end if11 end if12 end
is detected, it is necessary to find out the type of software fault.
The fault diagnosis process receives sensor information at time t can
be defined as follows: The proposed scheme detects the major types of time-based failures such
( ) as permanent, intermittent, and transient faults of the deployed nodes.
x(t) = c1 Zij (t), α (2) Algorithm 3 depicts the step-by-step procedures involved in a time-
where x(t) provides the non-optimal decision result of the sensor based fault diagnosis scheme. Primarily, every node sends its current
node (fault and non-fault results), c1 () represents the conversion func­ sensed data (NCR ) to its single-hop neighbors for u rounds. Despite of
tion from node reading data to fault diagnosis results and α refers to the detecting the time-based faults, every node additionally maintains a
counter variable (Ci ) and a Boolean variable (Bi ;whereBi ∊{0,1}) that are
set of parameters involved in the conversion function. In addition, Zij (t)
calculated for every time instance. Next, the sensed data of every node
indicates a fault diagnosis dataset that holds input data for all nodes and
and its neighbors are compared with every instance of the time cycle. If
it can be evaluated as: ⃒ ⃒
⃒nij (NCR ) − nji (NCR ) ⃒ is greater than the predefined threshold (T4 ) value,
⎡ ⎤
1
then the neighbor nodes increment the Ci value for every testing process.
⎢ z1 z12 ⋯ z1i ⎥ The evaluation of T4 depends on the user-specific and IWSN appli­
⎢ 2 ⎥
⎢ z1
⎢ z22 ⋯ z2i ⎥
⎥ cation. After the testing process, the Ci value is compared with majority
Zij (t) = ⎢ ⋮ ⎥ (3)
⎢ ⋮
⎢
⋮ ⎥
⎥ neighbors (NM ), where NM = neighbors(n
2
i)
. If the Ci value is greater than or
⋱
⎣ j
z1 zj2 zij ⎦ equal to NM , then set Bi variable as 1 for that time instance (t), otherwise
Bi (t) = 0. Thereafter, a time-based fault classification function (FCi ) has
been calculated for every node.
where zji (t) represents the ith type of data gathered in the jth node at
time t. It is difficult to employ the attained data directly to classify the ∑
u
1
FCi = Bi (t) × (8)
fault category. Hence, the corresponding optimal features in the sensor t=1
u
data should be formulated. The software fault type formulation of the
node at time t is depicted as follows: An assumption has been made for the following threshold range that
( ) would fix for a different type of time-based faults. In the case of a per­
x(t) = c2 Ojk (t), β (4) manent fault, the lower bound threshold value is LTp = 0.8 and the
higher bound threshold value is HTp = 1.0. The lower and higher bound
where c2 (.)represents the conversion function from the optimal
threshold values of intermittent fault are LTi = 0.3 and HT i = 0.79
feature to the detection result. β refers to the set of feature parameters
respectively. Finally, the fault rate of transient fault is considered as
involved in the conversion function. x(t) intends the type of software
LTt = 0.01 andHTt = 0.29. The FCi value for sensor ni will be compared
faults that comprises gain, offset, and stuck-at faults. Ojk (t) indicates an with various fault rates to identify the real fault condition of the node.
optimal feature set to measure the fault type. The various software fault The real fault condition RF(ni ) of node ni can be computed as follows:
types can be acquired through the attribute extraction of Zji (t)and it is ⎧
formulated as ⎪
⎪ LT ≤ FCi ≤ HT p permanentfault
⎪ p
⎪
⎨
⎡ ⎤ LT i ≤ FCi ≤ HT i intermittentfault
RF(ni ) = (9)
1
o12 o1k ⎥ ⎪
⎪ LT t ≤ FCi ≤ HT t transientfault
⎢ o1 ⋯ ⎪
⎪
⎢ 2 ⎥ ⎩
⎢ o1 o22 ⋯ o2k ⎥ otherwisenon − faulty
⎢ ⎥
Ojk (t) = ⎢ ⋮ ⎥ (5)
⎢ ⋮
⎢
⋮ ⎥
⎥ Every sensor can be determined as faulty by NM neighbors in the
⋱
⎣ j
o1 oj2 okj ⎦ sensing field. If a node ni inclined towards a healthy node is identified by
the diagnosis algorithm, then the surrounding neighbors are designated
as healthy neighbors in the network.
where ojk (t) represents the kth optimal feature value in the jth node at Algorithm 3 – Time-Based Fault Diagnosis Algorithm
time t. The optimal features are obtained from the attribute information
of the node. The design of the optimal feature of the fault type (continued on next page)

6
S. Lavanya et al. Measurement 183 (2021) 109771

(continued ) { }
1 τ ∑n
Algorithm 3 – Time-Based Fault Diagnosis Algorithm min β β+s σi (12)
β,e,σ 2
BeginN = {n1,n2,….,nk} /*set of deployed nodes*/ Ntf = {ntf1, ntf2,…., ntfq} /*set of time- i=1

based faulty nodes*/1 Initialize the variables Ci = 0 and predefined T4 value;2 for
Subject to,
each node ni ∊N do /* Fault diagnosis process*/3 for each instance t = 1 to u do4 Node
ni sends NCR to its set of neighbors;5 end for6 for each instance t = 1 to u do7 for wi = (βτ vi − e) ≥(1 − σ i ) (13)
⃒ ⃒
each neighbor do8 if (⃒nij (NCR ) − nji (NCR ) ⃒ > T4 ) then9 Increment Ci variable;10
end if11 end for12 if (Ci ≥ NM ) then13 Bi = 1;14 else15 Bi = 0;16 end if17 end where σ intends the slack variable exploited to represent the allow­
for18 Compute FCi by using Eq. (8);19 if LTp ≤ FCi ≤ HTp then /* Time-based fault able degree of classification error. In this work, the value of σ is
diagnosis*/ 20 Node ni is permanent faulty node ;21 Nodeni is inserted intoNtf ;22
considered as 0.001. The soft parameter is used to balance the decision
else if LTi ≤ FCi ≤ HTi then23 Node ni is intermittent faulty node ;24 Nodeni is
inserted intoNtf ;25 else if LTt ≤ FCi ≤ HTt then26 Node ni is transient faulty node
margin and train error.
;27 Nodeni is inserted intoNtf ;28 else29 Nodeni is non-faulty;30 end if31 end for32 A kernel function has been utilized in the SVM classifier for sepa­
The status of fault information is broadcasted in the network;33 end rating the data nonlinearly and it is given by
( )
∑
n
f (vi ) = sign wi γi k(vi , v,i ) + e (14)
3.2. Proposed Hybrid TSVM-HGOA (TH) classifier i=1

where γ depicts the Lagrange multiplier, and k(.) signifies the kernel
The proposed Hybrid TH classifier has been exploited for classifying
function which is the inner product of the two vectors vi and v,i .
the heterogeneous faults in IWSN. Here, the optimal values of TSVM
In this study, a radial basis function can be employed as a kernel
parameters are chosen through HGOA during the training process. The
parameter that enlarges the gap between the two classes, thereby
key reason for selecting the TSVM classifier is to provide a global opti­
structuring the hyperplane optimum. The radial basis function is
mum and does not converge into local minima. The selection of appro­
computed as follows:
priate TSVM parameters is most essential for better performance results
( )
and achieves higher accuracy for heterogeneous fault classification in ∑
n
( )2
IWSN. k(vi , v,i ) = exp − δ vij , v,ij (15)
j=1

3.3. TSVM model
Support Vector Machine (SVM) is characterized with conventional
supervised models that are enlightened as one of the robust techniques
in the machine learning applications [34]. To implement the idea of
structural risk minimization, the SVM attains superior generalization
ability in the training phase. It has better competence to exploit com­
posite and non-linear relations. The lesser susceptibility to local minima
and an efficient outcome demonstration are additional features of the
SVM classifier. These amenities stimulate SVM as one of the most dy­
namic classifiers for heterogeneous faults in IWSN.
A typical heterogeneous fault classification issue would be trans­
posed to optimization constraints through implementing the meta­
heuristic algorithm with the SVM classifier. In practice, the SVM
operates for both regression and classification processes. It adopts to
produce high-attribute decision boundaries based on the selected subset
of gathering data. The feature subset selection can be exploited to train
the SVM classifier which has a direct impact on the classification accu­
racy. It plays a vital role in supervised learning while operating in high-
dimensional datasets.
The SVM classifier generally creates linear separating hyperplanes in
high-dimensional feature data set where every data can be noticed as
(vi , wi ), wherevi represents the feature vector and wi refers to class label
value{-1,1}. In addition, vi is expressed as follows:
vi = vi1, vi2,⋯, vin ; i = 1, ⋯, m
where n and m intend the number of features and the number of
training data respectively.
The sample features are bisected with a hyperplane in a classification
phase as,
w = f (vi ) = βτ v − e
where β denotes the weight vector, τ indicates the transpose operator
and e represents the bias of the hyperplane. The training dataset of SVM
heterogeneous fault classification belongs to one of the two classes of the
target label {-1, 1}. The two classes are bisected through a gap which is
termed a hyperplane. The optimum hyperplane can be structured in
order to make the most efficient heterogeneous fault classifier. An op­
timum hyperplane is acquired through minimization of the soft
parameter (s) given as
where δ denotes the kernel parameter which controls the radius of
the hyperplane.
The merits of the SVM classifier are to offers a global optima and does
not converge into local minima. The performance of the SVM classifier
predominantly depends on selection of kernel function and its parame­
ters (δ, s). However, the standard SVM fails to choose the optimal pa­
rameters that cause the classification errors in the training process. To
mitigate this constraint, a new enhanced SVM called Tuned SVM
(TSVM) classifier has been proposed in which HGOA is utilized to
optimize the parameters (s, δ). This aids to reduce the running time and
the fault classification errors generated by the SVM.
The main phenomenon is to select the HGOA as optimizing algorithm
for TSVM classifier is summarized as follows: (a) HGOA is widely
employed in several optimization problems owing to its simple
deployment and effective accuracy; (b) It has better exploration abili­
ties; (c) It does not necessitate the gradient data of the searching point;
(d) It attains maximum repulsion rate when determining multi-model
and complex test functions.
3.4. Hierarchy based GOA (HGOA)
The traditional GOA is categorized with the population-based met­
aheuristic algorithm which was mimicked from the swarm intelligence
of grasshoppers to solve the optimization constraints [35]. Typically, the
grasshoppers are isolated individually, but they are formed as fasci­
(10) nating groups of creatures. The nature mimicked metaheuristic algo­
rithms logically split the search procedure into two phases namely:
exploration and exploitation. Here, the exploration resembles the adult
swarm behavior whereas the exploitation mimics the larva swarms. The
mathematical model formulated to invoke the swarm intelligence of
grasshoppers is depicted as:
(11)
Yk = Rk + Fk +Wk (16)
th
where Yk intends the position of the k grasshopper, Rk refers to the
social relationship, Fk signifies the gravity force on the kth grasshopper
and the wind advection is indicated as Wk . These three factors will affect
the flying movement of grasshopper in a swarm. The key component of
the traditional GOA is the social relationship between the grasshoppers
(Rk ) and it is expressed as:

7
S. Lavanya et al. Measurement 183 (2021) 109771

⎧ ∑M
⎪
⎨ Rk = r(dkl ) dkl ⎛ ⎛ ⎞ ⎞
l=1,l∕
=k
(17) ⎜ ⎜ ∑ ⎟ ⎟
⎪
⎩ ⎜ ⎜ M (uBd − lBd ) ( ⃒⃒ d ⃒) ⎟
d ⃒ (yl − yk ) ⎟
⎟ Hk
− q
r(q) = ge t − e− q
Ykd = ⎜a ⎜ a r y − y + Z ⎟* (23)
⎜ ⎜
1 2
2 l k
d ⎟ d ⎟ S
⎝ ⎝ l = 1, kl ⎠ ⎠
where r(q)indicates the function that states the social relationship
l∕
=k
strength, M denotes the number of grasshoppers, dkl refers to the dis­
tance between the kth and lth grasshoppers which is expressed as dkl = The overall process of the proposed TH classifier is illustrated in
|yl − yk |, the unit vector from the kth grasshopper to lth grasshopper is Fig. 4. The quality of each grasshopper can be measured using a pre­
represented as dkl and the value of dkl = (yl − yk )
. Further, g intends the defined fitness function. In general, the fitness function is assessed for
dkl
manipulating the finest solution from a set of results where it has been
attraction intensity and the scale of attraction length is denoted as t. The
considered as a maximization value. As a result, the solution that ac­
algorithm achieves better optimization performance by considering f =
quires the higher fitness value is designated as the finest solution.
0.5 and t = 1.5 values during the evaluation process [36].
The fitness function of the proposed Hybrid TH classifier depends on
Let f be the gravitational constant, ef denotes the unity vector
the average classification accuracy of the TSVM method. This can be
directing to the center of the earth and Fk component in Eq. (18) is
trained using the parameters specified in each grasshopper. It is calcu­
computed as:
lated as follows:
Fk = − fef (18)
1 ∑N

Let w be the constant drift, ew intends the unity vector in the direction Fitness = α(c(xi ), yi ) (24)
N
of the wind and Wk component in Eq. (19) is estimated as: i=1

Wk = wew (19) where N indicates the number of occurrences in the testing process,
c(xi )represents the classification output of ith heterogeneous fault
Now substituteRk , Fk , and Wk into Eq. (16) to obtain: occurrence, yi denotes the actual class of ith occurrence, αspecifies the
∑
M relation between c(xi ), and yi . Moreover, α is computed as follows:
(yl − yk )
Yk = r(|yl − yk | ) − fef + wew (20) {
dkl 1c(xi ) = yi ,
l=1
α= (25)
0otherwise.
Nevertheless, the aforesaid mathematical model cannot be employed
directly to resolve the optimization constraints. This is because grass­ In this work, α = 1 constitutes the correct classification of hetero­
hoppers rapidly reach the comfort zone and do not converge to a tar­ geneous faults in IWSN whereas α = 0 designates an incorrect classifi­
geted region. Hence, the modified updated equation is proposed as cation. According to fitness value, every grasshopper attains a hierarchy
follows: in the proposed HGOA. The high attribute grasshoppers with superior
⎛ ⎞ fitness values will have a lesser priority and thus there will be minor
refinement in their locations. This facilitates the proposed HGOA to
⎜ ∑ ⎟
⎜ M (uBd − lBd ) ( ⃒⃒ d ⃒) ⎟
d ⃒ (yl − yk ) ⎟ explore locally and do more exploitation during the search process in the
Ykd = a1 ⎜ a r y − y ⎟ + Zd (21)
⎜ targeted region. This is advantageous when the grasshopper is nearer to
2 l k
⎝ l = 1, 2 dkl ⎠
the optimum result because it will maximize the feasibility to attain
l∕
=k
global optima.
where lBd and uBd indicate the lower and upper limit values of dth On contrary, the low attribute grasshoppers with lesser fitness values
dimension respectively, Zd refers to the optimal position after every will be allocated higher priority. It directs them to intensely varying
iteration, the parameters a1 and a2 are key controlling factors which their locations and explore globally to search more areas in the targeted
linearly reduces during the iteration process and it is evaluated as region. Consequently, the proposed HGOA balances the exploitation and
follows: exploration in searching space owing to the incorporation of a hierarchy
strategy. This stimulates the HGOA to obtain the optimal values for
amax − amin
a1 = a2 = amax − v (22) TSVM parameters (s, δ) during the training process.
V
where amax and amin signify the maximum and minimum search ca­ 4. Performance evaluation
pabilities respectively, v denotes the current iteration, V intends the
maximum number of iterations. According to [36], amax = 1 and amin = The performance of the proposed E2HFM scheme was assessed and
0.00001 values are considered in this work. compared with three candidate schemes such as HFD [26], INSA [17],
Most of the existing works have proven that the traditional GOA is and FDRFC [23]. To ensure that the comparisons were fair, all the above
highly effective in identifying the optimum solutions for various chal­ schemes employed the same distributed methodology to identify the
lenging optimization problems. However, the performance of traditional fault diagnosis of scattered sensor nodes.
GOA extremely depends on accomplishing a balance between exploita­ Each existing schemes have unique characteristics to identify the
tion and exploration. Furthermore, the fitness value of a grasshopper occurrence of one or two faults in the network. Further, these schemes
cannot be taken into account during the position updating of a grass­ are employed under the condition of the selection of the same training
hopper. This illustrates that low-quality grasshopper will vary their parameter and utilization of the same generated fault dataset. Among
position as high-quality results in the search region. Thereupon, it de­ them, the common parameters of every scheme are set based on the
grades the classification performance of the traditional GOA. corresponding literature. To validate the robustness of the performance
To mitigate this problem, a Hierarchy based GOA (HGOA) has been results and minimize their randomness, under the same conditions, the
proposed in which the position of a grasshopper is updated towards the simulation scenario can be run 50 times and thus, the average value is
finest global solution based on the searching hierarchy. The funda­ taken for different performance metrics. Moreover, the HFD scheme was
mental update process of a grasshopper in the traditional GOA is framed manipulated with hardware and time-based faults whereas the INSA and
in Eq. (21) which will be reformed into a new update process as FDRFC schemes addressed the time-based and software faults respec­
demonstrated in Eq. (23). Here, the newly added hierarchy term is HSk tively. Nevertheless, most of the existing schemes do not focus on
where Hk signifies the hierarchy of the kth grasshopper in the swarm and simulating these three faults simultaneously. The best fault detection
the size of the swarm is denoted as S. scheme for each fault has been anticipated for the comparison process.

8
S. Lavanya et al. Measurement 183 (2021) 109771

Fig. 4. Schematic diagram of the proposed TH classifier.

Therefore, the proposed scheme was compared with HFD, INSA, and
Table 2
FDRFC schemes.
Simulation setting parameters.
To prove the robustness of the proposed fault diagnosis scheme, the
Intel Lab dataset [37] is utilized in this experiment where the sensor data Type Parameters Value
is recorded between 28th February 2004 and 5th April 2004. Here, 54 Network Topology Network size 1000 m × 1000 m
sensors are distributed in the Intel Berkeley Research lab to gather the Number of nodes 100, 200, 300, 400, 500
sensor data. The gathered data consist of temperature, humidity, and Node deployment Random
Transmission range 150 m
light. Every sensor gathers one data every 31 s for a total of 2.3 million Antenna model Omni-directional antenna
records. Afterward, different types of heterogeneous faults were injected Propagation model Two-Ray Ground
randomly under different probability rates which are described as fol­ Traffic Range Constant Bit Rate
lows: The temperature data of all sensors are preferred between 1st No. of Heterogeneous 3
Faults
March and 7th March 2004 where the first sensor data is desired from
Sink position (X,Y) (500 m , 500 m),(1000 m , 500
1st March for the fault injection process. The time interval of the sam­ m)
pling is 10 min and 400 samples were engendered continuously. These Sensing range 5m
generated samples are characterized as: the samples of 1–100 are normal Simulation time 200 s
data, 101–200 are hardware fault samples, 201–300 are software fault Channel Wireless
Channel type Bidirectional
samples, and 301–400 are time-based fault samples. Finally, the corre­ Initial energy 5J
lated heterogeneous fault category tags for the distinct samples are Hybrid TH amax 1
aggregated and thus, the fault dataset is generated. Afterward, the classifier amin 0.00001
performance of all schemes has been assessed using the generated fault Number of search agents 50
dataset. Number of iterations 100
The performance of the proposed as well as existing schemes (HFD, σ 0.001
INSA, and FDRFC) have been evaluated using MATLAB (2017b) envi­
ronment. The detailed simulation settings are enumerated in Table 2.
simulation cases: IWSN#1 and IWSN#2. The sink was located at the
The IWSN simulation setting encompasses multiple sensor nodes and
center point (500, 500) in Case 1 whereas, the sink was positioned at
one sink node. These sensor nodes are randomly scattered within the
(1000, 500) in Case 2. In most of the previous studies, simulation en­
sensing field of 1000 m × 1000 m. A variety of heterogeneous faults were
vironments are accomplished in IWSN#1 case only but they do not
appended into the IWSN under two different cases. Every distributed
focused on IWSN#2 simulation. To resolve this, the proposed research
node is connected with single-hop neighbors within the communication
can be implemented in a new simulation case (IWSN#2) along with
range. The battery energy of all nodes is 5 J. The energy model is
IWSN#1. The main intention of examining the IWSN#2 environment is
operated with 20 and 50 bytes for control and data packets respectively.
to test the performance of proposed and existing schemes under different
An assumption has been made for the simulation that no collisions arise
sink locations. These two different sink sites are shown in Fig. 5 (a) and 5
in data communications during the initial state. Every simulation has
(c).
been performed ten times and thus the average value was calculated. All
From Fig. 5, the green color nodes represent the randomly scattered
the experiments were executed in a static environment, i.e., the sink and
nodes with fault-free circumstances and the sink is signified by the pink
sensor nodes were stable.
triangle. After applying the proposed E2HFM scheme, the various faults
Another key contribution of this research is to examine the effec­
are detected in the sensing field which is indicated as red color nodes
tiveness of the proposed as well as existing schemes under two different
(Fig. 5b and 5d). The proposed E2HFM scheme assesses the various types

9
S. Lavanya et al.
Fig. 5. (a) Early network condition of IWSN#1, (b) Detection of faulty nodes in IWSN#1 (c) Initial network circumstance of IWSN#2, (d) After fault diagnosis
in IWSN#2.
of heterogeneous faults such as hardware, software, and time-based
faulty nodes together in IWSN.
5. Simulation results and discussion
The proposed E2HFM scheme is assessed as well as compared with
other existing state-of-art schemes through several performance metrics.
These metrics include Detection Accuracy Rate, False Alarm Rate, Network
Stability Rate (NSR), Network Coverage Ratio, Diagnosis Latency Rate,
Network Lifetime, and Average Energy Consumption. Here, NSR is the novel
metric to demonstrate the stability and reliability of the proposed
scheme.
10
Measurement 183 (2021) 109771
5.1. Impact of detection accuracy rate
The Detection Accuracy Rate (DAR) [38] is computed by
ND
DAR(%) = x100 (26)
NA
where ND indicates a set of faulty sensor nodes diagnosed through
implemented schemes and NA signifies a set of original failure nodes.
Fig. 6a and 6b expose the DAR value for case 1 and case 2 respec­
tively. These results manifest that the DAR of the proposed E2HFM is
better as compared with existing schemes for both IWSN#1 and
IWSN#2 cases. Especially, when the fault probability rate exceeds 35%,
the DAR of the FDRFC, INSA, and HFD schemes reduces linearly, while
S. Lavanya et al. Measurement 183 (2021) 109771

5.2. Impact of false Alarm rate

The False Alarm Rate (FAR) [30] is calculated by

NF
FAR(%) = x100 (27)
NN
where NF indicates a set of non-faulty nodes diagnosed as damaged
nodes and NN signifies a set of non-faulty nodes.
In Fig. 7, while the node has a lesser fault probability in the network,
the initial FAR of three existing schemes is minimum. But, they depict
the higher FAR value when the iteration process increases. Alterna­
tively, the proposed scheme exposed better performance than existing
schemes for a higher fault probability rate. The proposed E2HFM still
remains 0.012 FAR at 40% fault probability, whereas the FDRFC, INSA,
and HFD are having 0.053, 0.043, and 0.028 FAR respectively. It also has
fewer FAR values in IWSN#1 than IWSN#2 environment.
The key reason behind this minimum FAR is the execution of a
proper heterogeneous fault classification model (Hybrid TH classifier) in
the proposed E2HFM. The optimum tuning parameter selection from the

Fig. 6. Comparison of DAR under different fault probability rates in (a)

IWSN#1 (b) IWSN#2.

the DAR of the proposed E2HFM scheme still exceeds above 99%. This
can be attributed to monitoring the hardware, software, and time-based
fault status simultaneously of the deployed nodes in IWSN. These
different types of faults are classified using the multi-class TSVM
method. The detection algorithm in the proposed E2HFM scheme assists
in identifying the heterogeneous faults rapidly. It stimulates the pro­
posed E2HFM to detect all types of faulty nodes significantly in the
network. In contrast, the DAR of the FDRFC scheme attains lower DAR
because it entirely depends on the training dataset. It does not detect the
faults induced in real-time readings.
In the case of INSA, the neighbor nodes are recognized as failure
node occasionally. This will direct an unreliable decision on the
currently considered node in IWSN. Unlike the two schemes, the HFD
scheme detects the hardware and time-based faults of the scattered
nodes. Subsequently, the software failures are not identified which leads
to degrading the performance of HFD than the proposed E2HFM scheme.
Table 3 depicts the DAR of different fault types of the proposed E2HFM
scheme. While considering DAR, the proposed E2HFM outperformed Fig. 7. Analysis of FAR values at varying fault probability rates in (a) IWSN#1
FDRFC by 17%, similarly 13% for INSA and 8% for HFD. (b) IWSN#2.

Table 3
DAR value of proposed E2HFM scheme under different fault types.
Cases Fault types Fault Probability Rate (%)
5 10 15 20 25 30 35 40

IWSN#1 Hardware 100% 100% 100% 100% 99.8% 98% 99% 99%
Software 100% 100% 100% 100% 99.8% 97% 97% 98%
Time-based 100% 100% 100% 100% 100% 97.5% 98% 100%
IWSN#2 Hardware 100% 100% 100% 100% 99.8% 97% 98% 98.3%
Software 100% 100% 100% 99.7% 99.7% 97% 97% 97.8%
Time-based 100% 100% 100% 100% 100% 98% 99% 99%

11
S. Lavanya et al.
HGOA optimization algorithm yields lesser fault classification errors in
the TSVM classifier. In addition, the HGOA balances the exploitation and
exploration in searching space owing to the incorporation of a new hi­
erarchy strategy. It allows the searching agent to explore the entire re­
gions for attaining optimal solutions. Conversely, the FAR value is
maximum for the FDRFC scheme which is closely followed by INSA and
HFD schemes. The improper selection of optimization algorithm in
existing schemes causes higher classification errors in the network.
5.3. Impact of network stability rate
In this study, the NSR has been assessed based on operating status
and estimating detection time. The operating status incorporates three
statuses such as status 1, status 0.5, and status 0. Status 1 implies that the
network works consistently without any damaged node, whereas status
0.5 symbolizes the identification of the damaged node. The damaged
node that appeared in the network is signified by status 0. Then, the
detection time (DT ) is calculated by
DT = Td − Tf (28)
where Td represents the time at which the damaged node is detected
and Tf intends the time at damaged node appears. Normally, the scheme
has lesser DT provides higher NSR in IWSN.
From Fig. 8, it is noticed that the proposed E2HFM scheme has lesser
DT than FDRFC, INSA, and HFD schemes. Particularly, the first DT of
E2HFM, HFD, INSA, and FDRFC are 1 s, 3 s, 3 s, and 5 s respectively.
These statistical results confirm that the proposed E2HFM can sustain the
network for an extensive time period. This is because of the imple­
mentation of distributed heterogeneous fault model in the proposed
scheme to maintain a fault-free environment during data routing.
The NSR metric of the proposed scheme increases while moving to
Fig. 8. Comparison of NSR in (a) IWSN#1 (b) IWSN#2.
12
Measurement 183 (2021) 109771
the dense network environment. Besides, these stability improvements
are acquired through tuning the parameters of s and δin the Hybrid TH
classifier. On contrary, the HFD and INSA have a better diagnosis model,
however, they failed to establish the appropriate classifier model. The
SVM parameters are also not optimized in the INSA scheme which elu­
cidates lower NSR in the IWSN. Subsequently, FDRFC does not diagnose
as well as classify the faulty nodes effectively as compared with other
schemes.
Table 4 illustrates the results of tuned parameters for the different
classifiers. It is clearly observed in Table 4 that the proposed TH clas­
sifier has obtained superior tuned parameter results as compared with
existing classifiers. These significant improvements are owing to the
incorporation of the HGOA algorithm in the proposed TH classifier. So,
the proposed E2HFM scheme identifies heterogeneous faults accurately
and improves the network stability in IWSN.
5.4. Impact of network coverage ratio
The Network Coverage Ratio (NCR) [38] is determined by
Ct
NCR(%) = x100 (29)
Cmax
where Ct intends the total area covered by the deployed node and
Cmax denotes the maximum area covered by the deployed node.
The comparison of NCR for different schemes is demonstrated in
Table 5. The observance of the proposed E2HFM attains 37% better NCR
over FDRFC in IWSN#1 setting, 23% over INSA, and 12% over HFD for
varying node density in the sensing field. These statistics demonstrate
that when the number of scattered nodes increases, the performance of
the E2HFM scheme also increases linearly. In particular, the E2HFM still
remains 96% NCR for the 500 nodes environment, whereas the FDRFC,
INSA, and HFD are having 62%, 70%, and 76% NCR respectively. These
outcomes are due to the determination of the optimal fitness function in
E2HFM.
The extensive hardware, software, and time-based fault analyses are
accomplished within the heterogeneous fault detection process itself.
This strategy decreases the network partitioning issues throughout the
whole network. Moreover, the coverage sustain in E2HFM facilitates to
obtain the maximum NCR and increases the packet delivery ratio in
IWSN. Instead, unreliable connectivity within the sensing region stim­
ulates frequent network partitioning problems in FDRFC, INSA, and HFD
schemes. They met the challenge of deliberate classification in a higher
fault probability circumstance. As a consequence, they attain fewer NCR
values than the E2HFM scheme. In the same manner, the NCR of
IWSN#1 circumstance is better than the IWSN#2 because of the center
location of the sink node which monitors all sensing regions in the IWSN.
This center location guarantees the detection of sensing points more
accurately and conveyed to the central sink. The capability of network
coverage is poor in the IWSN#2 environment that reduces both NCR
value and the number of data packets received at the sink node.
Table 4
Tuned parameters results of different classifier.
Classifier Parameters Radial basis function
Proposed TH s 415
δ 1.17
GWO-SVM s 406
δ 1.13
DE-SVM s 405
δ 1.12
GA-SVM s 402
δ 1.12
Traditional SVM s 298
δ 1
S. Lavanya et al. Measurement 183 (2021) 109771

Table 5
Comparison of NCR for different schemes.
Number of deployed nodes NCR (%)
IWSN#1 IWSN#2
FDRFC INSA HFD Proposed E2HFM FDRFC INSA HFD Proposed E2HFM

100 34 41 51 58 33 39 48 56
200 43 49 56 67 42 46 52 63
300 49 55 63 75 50 51 59 73
400 56 63 69 84 54 60 64 81
500 62 70 76 96 58 67 72 94

5.5. Impact of diagnosis Latency rate
The total time period involved for diagnosing all the failure nodes in
the IWSN is termed as Diagnosis Latency Rate (DLR). Fig. 9 depict the
DLR analysis for two different cases with regard to the various fault
probability rate. By considering 500 nodes, the percentage of fault
probability rate is increasing from 5% to 40%. The proposed scheme
obtains a lesser DLR value as compared with the existing schemes. When
the fault probability rate increases, the average DLR for the proposed
scheme is 2.22 s, whereas, 2.58 s, 3.23 s, and 3.27 s for INSA, HFD, and
FDRFC schemes respectively.
The rationale behind the DLR reduction is due to fast diagnosis as
well as accurate classification of the heterogeneous faults in the pro­
posed scheme. In the case of FDRFC, every sensor node is exhibited with
the earlier sample of its own data along with neighbor’s data. The faulty
data can be trained and classified using generated RF decision tree. As a
result, the computation of the FDRFC scheme takes more delay than all
existing schemes. Similar to FDRFC, the HFD also requires the learning
data to detect and classify the faults. This neural network technique
increases the DLR value in the HFD scheme. On the other hand, the INSA
scheme follows a voting-based methodology for fault identification.
Every sensor node collects the voting information for classifying the
faults. It creates a higher computational time in the sensing field.
5.6. Impact of network lifetime
The Network Lifetime (NL) has been estimated with the aid of the
first sensor node die metric. It calculates the simulation rounds before
the first node dies in the network. From Table 6, it can be noted that the
NL of the proposed E2HFM is 37%, 23%, and 12% higher than the
FDRFC, INSA, and HFD schemes respectively. So, the proposed E2HFM
scheme manifests better results in terms of NL metric as compared with
existing schemes.
The significant enhancements are owing to the fact that the E2HFM
scheme maintains a larger number of alive nodes in both environment
cases. This enlarges the uninterrupted data packet communication in the
IWSN that facilitates the proposed scheme to achieve a maximum NL of
around 1430 rounds. However, the existing schemes have a lesser
number of alive nodes and it directs to obtain a minimum NL than
E2HFM. Furthermore, the hardware and software fault reveals the dy­
namic behavior in IWSN. Hence, the existing schemes do not properly
detect the faulty nodes during higher fault probability rates. At the same
time, the proposed E2HFM diagnoses these types of heterogeneous faults
simultaneously with the aid of three appropriate diagnosis algorithms.
The energy depletion is uniformly scattered across all the sensor nodes
due to its simultaneous diagnosis methodology. The proposed E2HFM
manipulates one-hop neighbor node information to identify the het­
erogeneous faults which significantly extends NL of IWSN.

5.7. Impact of average energy consumption

The Average Energy Consumption (AEC) of various schemes is

Fig. 9. Comparison of DLR under different fault probability rates in (a)
IWSN#1 (b) IWSN#2.
illustrated in Fig. 10. The results illustrate that the proposed E2HFM
reduces AEC up to 44% than FDRFC, up to 37% than INSA, and up to
33% over HFD for varying node density in the sensing field. The varia­
tion in AEC is higher for the FDRFC scheme.
The AEC of the INSA scheme matches with the HFD scheme till 300
node density. Thereupon, the gradual increase in the node density, it
becomes slightly larger than that of the HFD. The lowest value of AEC is
0.0164 J which is acquired through E2HFM at IWSN#1 scenario. It is
owing to the utilization of the robust fault maintenance strategy and
earlier detection of battery unit failure. The lesser AEC rate confirms that
the proposed technique significantly avoids premature death of the
IWSN. Besides, the consistent routing path is exploited for several
rounds. This elucidates lesser control packet overhead and accordingly
reduces the AEC of nodes, which consequently prolonging the lifetime of
the network.
On contrary, the FDRFC and INSA could not properly detect the
battery unit failure of deployed nodes. This cause premature death of the
nodes along with inconsistent routing path in IWSN. Thus, the above two
schemes attain higher AEC than HFD and E2HFM. The HFD scheme
efficiently diagnoses the battery unit failure and time-based faults.
Nevertheless, the probability of false diagnosis is maximal in HFD that
arises higher AEC than the proposed E2HFM scheme.

13
S. Lavanya et al. Measurement 183 (2021) 109771

Table 6
Comparison of NL for different schemes.
Number of deployed nodes NL (Rounds)
IWSN#1 IWSN#2
FDRFC INSA HFD Proposed E2HFM FDRFC INSA HFD Proposed E2HFM

100 1089 1198 1307 1430 1078 1191 1304 1429

200 1023 1165 1276 1400 1020 1160 1270 1395
300 997 1092 1210 1387 985 1089 1203 1379
400 956 1050 1167 1310 952 1046 1160 1299
500 877 989 1102 1250 862 976 1090 1237

for the dense network (500 nodes). It is manifest from Table 7 that the
performance of the proposed scheme is almost similar in both IWSN#1
and IWSN#2 environments. In particular, there are insignificant impacts
on the performance of the proposed scheme owing to change in the sink
position. In the case of NL and DLR, the proposed scheme achieves
slightly better values in IWSN#1 when compared with IWSN#2 envi­
ronment. This is owing to the uniform position of the sink that gua­
rantees all sensing areas are properly monitored and report back to the
central sink. In contrast, fewer sensor nodes in existing schemes deplete
more battery energy when compared with other nodes because the
average distance between the origin nodes and the sink node is higher in
case of the IWSN#2. From the aforesaid analysis, it can be concluded
that the position of the sink has an insignificant impact on the proposed
scheme whereas it shows notable changes in the case of existing
schemes.
Based on Table 7, it can be concluded that: (a) The overall perfor­
mance of the proposed E2HFM is far better than existing schemes owing
to the implementation of effective fault diagnosis and classification
methods. (b) The stability of E2HFM has been sustained constantly while
increasing the number of damaged nodes. (c) In the proposed scheme,
InfoHFD message-based fault diagnosis model is facilitated to increase
DAR and reduce FAR significantly. Further, the higher value of DAR in
the proposed method has proven that it is the most adopted technique
for IoT-based applications. (d) The proposed scheme achieves better
network lifespan and energy efficiency due to earlier detection of faulty
nodes in the network.

6. Conclusion

Owing to the limited battery resources of sensor nodes and diverse

deployment environments, fault diagnosis in IWSN has become a more
challenging task. Almost no existing fault management scheme has been
Fig. 10. Analysis of AEC with respect to varying network density in (a)
established to identify and classify the different types of heterogeneous
IWSN#1 (b) IWSN#2.
faults in the IWSN. But, the proposed E2HFM scheme considered all the
heterogeneous faults (hardware, software, and time-based faults) during
The overall energy consumption of the network depends upon the
the fault diagnosis stage itself. The initial step of the proposed E2HFM
total number of data transferred and received by distributed nodes
scheme is to identify the heterogeneous faults of the deployed sensor
during the fault detection stage. This metric facilitates the investigation
nodes in IWSN. Subsequently, the Hybrid TH classifier has been
of the energy balance among the scattered nodes in the network. The
exploited to classify the faults into their corresponding types. The HGOA
energy balance of the overall network outperforms well in the proposed
in Hybrid TH classifier balances the exploitation and exploration in
E2HFM scheme whereas the existing schemes provoke unconventional
searching space which stimulates the HGOA to obtain the optimal values
energy balance in the network.
for TSVM parameters (s, δ) during the training process.
Table 7 illustrates the performance comparison of various schemes
The simulation results are assessed in the MATLAB platform where a

Table 7
Performance comparison of various schemes for the dense network.
Metrics IWSN#1 IWSN#2
FDRFC INSA HFD ProposedE2HFM FDRFC INSA HFD ProposedE2HFM

DAR(%) 69 72 81 99 67 69 80 98
FAR 0.053 0.043 0.028 0.012 0.058 0.043 0.034 0.013
NSR [DT ](s) 5 3 3 1 3 3 1 1
NCR(%) 62 70 76 96 58 67 72 94
DLR(s) 5.5 4.1 5.8 3.3 5.8 4.7 6.0 4.0
NL(rounds) 877 989 1102 1250 862 976 1090 1237
AEC(J) 0.19 0.16 0.14 0.10 0.19 0.17 0.16 0.12

14
S. Lavanya et al.
set of experiments are evaluated under two different IWSN scenarios.
The obtained results manifest that the proposed E2HFM scheme out­
performs the FDRFC, INSA, and HFD schemes for large-scale IWSN.
Particularly, the results of the NSR metric proven that the proposed
E2HFM scheme perpetuates the network for an extensive period. The NL
of E2HFM has enhanced by 37%, 23%, and 12% when compared with
FDRFC, INSA, and HFD schemes respectively. Furthermore, the E2HFM
still sustains with 0.1 J AEC, 96% NCR, and 99% DAR for dense IWSN
circumstances. The AEC results validate that it depletes lesser battery
energy for the diagnosis as well as classification stages. Henceforth, it
can be concluded that the proposed E2HFM scheme prolonging the
network stability by the implementation of an effective fault manage­
ment strategy in IWSN.
The proposed E2HFM scheme does not address the recovery mech­
anism and mobility approach in IWSN. In future work, it has to be
extended and incorporated the mobile node along with static node en­
vironments. It also investigates some novel recovery mechanisms to
enhance the packet delivery ratio in IWSN.
CRediT authorship contribution statement
S. Lavanya: Conceptualization, Methodology, Visualization. A.
Prasanth: Software, Supervision, Data curation, Writing - original draft.
S. Jayachitra: Investigation, Software, Writing - review & editing. A.
Shenbagarajan: Formal analysis, Resources, Validation.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
References
[1] P. Nayak, G.K. Swetha, S. Gupta, K. Madhavi, Routing in Wireless Sensor Networks
Using Machine Learning Techniques: Challenges and Opportunities, Measurement
170 (108974) (2021) 1–18.
[2] M. Ammar, G. Russello, B. Crispo, Internet of Things: A survey on the security of
IoT frameworks, Journal of Information Security and Applications 38 (2018) 8–27.
[3] S.li, Z.liu, Z.huang, DynaPro: Dynamic Wireless Sensor Network Data Protection
Algorithm in IoT via Differential Privacy, IEEE Special Section on Data Mining for
Internet of Things, IEEE Access, 7 (2019) 167754 – 167765.
[4] A. Prasanth, S. Pavalarajan, Zone-based sink mobility in wireless sensor networks,
Sensor Review 39 (2019) 874–880.
[5] G. Kaur, P. Chanak, M. Bhattacharya, Memetic Algorithm-Based Data Gathering
Scheme for IoT-Enabled Wireless Sensor Networks, IEEE Sens. J. 20 (2020)
11725–11734.
[6] Z. Ding, H. Chen, N. Ansari, Energy-Efficient Relay-Selection-Based Dynamic
Routing Algorithm for IoT-Oriented Software-Defined WSNs, IEEE Internet Things
J. 7 (2020) 9050–9065.
[7] A. Prasanth, S. Jayachitra, A Novel Multi-Objective Optimization Strategy for
Enhancing Quality of Service in IoT enabled WSN Applications, Peer-to-Peer
Networking and Applications 13 (2020) 1905–1920.
[8] P. Chanak, I. Banerjee, Congestion Free Routing Mechanism for IoT-Enabled
Wireless Sensor Networks for Smart Healthcare Applications, IEEE Trans. Consum.
Electron. 66 (2020) 223–232.
[9] A. Prasanth, S. Pavalarajan, Implementation of Efficient Intra-and Inter-Zone
Routing for Extending Network Consistency in Wireless Sensor Network, Journal of
Circuits, Systems and Computers 29 (2020) 1–25.
[10] K. Karunanithy, B. Velusamy, Energy efficient cluster and travelling salesman
problem based data collection using WSNs for Intelligent water irrigation and
fertigation, Measurement 161 (107835) (2020) 1–21.
[11] A.Prasanth, S.Pavalarajan Particle Swarm Optimization Algorithm Based Zone
Head Selection in Wireless Sensor Networks, International Journal of Scientific &
Technology Research, 8 (2019) 1594-1597.
[12] M.R. Skydt, M. Bang, H.R. Shaker, A probabilistic sequence classification approach
for early fault prediction in distribution grids using long short-term memory neural
networks, Measurement 170 (108691) (2021) 1–17.
[13] E. Moridi, M. Haghparast, M. Hosseinzadeh, S.J. Jassbi, Fault management
frameworks in wireless sensor networks: A survey, Comput. Commun. 155 (2020)
205–226.
[14] S. Zidi, T. Moulahi, B. Alaya, Fault Detection in Wireless Sensor Networks through
SVM Classifier, IEEE Sens. J. 18 (2018) 340–347.
[15] S. Mitra, A. Das, Distributed Fault Tolerant Architecture for Wireless Sensor
Network, Informatica 41 (2017) 47–58.
15
Measurement 183 (2021) 109771
[16] T. Muhammed, R.A. Shaikh, An analysis of fault detection strategies in wireless
sensor networks, Journal of Network and Computer Applications 78 (2017)
267–287.
[17] S. Mohapatra, P.M. Khilar, Fault diagnosis in wireless sensor network using
negative selection algorithm and support vector machine, Comput. Intell. 36
(2020) 1374–1393.
[18] V.K. Menaria, S.C. Jain, N. Raju, R. Kumari, NLFFT: A Novel Fault Tolerance Model
Using Artificial Intelligence to Improve Performance in Wireless Sensor Networks,
IEEE Access 8 (2020) 149231–149254.
[19] M. Masdari, S. Ozdemir, Towards coverage-aware fuzzy logic-based faulty node
detection in heterogeneous wireless sensor networks, Wireless Personal
Communication 111 (2020) 581–610.
[20] W. Gui, Q. Lu, M. Su, F. Pan, Wireless Sensor Network Fault Sensor Recognition
Algorithm Based on MM* Diagnostic Model, IEEE Access 8 (2020)
127084–127093.
[21] Y. Gao, F. Xiao, J. Liu, Distributed Soft Fault Detection for Interval Type-2 Fuzzy-
Model-Based Stochastic Systems with Wireless Sensor Networks, IEEE Trans. Ind.
Inf. 15 (2019) 334–347.
[22] L. Rui, X. Wang, Y. Zhang, X. Wang, A self-adaptive and fault-tolerant routing
algorithm for wireless sensor networks in Microgrids, Future Generation Computer
Systems 100 (2019) 35–45.
[23] Z. Noshad, N. Javaid, T. Saba, Z. Wadud, Fault Detection in Wireless Sensor
Networks through the Random Forest Classifier, Sensors 19 (1568) (2019) 1–21.
[24] N. Moussa, A. Alaoui, A cluster-based fault-tolerant routing protocol for wireless
sensor networks, International Journal of Communication System 32 (e4131)
(2019) 1–17.
[25] R.R. Swain, P.M. Khilar, T. Dash, Neural network based automated detection of
link failures in wireless sensor networks and extension to a study on the detection
of disjoint nodes, J. Ambient Intell. Hum. Comput. 10 (2019) 593–610.
[26] R.R. Swain, P.M. Khilar, S.K. Bhoi, Heterogeneous Fault Diagnosis for Wireless
Sensor Networks, Ad Hoc Netw. 69 (2018) 15–37.
[27] S. Acharya, C.R. Tripathy, An ANFIS estimator based data aggregation scheme for
fault tolerant Wireless Sensor Networks, Journal of King Saud University –
Computer and Information Sciences 30 (2018) 334–348.
[28] Y. Cheng, Q. Liu, J. Wang, Distributed Fault Detection for Wireless Sensor
Networks Based on Support Vector Regression, Wireless Communications and
Mobile Computing 2018 (4349795) (2018) 1–8.
[29] O. Kaiwartya, A. Abdullah, Y. Cao, J. Lloret, Virtualization in Wireless Sensor
Networks: Fault Tolerant Embedding for Internet of Things, IEEE Internet Things J.
5 (2018) 571–580.
[30] J. Gu, Z. Wang, J. Kuen, L. Ma, Recent advances in convolutional neural networks,
Pattern Recogn. 77 (2018) 354–377.
[31] X. Wu, H. Chen, Y. Wang, L. Shu, BP neural network based continuous objects
distribution detection in WSNs, Wireless Network 22 (2016) 1917–1929.
[32] R.R. Swain, P.M. Khilar, Composite fault diagnosis in wireless sensor networks
using neural networks, Wireless Pers. Commun. 95 (2017) 2507–2548.
[33] K. Rajeswari, S. Neduncheliyan, Genetic algorithm based fault tolerant clustering
in wireless sensor network, IET Commun. 11 (2017) 1927–1932.
[34] P. Ray, D.P. Mishra, Support vector machine based fault classification and location
of a long transmission line, Engineering Science and Technology, an International
Journal 19 (2016) 1368–1380.
[35] S. Saremi, S. Mirjalili, A. Lewis, Grasshopper Optimisation Algorithm: Theory and
application, Adv. Eng. Softw. 105 (2017) 30–47.
[36] S.S. Guo, J.S. Wang, W. Xie, Improved Grasshopper Algorithm Based on Gravity
Search Operator and Pigeon Colony Landmark Operator, IEEE Access 8 (2020)
22203–22224.
[37] Sensor Data from Intel Berkeley Research Lab. Available online: https://www.
kaggle.com/divyansh22/intel-berkeley-research-lab-sensor-data/version/1.
[38] A. Prasanth, Certain Investigations on Energy-Efficient Fault Detection and
Recovery Management in Underwater Wireless Sensor Networks, Journal of
Circuits, Systems, and Computers 30 (2150137) (2020) 1–23.
Dr.S.LAVANYA received Bachelor of Engineering degree in Computer Science and Engi­
neering from Mahendra Engineering College, Chennai, Tamilnadu, India in 1999, Master
of Engineering degree in Computer Science and Engineering from Muthayammal Engi­
neering College, Affiliated to Anna University, Chennai, Tamilnadu, India in 2008, Doctor
of Philosophy degree in Information Science and Engineering from Anna University,
Chennai, India in 2016. Currently, she is working as Associate Professor in the Department
of Computer Science and Engineering at Muthayammal Engineering College, Affiliated to
Anna University, Chennai, Tamilnadu, India. Her research interests include Wireless
sensor Networks, Mobile Computing, Medical Image Processing, Soft Computing and
Internet of Things.
Dr.A.PRASANTH received the B.E. degree in Electronics and Communication Engineering
from Anna University, Chennai and the M.E degree in Computer Science and Engineering
(with specialization in Networks) from Anna University, Chennai and also received the Ph.
D degree in Information and Communication Engineering from Anna University, Chennai,
India. He is currently working as an Assistant Professor at Sri Venkateswara College of
Engineering, Sriperumpudur, Tamilnadu, India. He has published more than 25 research
articles in International Journals/Conference proceedings. Moreover, he has published 8
books under reputed publisher. His research interests include Internet of Things, Wireless
Sensor Networks, Blockchain, Medical Image Processing, and Computer Networks.
Ms.S.JAYACHITRA received the B.E. degree in Electronics and Communication Engi­
neering from Anna University, Chennai and M.E Degree in Communication System from
S. Lavanya et al.
Anna University, Chennai. She is currently working as an Assistant Professor at Karpagam
Academy of Higher Education, Coimbatore, India. She has published more than 10
research articles in International Journals/Conference proceedings. She has got 7 years of
teaching experience and her research interests include Image Processing, Signal Processing
and Wireless Sensor Networks.
Dr.A.SHENBAGARAJAN received Bachelor of Engineering degree in Electrical and
Electronics Engineering from P.S.R. Engineering College, affiliated to Anna University,
16
Measurement 183 (2021) 109771
Chennai, Tamilnadu, India in 2005, Master of Engineering degree in Computer Science and
Engineering from National Engineering College, Affiliated to Anna University, Chennai,
Tamilnadu, India in 2008, Doctor of Philosophy degree in Computer Science and Engi­
neering from Annamalai University, Tamilnadu, India in 2018. Currently he is working as
Assistant Professor in the Department of Computer Science and Engineering at AAA Col­
lege of Engineering and Technology, Affiliated to Anna University, Chennai, Tamilnadu,
India. His research interests include Medical Image Processing, Soft Computing and
Internet of Things.