FAULT DETECTION ON THE EDGE AND ADAPTIVE COMMUNICATION FOR

STATE OF ALERT IN INDUSTRIAL INTERNET OF THINGS

Industrial production and manufacturing systems require automation, reliability, as well

as low-latency intelligent control. Industrial Internet of Things (IIoT) is an emerging paradigm
that enables precise, low latency, intelligent computing, supported by cutting-edge technology
such as edge computing and machine learning. IIoT provides some of the essential building
blocks to drive manufacturing systems to the next level of productivity, efficiency, and safety.
Hardware failures and faults in IIoT are critical challenges to be faced. These anomalies can
cause accidents and financial loss, affect productivity, and mobilize staff by producing false
alarms. In this context, this article proposes a framework called Detection and Alert State for
Industrial Internet of Things Faults (DASIF). The DASIF framework applies edge computing to
execute highly precise and low latency machine learning models to detect industrial IoT faults
and autonomously enforce an adaptive communication policy, triggering a state of alert in case
of fault detection. The state of alert is a pre-stage countermeasure where the network increases
communication reliability by using data replication combined with multiple-path
communication. When the system is under alert, it can process a fine-grained inspection of the
data for efficient decison-making. DASIF performance was obtained considering a simulation of
the IIoT network and a real petrochemical dataset.
 CHAPTER 1

1. INTRODUCTION
Three industrial revolutions have taken place previously because of electricity, &
digitalization improvements.At present, industry 4.0 is dominated by the Internet of Things (IoT)
and 5G networks [1,2]. According to the McKinsey Global Institute, 5% of the more than 2000
occupational activities they examined could be completely automated, & 60% of those could be
automated to some extent [3]. The IoT, 5G/B5G, &Machine Learning (ML) were designed in a
symbiotic manner and also have the technology to execute mentally and physically industry
tasks, increase the capacity, effectiveness, & security of manufacturing processes, reduced costs
& mistakes, & enhance performance and effectiveness [4]. One way that IoT would benefit smart
manufacturing would be growth in recent years through monitoring and implementation. By
managing and monitoring products and machinery, IoT would improve supply chain operations
in manufacturing industries. This would rapidly identify defects that could compromise the
production process. IoT has been used in the coal industry toenhance the establishment,
management, traceability, and accessibility of mining operations [5].In contrast, 5G networks
would offer improved large-scale communications, ultra-reliable, low-latency interaction, and a
mobile broadband communications infrastructure to enable a variety of smart manufacturing
floor scenarios [6]. On factory floors, several Fieldbus & industry Ethernet types of technologies
are being utilized to build islands of linked industrial machines, making it difficult for data to
move between gadgets on various islands [7].Private or local 5G networks are used shortly to
remain connected to manufacturing equipment and reduce the need for cable communications
technology. This would lower maintenance costs & limit people's exposure to potentially
dangerous situations, enhance data flow & adjustability when trying to connect industrial
machinery support smartphones such as smartphone robotics and autonomous-focused
automobiles [[8], [9], [10]].

A key barrier to the use of ML in IIoT was the lack of labeled training data.However, many
manufacturing machines contain built-in detectors for formally regular, and real-time
surveillance, effective and high information from industrial areas is complicated and expensive
to collect [11]. For security and efficiency reasons, IIoT would rely primarily on private
deployments of automated technologies and telecommunications [12]. As a result, information
from certain deployments was perhaps not generally available.Additionally, ML models perform
best when training and validation information come from the same allocation & feature space,
which may not be the circumstance in IIoT applications due to unique deployments of smart
manufacturing, utilized manufacturing equipment, geographic locations, & organization
regulations and processes [13]. This would be true even though many commercial uses hold
common systems and tasks.

Therefore, the characteristics to be followed could be influenced by data analysis methods

used for multi-dimensional data, whether in geographic, frequency domains, or temporal.
Whenever we work with large-scale information, ML as one of the paths to real AI is crucial
[14]. Fault diagnosis has been one of the special goals of predictive maintenance, particularly in
Industry 4.0, because of the significant influence that defects have on the effectiveness and
productivity of industrial facilities. By using classifiers based on ML algorithms, the detection
efficiency of fault diagnosis systems can be improved [15]. Using the information provided by a
collection of dispersed devices, they could be trained to identify when a problem exists or even
to identify the specific type of fault that would be hurting the plant. A common strategy would be
to move the reasoning step to the peripherals and position the learning phase in the cloud. Key
benefits of quasi-detection decision-making include reducing transaction costs and response
times and improving the privacy framework [16].The commercial domain benefits from all these
computational power paradigm representational properties. Specifically, accelerating the
detection of outliers can help prevent dangerous conditions that would otherwise be unavoidable
[17]. The development of specialized local AI devices such as Google Coral, which specializes
in DL has dramatically expanded the possibilities of onboard computing in recent years. For this
reason, direct integration of LE into IoT infrastructure is currently considered to be a viable
option. In the context of IoT, border nodes can be characterized as integrated connected systems
that have sensor and compute components and communication networks, usually wireless and
based on the Internet [18].

The processor at the edges enables the manufacturer to distribute choice authority so that
clusters of nearby sensors could be collected to infer a characteristic remotely to use a learned
ML model. When localized but centralized technologies have been used, spreading out would
also reduce the costs of factory communications and the latency caused. To transfer the fault
diagnosis systems into Industrial 4.0, a new framework hybrid approach was proposed in this
study [19]. It must be dependent on the deployment of multiple Machine Learning models
through the plants, one for every node. According to experimental results, this framework offers
lightweight ML solutions based on reduced-feature ML models that meet the highest standards in
the field of detection quality. This structure can create a fault detection infrastructure that would
be scalable and uses networked sensor sets that seem to be qualified to identify various defects
[20,21]. The physical needs of the edge layer processing nodes have been reduced through
decentralization and the use of fewer sensors. Since only one sample was required to generate a
forecast, the solutions provided by this methodology differ in terms of lag [22]. Since the
methods in this article require fewer sensors to detect a problem than the state of the art,
researchers are also reducing interactions in the factory while maintaining accurate predictions.
One way to decentralize a network is to consider using a single class grader to define the
particular problem rather than a global multi-class classification. In addition, compared to multi-
class classifications, fault-specific classifications generally offer a higher detection quality

Machine learning (ML) has been largely agreed upon as a key building block for the
Industrial Internet of Things (IIoT). ML can support intelligent and quick decisions needed in
IIoT environments [1]. Without fast and precise detection and decision-making, manufacturing is
susceptible to all sorts of delays, which can cause financial loss and may represent a security
threat.

In IIoT systems, the benefits of ML are potentiated by the capabilities of edge computing. By
definition, edge computing has emerged as a paradigm to increase the computational power of
IoT systems, providing extra storage and processing [2]. Edge computing devices are deployed
near IoT devices to reduce the time necessary to complete computationally demanding tasks.

Hardware failures and faults in IIoT are critical challenges to be faced. A vast set of
faults can occur in IIoT devices, including drift, spike, stuck, offset, bias, gain, out-of-bounds,
saturation, and precision degradation [3]. All these anomalies can cause accidents, halt
production, and mobilize staff by producing false alarms.

To mitigate the problems caused by IIoT faults, the next generation of IIoT systems
requires: (i) precise and early fault detection; and (ii) autonomic response and reaction. The first
requirement demands constant monitoring of the produced data, and the second, not least
important, consists of autonomously reacting when the fault has been detected. In this context,
modern industrial business models consider that solving a fault in its early stages prevents
collateral damage and results in fewer costs.

Many works in the literature propose solutions to meet the requirements of IIoT. Most of
them are based on ML classifiers to support precise fault detection. Only some works seek to
minimize the response time during the detection. Additionally, a reduced number of works
provides any autonomic reaction for detected faults. To fill this gap, this article proposes a
framework called Detection and Alert State for Industrial Internet-of-Things Faults (DASIF).
The DASIF framework applies highly precise and low latency machine learning models to detect
IIoT faults and also autonomously enforce a state of alert in case of detecting the fault. The faults
are double classified on the edge computing devices using a decision tree and Gaussian naive
Bayes. The state of alert enforced by DASIF is the first-stage countermeasure for fine-grained
data analysis. Under this state, the network decreases the communication interval and uses data
replication and multiple-path communication to achieve higher communication reliability. In our
previous work [4], we conducted a study to select the best ML models for the DASIF framework.
In [4], we evaluated six machine learning classifiers measuring accuracy, precision, recall, F1
score, training time, and response time.

The contributions of this article are the following:

to propose an ML-based framework running on edge computing for detecting Internet of Things
(IoT) faults in industrial environments;

to present a Markov chain-based algorithm to inject a set of faults into IIoT datasets;

to enforce an adaptive communication policy, instituting a state of alert as an autonomic reaction

when a fault has been detected.

High Impedance Faults (HIFs) are undetectable by conventional protection technology

under certain conditions. These faults occur when an energized conductor makes undesired
contact with a quasi-insulating object, such as a tree or a road. Certain features of the
configuration of distribution networks determine the difficulty of the HIF detection; therefore,
the utilities have to evaluate the necessity of using specific HIF protection for each network. The
worst scenario for HIF detection is multiple-grounded distribution system using single-phase
transformers. In this situation, which is very common in America, there is the clear need of using
new and specific HIF detection functions.

A high impedance fault results either from high impedance fault object or when a primary
circuit conductor makes an unwanted electrical contact, which restricts the flow of current below
the detection level of the protective devices. High impedance faults often occur when a
conductor breaks and falls to the ground. Such a situation leaves an energized conductor at
ground level, creating a public hazard, and any unsafe condition is of concern to utilities. For this
reason, the detection of high impedance faults (HIF) in electric distribution systems has been the
subject of intense interest over the history of utility systems. When a HIF occurs, the sparks are
usually created between downed conductor and contact surface and the conductor is bouncing
until the protection system remove the faulted system from the energy source. It causes to
increase the public hazard.

Traditionally, the technology for detecting phase-to-ground faults in distribution

networks is over current protection. Abnormal situations create excess of current that could
damage the system. Over current protection devices identify that unusual high level of current
and indicate fault. Their performance is successful in case of low impedance faults, when fault
current is important; but not in case of high impedance faults, when fault current can be lower
than the triggering threshold of the over current protection. In European distribution networks the
residual current under normal conditions is nearly zero, so ground fault currents above 5A or
10A are interpreted as faults by the protection. However, the residual current in American
distribution networks can reach 150A or 200A, so the current tap setting of the protection cannot
be set below this value, and consequently HIF currents are not detected.

There has been much research on arc fault detection methods at present. Some
researchers have studied the mathematical modeling of arcs by collecting experimental data and
thus obtained the parameters of the arc model. However, the research on arc fault models has
merely stayed in the simulation stage. There will be strong light and a high temperature,
electromagnetic radiation, noise and other physical phenomena when an arc fault occurs,
therefore, researchers are able to identify arc faults by detecting the occurrence of these physical
phenomena. Based on the features extracted from the electromagnetic radiation signal, least
squares and extreme learning machine are used to identify arc faults by collecting the voltage
signals in the circuit. However, the arc faults are very random and their locations are uncertain. It
is difficult to apply the detection methods which are based on sound, light, heat, voltage and
electromagnetic signals in the actual environment because the sensors are placed in fixed
positions. In order to solve the problems of fault detection in terms of location and isolation on
the transmission lines, a lot of research has been conducted. A fault location method that uses the
faulted negative-sequence voltage and locates the faulted sections was proposed by applying the
relationship between the fault distance and the clustered measurement groups. For a load change
that would not lead to a 180◦ phase angle change, the current-only method was able to detect the
over current fault without requiring the measurement of voltage. The EnKF-based approach is
able to accurately locate the short-circuit faults on transmission lines by ignoring the
foreknowledge of either the fault type and an approximate guess of the fault location. Arc faults
are very random and intermittent when compared to short-circuit faults and high-impedance
faults, therefore, it is unsuitable to apply the methods for fault detection and the detection of their
location in the transmission line mentioned above in the detection of arc faults.

1.1 OVERVIEW

A fault on a distribution feeder is an abnormal circuit condition which results in energy

being dissipated in a manner other than the serving of the intended load. Also known as a "short
circuit", a fault may result in damage to the electrical system, loss of power to customers, and/or
possible unsafe conditions. The traditional method of detecting and isolating such abnormal
conditions is over current protection. The primary purpose of over current protection is to protect
the electrical system, and is based on the electrical properties of the circuit. Excessive current for
too long will damage or interfere with normal operation of the system. The parameters (current
and time) are easy to measure. Devices to detect an over current event are well established, with
decades of field experience. In practice, the removal of safety hazards caused by the fault is an
additional motivation for over current protection.

Most of the faults on power systems result in a substantial increase in current flow
towards the fault point. Over the years, conventional over current based protection schemes have
been successfully used to detect and protect against these "low impedance" faults. However, for
HIFs on distribution systems, the high impedance of the fault does not result in a substantial
increase in current. Thus they cannot be reliably detected using conventional methods. This is
often the case with downed conductors on distribution systems. These downed conductors are a
major concern for electric utilities as a public safety hazard.

For many years, protection engineers and researchers have looked for solutions to this
long standing problem. This search has resulted in the development of several techniques to
detect HIFs. These groups of faults, in addition to being characterized as low current faults, tend
to exhibit random behavior with unstable and wide fluctuations in current levels. They are also
characterized by the presence of harmonics and high frequency components. Most of the
research on HIFs has focused on the development of sensitive detectors to reliably detect such
faults. Some of the techniques used standard substation relaying inputs; a few others required
special equipment.

1.2 General considerations

By analyzing of HIF characteristics, it can be seen that the HIF frequency spectrum is
very similar to capacitor switching and other events in distribution systems. Thus, this problem
sophisticates the detection of HIF from other events. Many researchers have reviewed and
documented the methods to detect of HIF and others have denoted the challenges to improve the
safety of detecting these faults. In several complexes technical, legal, economical, and
operational problems involved in high impedance fault detection are discussed. Reference
discusses technical as well as non-technical issues associated with applying high impedance fault
detectors. Besides, there is not a single method to completely solve the problem of HIF
identification

Many types of electrical loads can give false indications for arcing. With these false
alarms, it’s inevitable that workers become complacent and start ignoring any alarm altogether.
This must be avoided by developing a system that produces the highest confidence that an arcing
fault is occurring in their area. Currently, no such commercial apparatus exist to meet this need.
The scope of this thesis is to build on this previous research by applying today’s new technology
and see if any advancement can be made with new sensors and algorithms to design and develop
a manhole arc recognition system.
 CHAPTER 2

2. LITERATURE SURVEY

A vast literature applies machine learning models to support smart decisions in industrial IoT
environments. Most of these works propose solutions for anomaly detection related to security
aspects. A reduced number of works aims to detect anomalies in the sensed data produced by IoT
devices and caused by internal hardware and software malfunctions or external factors, such as
vibration.

To detect faults and errors in IoT devices using machine learning classifiers, Jan et al. [5,6]
propose a diagnosis system to detect sensor fault. The authors consider devices with limited
computation resources, such as memory, processing, and energy. This solution is distributed and
based on a Support Vector Machine (SVM) model, where response time is not the top priority
for the application.

Saeed et al. [7] analyze different classifiers using a dataset with drift fault injection. The faults
were injected using real data from a digital relative temperature/humidity sensor (DHT22) and an
Arduino controller. For detection, a Raspberry Pi is used. The performance was measured using
a large set of machine learning models: SVM, ANN, naive Bayes, KNN, and decision tree. The
authors compare the performance of the machine learning models in terms of precision, recall,
F1 score, and total accuracy.

Javaid et al. [3] aim to detect and diagnose faults based on decision fusion with different
classification techniques named: Enhanced K-Nearest Neighbor (EKNN), Enhanced Extreme
Learning Machine (EELM), Enhanced Support Vector Machine (ESVM), and Enhanced
Recurrent Extreme Learning Machine (ERELM). The authors consider the offset, gain, stuck,
and out-of-bounds faults.

Zidi et al. [8] apply machine learning for fault detection in wireless sensor networks. The
solution considers an SVM model for dealing with offset, gain, stuck-at, and out-of-bounds faults
in a real dataset from the University of North Carolina at Greensboro [9].

In a nuclear power plant scenario, Naimi et al. [10] propose fault detection and diagnosis based
on neural networks and a KNN algorithm applied to a pressurized water reactor. First, the neural
network performs detection. Second, the KNN algorithm classifies the faults. The KNN
performance is also compared to neural networks and SVM. This work includes bias, drift,
actuator offset, and saturation faults.

Khodabaksh et al. [11] consider a method for real-time data validation, gross error detection, and
classification. This work is based on data from petrochemical power plants of an oil refinery.
The injection of bias, drift, and precision degradation failures was based on statistical studies of
sensor data and the observation of changes in mean and variances. The classification relies on the
complex decision tree, neural network, and KNN algorithms. The performance was measured
using precision and recall. The authors made available the dataset for academic use [12].

So far, all the mentioned works do not consider any countermeasure to be applied after the faults
have been detected. Dofe et al. [13] present a comprehensive perspective on countermeasures
against IoT attacks. Mustafa et al. [14] propose three countermeasures to preserve privacy when
abnormal behavior is detected in assisted living applications.

Summing up, many works explore fault detection through the SVM classifier as can be observed
in Jan et al. [5,6] and Zidi et al. [8]. However, an industrial scenario requires a sensitive time
application to deal with faults, but the SVM demands higher computation time. Naimi et al. [10]
also do not consider the necessary time to apply a two-phase solution. Additionally, how to react
after the fault has been detected is an important aspect, but countermeasures are often considered
only from the security perspective. Therefore, there is a lack of novel solutions capable of
detecting and reacting against possible faulty situations in IIoT environments.
[1] Fang, W.; Xue, F.; Ding, Y.; Xiong, N.; Leung, V.C. EdgeKE: An on-demand deep
learning IoT system for cognitive big data on industrial edge devices. IEEE Trans. Ind.
Inform. 2023, 17, 6144–6152.

Motivated by the prospects of 5G communications and industrial Internet of Things (IoT), recent
years have seen the rise of a new computing paradigm, edge computing, which shifts data
analytics to network edges that are at the proximity of big data sources. Although deep neural
networks (DNNs) have been extensively used in many platforms and scenarios, they are usually
both compute and memory intensive, thus, difficult to be deployed on resource-limited edge
devices and in performance-demanding edge applications. Hence, there is an urgent need for
techniques that enable DNN models to fit into edge devices, while ensuring acceptable execution
costs and inference accuracy. This article proposes an on-demand DNN model inference system
for industrial edge devices, called knowledge distillation and early exit on edge (EdgeKE). It
focuses on the following two design knobs: first, DNN compression based on knowledge
distillation, which trains the compact edge models under the supervision of large complex
models for improving accuracy and speed; second, DNN acceleration based on early exit, which
provides flexible choices for satisfying distinct latency or accuracy requirements from edge
applications. By extensive evaluations on the CIFAR100 dataset and across three state-of-art
edge devices, experimental results demonstrate that EdgeKE significantly outperforms the
baseline models in terms of inference latency and memory footprint, while maintaining
competitive classification accuracy.

TECHNIQUE

 Deep neural networks (DNNs)

ADVANTAGES

 High accuracy
 Robust to fault distance

DISADVANTAGES

 It cause shock hazard

 Large disturbance characteristics
[2] Fazio, M.; Ranjan, R.; Girolami, M.; Taheri, J.; Dustdar, S.; Villari, M. A note on the
convergence of IoT, edge, and cloud computing in smart cities. IEEE Cloud Comput. 2022,
5, 22–24.

The purpose of the special issue is to cover all aspects of design and implementation, as well as
deployment and evaluation of solutions aimed at the osmotic convergence of IoT, edge, and
cloud computing, with specific reference to the smart cities application scenario. A high
impedance fault (HIF) is characterized by a small, nonlinear, random, unstable, and widely
varying fault current in a power distribution system. HIFs draw very low fault currents, and
hence are not always effectively cleared by conventional over current relays. Various schemes
are proposed to detect such faults. This paper presents a method to detect HIFs using a tool based
on mathematical morphology (MM). The method is implemented alongside the conventional
over current relay at the substation to improve the performance of this relay in detecting HIFs. It
is rigorously tested on standard test systems using PSCAD/EMTDC® to generate test
waveforms, and Matlab® to implement the method. Simulation results show that the proposed
method is fast, secure, and dependable. The transformation is carried out by a signal processing
function called Structuring Element. The structuring element is defined based on the type of
application and plays a pivotal role in MM operations. The real-time applications and the
performance of the MM operators have been reported. References discuss the effect of the
structuring element on the real-time performance of MM.

TECHNIQUE

 Mathematical morphology (MM)

ADVANTAGES

 Low fault current

 Improve performance

DISADVANTAGES

 High frequency components

 Large harmonics occur
[3] Javaid, A.; Javaid, N.; Wadud, Z.; Saba, T.; Sheta, O.E.; Saleem, M.Q.; Alzahrani,
M.E. Machine learning algorithms and fault detection for improved belief function based
decision fusion in wireless sensor networks. Sensors 2021, 19, 1334.

Decision fusion is used to fuse classification results and improve the classification accuracy in
order to reduce the consumption of energy and bandwidth demand for data transmission. The
decentralized classification fusion problem was the reason to use the belief function-based
decision fusion approach in Wireless Sensor Networks (WSNs). With the consideration of
improving the belief function fusion approach, we have proposed four classification techniques,
namely Enhanced K-Nearest Neighbor (EKNN), Enhanced Extreme Learning Machine (EELM),
Enhanced Support Vector Machine (ESVM), and Enhanced Recurrent Extreme Learning
Machine (ERELM). In addition, WSNs are prone to errors and faults because of their different
software, hardware failures, and their deployment in diverse fields. Because of these challenges,
efficient fault detection methods must be used to detect faults in a WSN in a timely manner.
Experimental results show that ERELM gave the first best result for the improvement of the
belief function fusion approach. The other three proposed techniques ESVM, EELM, and EKNN
provided the second, third, and fourth best results, respectively. The proposed enhanced
classifiers are used for fault detection and are evaluated using three performance metrics, i.e.,
Detection Accuracy (DA), True Positive Rate (TPR), and Error Rate (ER).

TECHNIQUE

 Enhanced K-Nearest Neighbor

ADVANTAGES

 Low current process

 Higher detection rates

DISADVANTAGES

 Less sensitive process

 High power consumption
 [4] Santo, Y.; Dalmazo, B.L.; Immich, R.; Riker, A. On the Performance of Machine
Learning at the Network Edge to Detect Industrial IoT Faults. In Proceedings of the 2022
IEEE 21st International Symposium on Network Computing and Applications (NCA),
Boston, MA, USA, 14–16 December 2022; Volume 21, pp. 291–295.

Industrial Internet-of-Things (IoT) massively deploys intelligent computing in industrial

production and manufacturing environments seeking automation, reliability, and control.
Machine Learning models provide intelligent decisions to drive manufacturing systems to the
next level of productivity, efficiency, and safety. One of the critical challenges that must be faced
is the deployment of Machine Learning models at the network edge to detect data anomalies
caused by Industrial IoT hardware failures, since industrial IoT devices are prone to errors and
failures. These anomalies can harm the industrial IoT system by producing false alarms,
consuming network resources, and affecting productivity. Because of that, it is critical to rely on
low latency and high precision detection systems to verify the data received from industrial IoT
devices. In light of this, we assessed key performance indicators of five machine learning models
running at edge computing, to provide in-depth discussions. The performance results were
obtained from an oil refinery scenario using a real industrial IoT dataset. The performance was
measured in terms of (a) Accuracy, (b) Precision, (c) Recall, (d) F1 score, (e) Training time, and
(f) Response time.

TECHNIQUE

 Machine Learning algorithm

ADVANTAGES

 Limited harmonic content

 Reduce heat accumulation

DISADVANTAGES

 Poor detection sensitivity

 It causing waveform distortion
[5] Jan, S.U.; Lee, Y.D.; Koo, I.S. A distributed sensor-fault detection and diagnosis
framework using machine learning. Inf. Sci. 2021, 547, 777–796.

The objective of this work is to design a sensor-fault detection and diagnosis system for the
Internet of Things and Cyber-Physical Systems. The challenge is, however, achieving this
objective within the limited computation, memory, and energy resources of the sensors. More
importantly, the detection of faults is time-sensitive, whereas the diagnosis does not necessarily
need to be fast. We propose a distributed sensor-fault detection and diagnosis system based on
machine learning algorithms where the fault detection block is implemented in the sensor in
order to achieve output immediately after data collection. This block consists of an auto-encoder
to transform the input signal into a lower-dimensional feature vector, which is then provided to a
Support Vector Machine (SVM) for classification as normal or faulty. Once detected, fault
diagnosis is performed at a central node, such as a network server, to reduce the computational
load on the sensor. In this work, a Fuzzy Deep Neural Network (FDNN) is used for diagnosis to
provide further information, such as the type of fault. Here, the input propagates through a deep
neural network and a fuzzy representation process. The output of these two components is then
fused through densely connected layers.

TECHNIQUE

 Fuzzy Deep Neural Network (FDNN)

ADVANTAGES

 High dependability and security

 Reliable features

DISADVANTAGES

 Lack of precision
 Less effective process
[6] Jan, S.U.; Lee, Y.D.; Shin, J.; Koo, I. Sensor fault classification based on support vector
machine and statistical time-domain features. IEEE Access 2017, 5, 8682–8690.

This paper deals with the problem of fault detection and diagnosis in sensors considering erratic,
drift, hard-over, spike, and stuck faults. The data set containing samples of the above mentioned
fault signals was acquired as follows: normal data signals were obtained from a temperature-to-
voltage converter by using an Arduino Uno microcontroller board and MATLAB. Then, faults
were simulated in normal data to get 100 samples of each fault, in which one sample is
composed of 1000 data elements. A support vector machine (SVM) was used for data
classification in a one-versus-rest manner. The statistical time-domain features, extracted from a
sample, were used as a single observation for training and testing SVM. The number of features
varied from 5 to 10 to examine the effect on accuracy of SVM. Three different kernel functions
used to train SVM include linear, polynomial, and radial-basis function kernels. The fault
occurrence event in fault samples was chosen randomly in some cases to replicate a practical
scenario in industrial systems. The results show that an increase in the number of features from 5
to 10 hardly increases the total accuracy of the classifier. However, using ten features gives the
highest accuracy for fault classification in an SVM

TECHNIQUE

 Support vector machine (SVM)

ADVANTAGES

 Effective approach
 It ensure high safety and reliable process

DISADVANTAGES

 It require a high sampling rate

 Difficult to accurate Phasor results
[7] Saeed, U.; Jan, S.U.; Lee, Y.D.; Koo, I. Machine learning-based real-time sensor drift
fault detection using Raspberry PI. In Proceedings of the 2020 International Conference on
Electronics, Information, and Communication (ICEIC), Barcelona, Spain, 19–22 January
2020; pp. 1–7.

From smart industries to smart cities, sensors in the modern world plays an important role by
covering a large number of applications. However, sensors get faulty sometimes leading to
serious outcomes in terms of safety, economic cost and reliability. This paper presents an
analysis and comparison of the performances achieved by machine learning techniques for
realtime drift fault detection in sensors using a low-computational power system, i.e., Raspberry
Pi. The machine learning algorithms under observation include artificial neural network, support
vector machine, naïve Bayes classifier, k-nearest neighbors and decision tree classifier. The data
was acquired for this research from digital relative temperature/humidity sensor (DHT22). Drift
fault was injected in the normal data using Arduino Uno microcontroller. The statistical time-
domain features were extracted from normal and faulty signals and pooled together in training
data. Trained models were tested in an online manner, where the models were used to detect drift
fault in the sensor output in real-time. The performance of algorithms was compared using
precision, recall, f1-score, and total accuracy parameters. The results show that support vector
machine (SVM) and artificial neural network (ANN) outperform among the given classifiers.

TECHNIQUE

 Artificial neural network (ANN)

ADVANTAGES

 High efficiency
 More reliable, and efficient

DISADVANTAGES

 Challenging task to design a systematic detection technique

 High computational burden
[8] Zidi, S.; Moulahi, T.; Alaya, B. Fault detection in wireless sensor networks through
SVM classifier. IEEE Sens. J. 2017, 18, 340–347.

Wireless sensor networks (WSNs) are prone to many failures such as hardware failures, software
failures, and communication failures. The fault detection in WSNs is a challenging problem due
to sensor resources limitation and the variety of deployment field. Furthermore, the detection has
to be precise to avoid negative alerts, and rapid to limit loss. The use of machine learning seems
to be one of the most convenient solutions for detecting failure in WSNs. In this paper, support
vector machines (SVMs) classification method is used for this purpose. Based on statistical
learning theory, SVM is used in our context to define a decision function. As a light process in
term of required resources, this decision function can be easily executed at cluster heads to detect
anomalous sensor. The effectiveness of SVM for fault detection in WSNs is shown through an
experimental study, comparing it to latest for the same application.

TECHNIQUE

 Support vector machines (SVMs) classification method

ADVANTAGES

 Robust reference under conditions

 Reduce data loss

DISADVANTAGES

 Reduce reliability process

 More difficult to angle measurement process
[9] Suthaharan, S.; Alzahrani, M.; Rajasegarar, S.; Leckie, C.; Palaniswami, M. Labelled
data collection for anomaly detection in wireless sensor networks. In Proceedings of the
2010 Sixth International Conference on Intelligent Sensors, Sensor Networks and
Information Processing, Brisbane, QLD, Australia, 7–10 December 2010; pp. 269–274.

Security of wireless sensor networks (WSN) is an important research area in computer and
communications sciences. Anomaly detection is a key challenge in ensuring the security of
WSN. Several anomaly detection algorithms have been proposed and validated recently using
labeled datasets that are not publicly available. Our group proposed an ellipsoid-based anomaly
detection algorithm but demonstrated its performance using synthetic datasets and real Intel
Berkeley Research Laboratory and Grand St. Bernard datasets which are not labeled with
anomalies. This approach requires manual assignment of the anomalies' positions based on visual
estimates for performance evaluation. In this paper, we have implemented a single-hop and
multi-hop sensor-data collection network. In both scenarios we generated real labeled data for
anomaly detection and identified different types of anomalies. These labeled sensor data and
types of anomalies are useful for research, such as machine learning, and this information will be
disseminated to the research community.

TECHNIQUE

 Regression Algorithm

ADVANTAGES

 High-level findings
 High penetration of renewable resources

DISADVANTAGES

 High computational complexity

 Poor sensitive process
 [10] Naimi, A.; Deng, J.; Shimjith, S.; Arul, A.J. Fault Detection and Isolation of a
Pressurized Water Reactor Based on Neural Network and K-Nearest Neighbor. IEEE
Access 2022, 10, 17113–17121.
Nuclear power plants (NPPs) are complex dynamic systems with multiple sensors and actuators.
The presence of faults in the actuators and sensors can deteriorate the system’s performance and
cause serious safety issues. This calls for the development of fault detection and diagnosis
systems for detection and isolation of such faults. In this study, fault detection and diagnosis
(FDD) based on neural networks (NN) and K-nearest neighbour (KNN) algorithm is applied to a
pressurized water reactor (PWR). Fault detection is first determined based on the NN. Second,
the KNN algorithm is used to classify the faults. The proposed approach is capable of classifying
a variety of actuator faults, sensor faults, and multiple simultaneous actuator and sensor faults. A
set of simulation results is provided to demonstrate the accuracy of the FDD method. The
classifier performance is further compared with other machine learning techniques.

TECHNIQUE

 K-nearest neighbour (KNN) algorithm

ADVANTAGES

 Reduction of maintenance costs

 Reduce the faults of electrical systems

DISADVANTAGES

 Complex of control parameters

 Complex nature of the reliable mathematical description
 CHAPTER 3

EXISTING SYSTEM

AFM aims to ‘systematically support grid resiliency under grid disturbances rather than
negatively impacting the disturbed grid’.

The purpose of developing AFM for NMs is to address major hurdles in integrating NMs into a
distribution network.

First, a minor fault in the main grid could cause the microgrids to disconnect because the
inverter-dominant microgrids are highly sensitive to grid anomalies. Unexpected disconnections
can increase the risk of major blackouts during large grid disturbances.

A major difficulty for fault ride-through is the complex ride-through objectives and constraints
in microgrids.

DISADVANTAGES

 Challenges to the stable operation

 Unstable sub synchronous oscillations
 Instability phenomena
 Complicates online monitoring

PROPOSED SYSTEM

DA-AFM is formulated as a distributed optimization problem that supports customized

requirements and the plug and play of microgrids or microgrids components. Any AFM
objectives such as currents or powers can be added into or removed from the formulation easily.

A distributed and asynchronous surrogate Lagrangian relaxation (DA-SLR) method is

devised to coordinate various microgrids during faults. AFM for each individual micro grid is
computed on a separate core. This computation scheme leads to a genuine distributed and
asynchronous algorithm.
 A software-defined networking (SDN) architecture for enabling low-latency distributed
computing is integrated to empower DA-AFM, making it hardware-independent, fast and
programmable, and resilient to communication delay and link loss by the ultra-fast activation of
backup channels.

ADVANTAGES

 Large convergence performance

 Reduce communication delay and failures
 Increase computation efficiency
 Improve power quality

SYSTEM REQUIREMENTS

HARDWARE DETAILS

 Arduino Uno Controller

 Power Supply Unit
 Humidity sensor
 LCD Display
 ESP8266 WiFi IC

SOFTWARE REQUIREMENTS

 WINDOWS 10
 PYTHON IDE
PROPOSED BLOCK DIAGRAM

Arduino Voltage Power

Microcontroller Source control
PS Unit Analysis analysis

Load
Monitoring

ESP 8266 Wifi

Device Load Voltage
Analysis

Grid Humidity
output Analysis
 CHAPTER 4

4. PROPOSED PROCESS EXPLANATION

4.1 High Impedance Faults (HIFs)

High impedance faults (HIFs) happen when an energized conductor of the distribution
network falls on the ground, making unwanted electrical contact with a road, a sidewalk, or some
other surface with high resistive value. HIFs can also be caused by trees growing near lines,
whose branches might get in contact to an energized conductor. This contact restricts the flow of
fault current to a level below the one detectable by conventional over current devices. As a
consequence, the energized conductor on the ground surface can pose public danger, as well as
risk of fire due to the probable arc ignition. The damage derived from HIF concerns people,
animals, and properties rather than electrical equipment of the network.

Figure 3: HIF current

4.1.1 High impedance fault Overview

A high impedance fault (HIF) results when an energized primary conductor comes in
contact with a quasi-insulating object such as a tree, structure or equipment, or falls to the
ground. The significance of these previously undetectable faults is that they represent a serious
public safety hazard as well as a risk of arcing ignition of fires. A high impedance fault is
characterized by having impedance sufficiently high that it is not detected by conventional over
current protection, such as fuses and over current relays. Unlike low impedance short circuits,
which involve relatively large fault currents and are readily detectable by conventional over
current protection, these HIFs represent little threat of damage to power system equipment. High
impedance faults produce current levels in the 0 to 50 ampere range. Typically, an HIF exhibits
arcing and flashing at the point of contact. Throughout the utility industry, there has been
differences of opinion on how often HIFs occur. Normally, utilities do not keep good records on
the number of down conductor instances. It is seldom recorded on trouble reports unless it results
in a fuse or breaker operation. While it is likely that only a few percent (5-20%) of all
distribution faults are high impedance faults, means exist to detect a high percentage of HIFs.

4.2 MECHANICAL DETECTION METHODS

One type of mechanical HIF detection method consists of a device(s) mounted to a cross
arm or pole. A unit is mounted under each phase wire. It provides a low impedance ground fault
by catching the falling conductor. The force of the falling conductor releases an internal spring
that ejects a bus bar to make contact with the fallen wire and create a ground fault. The ground
fault created will cause conventional over current protection to operate. Sagging conductors that
do not come in contact with earth or a grounded object could be detected by this mechanical
method. The installation and maintenance costs are high. For bi-directional coverage, six units
would have to be mounted on each pole. Even though the cost may be high to allow usage on
every pole, utilities may install in certain areas, such as churches, schools, or hospitals. Another
type of mechanical HIF detection method uses a pendulum mounted aluminum rod with hooked
ends. It is suspended from an under-built neutral conductor. The falling conductor is caught and
produces a low impedance ground fault, which operates conventional over current protection.
Typically, two units are mounted per span. Sagging conductors that do not come in contact with
earth or a grounded object could be detected by this mechanical method. Ice, wind, and tree
growth could cause a false detection

4.3 High Impedance Fault Analysis System

This electrical HIF detection method measures the third harmonic current phase angle
with respect to the fundamental voltage. There is a distinct phasor relationship between the third
harmonic current and the faulted phase voltage. The device calculates and stores the average
ambient third harmonic current phasor. When a fault occurs, the new third harmonic current
phasor is vectorially subtracted from the stored value. A high impedance fault is issued if the
magnitude is above setting and angle matches a predetermined value for a down conductor. The
device acquires current and voltage values from the relaying current and voltage transformers.
Typically, one unit is installed in each distribution breaker.

4.4 Distribution system grounding

In power systems, several main methods for distribution system grounding are used. The
amplitude of the ground fault current depends on the system grounding method. In an isolated
grounded system, there is no internal connection to ground. Since the phase-to phase voltage
triangle is intact, the system can keep working without being disturbed. To compensate, the
insulation rate has to reach a phase-to-phase level. As with ungrounded systems, the high
impedance grounded method can also allow the utility to operate with a limited fault current.

Figure 4: Typical solidly grounded 11 kV distribution system

With an effective or solid grounding system, the risk of overvoltage during the ground
faults can be reduced. In this thesis, HIF in 11 kV distribution power systems is the primary
research focus. In Australia, the majority of the distribution power systems on 11 kV are 3-phase
3-wire uni-grounded system. The main advantage of 3-wire systems is that there are no severely
unbalanced loads. In this case, a sensitive ground relay can be used to monitor the residual
current, which would be investigated and used to detect HIF. One limitation of HIF detection in
this grounded system is that the residual current is polluted by the line capacitance current, non-
linear load and transformer saturation.

In truth, detecting a noisy arcing fault that produces high levels of harmonics and a
distorted current waveform is relatively achievable. However it is difficult to make an accurate,
definitive decision that unusual activity on a feeder event is in fact a fault, as opposed to one of
many normal distribution feeder events. HIF with several hundred amperes of current on a
distribution feeder can be indistinguishable to loads which can be of the same magnitude. It is
not always possible to find characteristics to describe a specific fault and uniquely identify it.

4.5 Problem Statement

In reality, there are a number of factors that make the detection of HIF more complicated
and some of them are discussed here.

4.5.1 Random behavior

Most HIF detection techniques proposed in the literature were developed based on
simulations or laboratory experiments. In real life, many external factors can contribute to the
randomness of HIF behavior. These factors are (but not limited to) wind speed, surface
conditions and moisture. The surface conditions can affect the HIF current magnitude greatly
such as the moisture content, coarseness and the dielectric properties of the surface. Different
kinds of contact surface produce different HIF currents. The magnitude and content of HIF
currents keep changing as it is happening due to various factors such as surface moisture
evaporation, chemical reaction, etc. The weather conditions affect HIF as well. In wet weather,
the HIF currents can be large enough to trigger conventional over current devices. The wind
speed and the electro-dynamic force also have significant impacts, it makes the conductors dance
and can change the gap distance between the conductor and the contact surface. The arc can
disappear momentarily for a few cycles and then re-ignite. The HIF currents would become
intermittent, the magnitude and content would change accordingly. Besides the contact surface
and weather conditions, the location where the HIF happens also has impacts on the current
signal. The HIF can be relatively easier to be detected if it happens close to the substation since
the fault currents are less attenuated and pass through fewer network devices.
4.5.2 Low fault current amplitude

In power system protection, over current faults are detected and eliminated by setting a
threshold in the over current sensing device. They can be an over current relay, a re-closer or a
fuse. To avoid false tripping which can cause unnecessary customer outage, a trade-off threshold
value is set. The ability of these devices to detect HIF is limited by the settings. In most cases,
HIF currents are much lower than the set threshold values.

4.5.3 High cost

Underground cables are being used extensively in distribution networks. Nevertheless,

the primary means of delivering power is still via bare overhead lines due to the large
geographical area coverage. One proposed solution is a mechanical mechanism to catch the
broken conductor and connect it the ground to produce a low impedance fault so that
conventional protection is able to clear the fault. This was found to be ineffective and the cost of
installing the device is high. Voltage loss sensors or action recognition cameras could be a
solution too but they suffer the same problem of high installation and maintenance cost. A
feasible but more cost-effective solution is required

Other events and load types

Another difficulty of detecting HIF is that some other events on the network share similar
characteristics with HIF. Some of the examples are transformer magnetizing inrush currents,
capacitor bank switching and non-linear load such as motor starting. The HIF-like currents can
falsely trigger the HIF detection devices.

4.6 HIGH IMPEDANCE FAULT DETECTION

Energy demand in the pre-industrial world was provided mostly by man and animal
power and to a limited extent from the burning of wood for heating, cooking and smelting of
metals. The discovery of abundant coal, and the concurrent technological advances in its use,
propelled the industrial revolution. Steam engines, mechanized production and improved
transportation, all fuelled directly by coal, rapidly followed. The inter war years saw the rise of
oil exploration and use. Access to this critical fuel became a key issue during the Second World
War. Post - war industrial expansion and prosperity was increasingly driven by oil, as was the
massive growth in private car use. More recently a new phase of economic growth has been
underpinned to a great extent by natural gas. A substantial proportion of coal and gas production
is used to generate electricity, which has been widely available now for over a century.
Electricity is a premium form of energy due to its flexibility and ease of distribution. Demand
worldwide is growing, driven by the explosion in consumer electronics, the associated industrial
activity and the widening of access to consumers in the developing world. Detection of downed
power lines is a long-standing problem to electric utilities. High impedance faults result in very
low currents which are often not detectable by conventional over current relays. A HIF occurs,
for example, when a conductor breaks and falls on a no conducting surface such as asphalts road,
sand, grass or a tree limb producing a very small current.

These faults are difficult to detect when the impedance at the point of fault is high enough
to limit the fault current to the region unprotected by conventional over current devices. When no
solid return path for the current is available, the fault exhibits arcing phenomena; these faults are
then referred to as “high impedance arcing faults”. HIFs are a dangerous phenomenon since risks
of electric shocks are posed to the public and fire hazard also exist. It is estimated that majority
of electrically caused fires are due to arc type, hot neutral intermittent faults. Therefore, the
principal motivation in high impedance fault detection is not just system protection, but to
improve safety. The threshold of over current relays must be set at a relatively high current level
to prevent tripping by inrush currents thereby causing unnecessary service interruption. Most
detection schemes involve the adjustment of the existing over current protection to be more
sensitive by lowering its setting. Such scheme have failed to operate in 32% of high impedance
faults and lead to several unexpected service interruptions
 Fig.5. Relation of high impedance fault current to over current.

In the past two decades many techniques have been proposed to improve the detection of
HIFs in power distribution systems, and recently the utilities have intensified research programs
searching for more efficient protection against this type of a fault. Some of the techniques used
to deal with this problem are mechanical methods where various mechanical devices are used to
provide a low impedance fault by catching the fallen conductor. Others have used electrical
methods and techniques

Difficulties in Detecting HIFs and Influence of the Network Configuration

The difficulty of detecting HIFs is determined by the configuration of the distribution

network and by the loads connected to the system. Hence in some countries and under some
circumstances, HIF detection is a considerable concern, while in others it remains a secondary
issue. Detecting electrical faults implies to extract information that reveals presence of fault from
some electrical signal. The difficulty in detecting faults remains in how easy is extracting that
information. Regarding HIFs, the crucial task is to recognize relevant characteristics of HIFs in
the residual current. The more unfavorable cases for the HIF detection are those where the
residual current contains several components and the HIF component is minimal, or where the
residual current under normal conditions presents characteristics similar to HIFs.

4.6.1 Approaching the High Impedance Fault detection Problem

The general lack of knowledge of HIFs is the first difficulty for developing a practical
detection method. Consequently, the initial stage of our study is aimed to understand and explain
the HIF detection problem. HIF research is not yet a well-known subject. For this reason it is
needed to accomplish some stages before being capable of developing a detection method. The
three stages of the procedure that we propose to characterize HIFs

Fig 6 Approach of high impedance problem

1. Learning about HIFs, identifying the difficulties of the detection and reviewing the
typical characteristics.
2. Collecting HIF current recordings to conduct a study of its characteristics. Simulation,
laboratory tests, and requesting field data are the three practices for collecting recordings.
3. Analyzing and processing the mentioned database in order to develop a list of indicators
able to characterize HIFs.

4.6.2 High frequencies in kHz range

Input impedance is not affected by normal operation, e.g. changing loads, but faults do.
The work in is based on this idea. One of the key features in HIF is the arcing. There is high
frequency in this transient. The detection approach of is based on feeder fault high frequency
signals of 2-10 kHz to detect the fault. A microcomputer-based prototype relay based on the high
frequency changes in the current was installed on the feeder

High frequency components of the current in the circuit are utilized to detect the HIF.
The high frequency components of each cycle are evaluated to compare the magnitude of the
high frequency components with the previous values. A prescribed pattern is used as reference to
determine the HIF occurrence if there is significant increase of the magnitude. In a similar
approach, a method to improve the system operation reliability was introduced. It is achieved by
applying the difference of high frequency components between normal and faulty operations.
 Relying solely on high frequencies as detection features is inherently not reliable. Having
smaller magnitudes, the high frequencies are easier to be masked out by noise and interference.
They also experience higher attenuation during propagation from the fault location to the sensor.

High Frequency Current Components (2-10 kHz)

In their preliminary research, they hypothesized that the key characteristic for identifying
HIF was the arcing associated with these faults. From their research, they found HIF are non-
periodic and transient in nature. Therefore, it is difficult to obtain a precise frequency spectrum
of an arc. Based on field tests, increased high frequency components were observed after
processing through Fourier Transform. The researchers explain that the high frequency
components caused by HIF are based on the theory that individual arcing impulses exhibit a
wideband sin x x frequency spectrum, shown in Figure. They relied on the increase in
magnitude of the high frequency component (2-10 kHz) of the current as an indicator of a fault
and used its duration to discriminate against the switching transient events. A prototype Arcing
Fault Detection (AFD) utilizing this detection technique was installed and monitored over
several months

4.7 HIF characteristics

By examining the waveform of the HIF currents, the following characteristics are found.

Low level

The current is only a few amps due to the high impedance path. This is one of the reasons
why it is difficult to detect HIF using the conventional over current devices.
Randomness

Another important feature found is the random behavior of HIF current. HIF random
characteristics; one is fluctuation of current amplitudes due to unstable impedance caused the
unstable physical contact and chemical reaction at the contact, and the other is that the arc
currents become zero randomly due to the unstable contact surface. Also the U-I curves are
different from each cycle by observing the U-I hysteresis loops.

Figure 7 Unpredictable fluctuations in the amplitude showing randomness characteristics of

HIF current

Current is intermittent

The resistance is hardly constant due to the chemical reaction, moisture evaporation, etc.
The randomness of the contact conductance causes intermittent current as shown in Figure

Figure 8: Intermittent current showing the intermittency characteristics of HIF

Zero crossing and harmonics

One important feature found is the distortion around the zero value; this produces
harmonics in the HIF current. The zero-crossing tends to become less obvious as the voltage
increases because less time is required to re-ignite the arcing.

Figure 9: Waveform showing distortion at zero crossing

4.8 ARC CURRENT NATURE

An arc is defined as a luminous electrical discharge flowing through a gas between two
electrodes. In the case of an arcing HIF, when an energized conductor touches the ground, the
electric contact is not solid. Due to the existence of air between the ground and the conductor,
the high potential difference across a short distance excites the appearance of an arc. Many
authors have worked on the theory and dynamics of voltages and currents in an electric arc, most
such studies are experimentally based. In a model explaining the phenomenon using a spark gap
was proposed. This air gap will not conduct till the applied voltage reaches the breakdown point.
Then the current flows and reaches a maximum when the applied voltage equals the arc voltage.
After that, the arc current decreases and becomes zero, i.e. the arc is extinguished.

When extinction occurs, the arc requires a potential, known as restrike voltage, to
reignite. This reignition will have the opposite polarity. This explains the typical voltage-current
waveform of an arc shown in Fig. Many electric models have been proposed describing arc
behavior as reviewed
 Fig. 10. Electric arc voltage and current shapes

In the context of downed conductors, Russell conducted staged HIF tests studying
dependencies of arc current magnitude on potential difference, gap distance features of the
grounding surface and environmental conditions of the grounding point. A high degree of
random behavior was observed due to impurities near the grounding point, heat from the arc that
is intense enough to fuse substances and the evolution of different paths for current flow on
surfaces accurately modeled so far. Some previous researchers have reached a consensus that
HIFs are nonlinear and asymmetric, and that modeling should include random and dynamic
qualities of arcing. Emanuel et al suggested two dc sources connected anti parallel with two
diodes to simulate zero periods of arcing and asymmetry

The high impedance fault model proposed is shown in Fig. and includes two DC sources,
Vp and Vn, which represent the arcing voltage of air in soil and/or between trees and the
distribution line; two resistances, Rp and Rn, between diodes, which represent the resistance of
trees and/or the earth resistance; and since most observed arcs occur in highly inductive circuits
two inductances, Lp and Ln, added to the circuit. The effect of the inductances leads to the
nonlinearity loop in the V-I curve and the desired asymmetrical shape for the HIF current. When
the line voltage is greater than the positive DC voltage Vp, the fault current starts flowing
towards the ground. The fault current reverses backward from the ground when the line voltage
is less than the negative DC voltage Vn.
 Fig 11. A two diode fault model for a HIF containing Rn, Rp, Ln, Lp.

In the case when the line voltage is in between Vp and Vn, the line voltage is
counterbalanced by Vp or Vn so that no fault current flows. As a direct result of the presented
model the typical high impedance fault current and V-I curves were produced and are shown in
Figures

When a conductor is left energized on the ground, it represents an increased hazard to the
public in the immediate vicinity of the downed wire. Therefore, the solution appears to be
obvious - employ HIF detection and clear the circuit immediately for such conditions.
Unfortunately, the decision is not that simple. No device, over current or otherwise, can protect
from initial electrical contact. The type of HIF detection currently available is relatively slow.
Therefore, its ability to prevent and protect against an injury resulting when the downed
conductor event and the occurrence of electrical contact close together is limited.

Fig.12. Current curve for HIF.

4.9 THE HARMONIC MODEL

A signal can be defined as a function that carries information, usually about a state or a
procedure of a physical system. However, signals can be represented in several ways.
Mathematically, a periodic and distorted signal can be suitably represented in terms of its
fundamental frequency and harmonic components expressed as a sum of sinusoidal waveforms
referred to as the Fourier series. Each frequency is an integer multiple of the fundamental system
frequency. In order to obtain an approximation of such waves, mathematical models are
employed
4.9.1 Mathematical Model

Arc models can be classified in two categories: the former based on dielectric recovery
and the latter based on the arc thermal characteristics. The thermal characteristic models are
widely used as they give a better interpretation of the arc behavior. Hence last researches in this
field have been focused on developing new thermal models, improving the existing ones and
validating them by comparing experimental and computed data.

The dynamic description of the arc conductivity has been improved and modified to
increase the models validity and to reduce the computational burden. It derives from the
Hochrainer arc description through the following differential equation:

4.9.2 Harmonic Principle

Harmonic is another important feature of HIF, as explained in an AC cycle, the resistance

near zero will be greater than the resistance near peak. Harmonics can be a representation of this
non-linearity. In this paper, third harmonic current with a phase difference around 180º
comparing to fundamental current is used to represent this non-linear feature

The detection criterion of harmonic principle is as following: 1) The fundamental

amplitude is above a threshold. 2) Amplitude ratio between 3rd harmonic and fundamental is
above a threshold. 3) Phase difference between 3rd harmonic and fundamental is within a range
around 180°. 4) Above three requirements last for non-trivial time duration.
 Fig 13 Harmonic principle

In order to eliminate the interferences of background harmonics such as steady non-linear

load, an average-samples based increment sample value (ASI) is used to extract the fault feature.
A one-cycle array of sampled points keeps updating to tracking the average shape of current
waveform. If the amplitude of the newly sampled cycle is greater than the amplitude of the
average samples array, the ASI will be calculated by subtracting the average samples array from
the newly sampled array. This ASI array then will be used in the harmonic feature extraction.
Once the criterion has been satisfied for a pre-set duration, an HIF harmonic flag will be issued.

4.9.3 Harmonics analysis of HIF signals

By analyzing the HIF current, many unique features and characteristics of the frequency
spectrum can be found. In detection method was proposed based on the third-order harmonic. In
field measurements were carried out to record the second and third harmonics. A theoretical
model together with laboratory measurements demonstrates that HIF detection can be achieved
by monitoring the second-order harmonic

Based on various HIF features discovered by previous researches proposed an integrated

HIF and LIF (low impedance fault) detection method. The features used in this method are:
magnitude of the fault current, the 3rd and 5 th harmonic currents, the 3rd harmonic current
angle, the angle difference between the 3 rd harmonics current and the fundamental voltage, the
negative sequence current. The logic circuit for implementation was provided to process these
features. There are always rich harmonics associated with high impedance arcing fault. To make
use of this feature, a detection method based on quasi-static ripple harmonics, sub and super
harmonic frequencies was proposed

These detection methods have relatively high detection rates but low accuracy. This is
because the methods relying only on harmonics lose the time information when extracting
harmonics features. Accuracy is about discriminating HIF from other types of faults or switching
events, especially those producing HIF-like signals.

4.10 FIELD DATA AND MODELING

Due to the lower steady state fault current, HIF features mostly come from the more detailed
fault characteristics. These are highly relying on the unpredictable and complex fault conditions
and environments. Various stage tests have been conducted to investigate these features of HIF.
As a summary, these features can be classified as following:

1) The arcing features: HIF burns with AC arc at the fault point, due to the lower fault
current, the arcing details features become phenomenal. The AC arc has a nature of
crossing zero quenches and re-ignitions: every half cycle, when arc current crosses zero,
the injection power goes down letting the arc path goes cooler, resulting in increase of
resistance and consequently, a short time quench when current cross zero, when the
voltage goes up, the arc will be re-ignited. Fig. borrowed from Taiwan Power’s stage
fault result shows this distinctive feature of the fault current.
2) The harmonic features caused by the ground fault nonlinearity. The path resistance is
varying periodically according to the thermal condition of each cycle: when current
increases, the heat of fault path accumulates, the conductance will increase as well.

4.10.1 Fundamental Principle

The erratic increases and decreases of the amplitude in the scale of cycles is another
focus for HIF detection. For normal loads and operations, the changes of waveform will be
normally predictable: there will be no such lasting-randomly changes from cycle to cycle. In
order to capture such HIFs features and to discriminate from normal operation, following aspects
should be considered

HIF fault burst current is the superimposed component mainly determined by the fault
path resistance. Therefore, the increment amplitude is preferred for detection algorithm. An
average-amplitude based increment value (AAI) is calculated for the fault detection. An
adjustable low-pass filter is used to track the average amplitude in the scale of the normal load
change rate. The AAI will be calculated by subtracting the average amplitude from the newly-
sampled current amplitude. Fault evaluation is based on identifying the status of burst or-
extinguish by comparing the AAI to a threshold. Fault will be captured by counting the changes
of burst-or-extinguish status of AAI within pre-set time duration

Fig.14 Illustration of feature extraction of fundamental principle

4.11 HIF arc models

A model which can encapsulate typical features of HIF provides an effective and flexible
way for research into HIF detection. Many previous researches on HIF detection relied heavily
on field tests. To overcome the complications of obtaining staged fault data or of studying
difficult abnormal conditions in the networks, a model which can accurately represent typical
HIF features is essential for the efficient and economical development of reliable detection
algorithms
 Figure 15: Emanuel arc model

Figure 16: Model using non-linear impedance

Figure 17: HIF model using time-controlled switching

Figure 18: A simplified arc model

 Figure 19: HIF combination model based on Emanuel arc model

4.12 HIF FEATURES USED IN DETECTION APPROACHES

Magnitude of Phase or Natural Fault Current and Voltage

There were approaches presented based on the magnitude of phase or natural fault current
measurement and as the HIF has a little current, these algorithms frequently failed to detect the
fault. Founded on these approaches, the authors proposed a proportional relaying algorithm and
introduced a new electromechanical relay. The unbalancing in phases currents in fault duration
was a criterion applied to detect the HIF. According to such criterion, in the ratio ground relaying
(RGR) algorithm and in other schemes appropriate for electromechanical relays were presented.
In a comparison was performed between a ratio ground relay and a proportional relay. The
proposed algorithm was based on the change of the neutral voltage and zero sequence currents.
In an algorithm consisting of monitoring the shunt resistance of each feeder derived from
substation was considered as a tool to detect the high impedance fault. In a modified directional
earth fault relay was suggested to identify HIF. The differences between HIF detection for
ungrounded systems and multigrounded systems are expressed and next, a method based on the
system fundamental frequency component of residual current and voltage was proposed to
determine HIF. In another protection scheme based on active power variation was developed,
and a dissipation factor (DF)-based criterion was proposed to identify load switching operation.
This method was implemented by cross-correlation analysis between phase voltage and residual
current in single-phase network. Reference proposed the algorithm which calculated the
influence of feeder admittance unbalancing in neutral voltage in order to detect the HIF.

As aforementioned, current waveforms of the HIF is contaminated by low and high order
harmonics and sub harmonics which are used as indexes to detect these faults. To extract the
noted low and high components from fault current waveforms, several authors have used the
approaches based on signal processing schemes as follows:

Low Order Harmonics, Sub-harmonics and Low Frequency Spectrum

The vast frequency spectrum can be derived via analyzing the HIF current. In this
subsection, the HIF features supporting its detection are explained. They are based on the
variation of one or more characterizing or noncharacterizing harmonics fault current signal. Also
using 3rd harmonic voltage and current changes, relay performance similar to a conventional
60Hz directional impedance relay was attained for fault currents detection. In the third harmonic
current relaying algorithm and in a more sophisticated technique that involved finding a
Chisquared test statistic using 60 Hz sequence components and harmonics were suggested to
detect HIF. In the authors proposed a method based on negative sequence current. An algorithm
based on the symmetric components was proposed to detect and identify the balance condition of
the system during the fault. Next, other ideas such as using the information contained in the low
frequency spectral behavior in terms of both magnitude and phase increasing frequency
components near to 60 Hz focusing on the low order harmonic current in order to obtain the
features to distinguish arcing faults from switching events and detection method using even order
harmonics were proposed.

The technique presented in monitors the unbalancing in the fundamental, third, and fifth
harmonic feeder currents at the substation and performs a statistical evaluation of the present
unbalancing relative to past levels of it. References introduced three criteria of an even order
power, an even order ratio, and an even order incremental variance criterion for the fault
detection. Energy variance criteria determination and threshold tuning scheme for high
impedance fault detection were explained and self-tuning of fault detection threshold was
described. In a relay was explained based on second order harmonic current algorithm. An
adaptive detection method founded on low order harmonics and harmonic ratio characteristics
was described. In an algorithm was proposed regarding to the increase in the energy level (off
harmonics) and the degree of randomness associated with arcing HIF. In a combined algorithm
was presented, which consists of five sub-algorithms including: Energy, Randomness, Arcing
phase signature, Spectral analysis, and Load analysis algorithm. In a comparison among four
high impedance fault detection algorithms was made, namely the proportional relaying
algorithm, the ratio ground relaying algorithm, the second order harmonic current relaying
algorithm, and the third order harmonic current relaying algorithm. In a communication link
relaying algorithm is explained as the sub-harmonic relaying algorithm. In the authors have
proposed a new adaptive algorithm with using harmonic characteristics of HIF. A multiple
algorithm based mainly on energy, randomness, and inter-harmonics was presented. In a scheme
was developed based on a concept of quasi-static ripple harmonics and sub/super harmonic
frequencies. In energy content of even, odd, and in-between harmonics is utilized as a criterion to
detect the HIF. In the authors have used fault features ranging from transient high frequency
harmonic distortion to fundamental intermittent in order to determine the HIF. The use of sub
harmonic oscillation and harmonic distortion was investigated in as a means of anticipating of
HIFs. The high impedance fault detection system reported in was based on algorithms that use all
harmonic and non-harmonic current components in all of the 3-phases and ground

High Frequency Spectrum

The HIF causes the creation of high frequency components in current signals. So, one of
the criteria to detect the HIF was localizing the high frequency energy level in current signals.
Naturally, normal events such as capacitor switching cause the high frequency energy level to
increase similar to HIF. Regarding that energy level increasing in HIF has random nature, thus to
solve such a problem, other criteria named randomness criteria were defined. Some researchers
have employed one or two of above mentioned criteria to detect the HIF. The monitored system
impedance was affected by high frequency components in current signal. Therefore in the HIF
was detected by high frequency impedance monitoring. Reference presented a novel algorithm
based on random behavior of HIF. The reference described a conventional detection technique
that compared energy values of load currents to preset thresholds in order to detect an arcing
fault on a power line. In the authors describe an identification approach involving low frequency,
high frequency, and inter harmonic parts for detecting HIF.

4.13 Fault Features

In many proposed algorithms phase currents are used to extract the features of HIAFs,
which, however, can be easily covered by the load components. Moreover, harmonics and
randomness produced by the loads can also greatly affect the reliability and security of a HIAF
detection algorithm if using the phase currents.

Zero sequence current is known as the superposition of three phase currents, and has been
applied in many practical protections. In the medium-voltage distribution system, loads are
commonly supplied through a step-down transformer, the high-voltage side of which is
commonly known to be wired without neutral lines especially at the 6-35kV system in China.
Therefore, the zero sequence current at the high voltage side will not be affected by the loads at
the lower voltage side. It is illustrated by supposing the delta wiring at the high-voltage side.
With filtering of the delta wiring, sum of load currents iX,H (X=A,B,C) at the high-voltage side
always equals to zero, and the zero sequence current i0 at the ‘measuring point’ is calculated as:

where, iX represents the phase currents at the measuring point, if represents the fault
current, iX ,c represent the leakage currents through the line to ground capacitors. That is to say
the zero sequence current mainly consists of the unbalanced component of the three phase
capacitor leakage currents and the fault current, where the former mainly depends on the
unbalanced network and is commonly small at the medium-voltage system.

Fig.20 Network with the D-Y wiring step-down transformer

Features of HIAFs vary greatly under different situations, including materials of

grounding surfaces, system operations, surface humidity, and weather conditions.

Type-1: After fault happens, the arc is with short ‘unstable arcing period’ and comes into
the ‘stable arcing period’ very early. When grounded with different surfaces, both the fault
distortions and current amplitudes could be various as shown in Fig

Type-2: The arc is with longer ‘unstable arcing period’ compared with the Type-1.
Furthermore, the HIAFs only with random fluctuations are further classified as Type-2A while
those both with random fluctuations and intermittences are classified
 Type-3: For some HIAFs, due to the unbalanced operations or the extremely high
grounding impedance, amplitudes of post-fault zero sequence current may have few differences
between that of the pre-fault. In this case, the current amplitude is much lower and waveform
distortions may be covered by the non-fault components more or less, leading to more
difficulties in the detections

Diverse types of HIAFs and their performances make it challengeable for an algorithm to
always keep high detection reliability with just single feature description and identification
method.

As introduced in Section I, harmonics, randomness and waveform distortions are three

most predominant features of HIAFs. Firstly, harmonics present abnormalities both during the
stable and unstable arcing period. However, as shown in Fig.6, significantly different amplitudes
or percentages of harmonics can be presented because the magnitudes and distortion degrees of
currents vary under different conditions. A better definition of harmonic intensity is important to
guarantee the detection reliability and security at the same time.

Secondly, also demonstrated in the occurrence, duration and fluctuation range of unstable
arcs are all uncertain for HIAFs. Moreover, fault arcs sometimes can be self-extinguished or
distinguished by the arc-suppression devices in seconds. As a result, only utilizing randomness
features cannot detect the HIAFs beginning with stable arcs or detect them as soon as possible.

As a result, it is necessary to combine the multiple features to respectively detect HIAFs

beginning with unstable or stable arcs, and keep reliability under different fault conditions.
Meanwhile, it is equally important to guarantee the detection security at the same time to avoid
the unnecessary blackouts.

4.13.1 INTEGRATED DETECTION ALGORITHM

In this section, two description methods for the features of HIAFs are presented,
including a global randomness index to reflect variations of the defined unified harmonic
energies, and a DWT-based method to extract distortions.

Global Randomness of Unified Harmonic Energy

 Harmonic is a useful characteristic quantity to describe the randomness generated by
HIAFs. In some traditional algorithms, randomness characteristic is extracted by finding the
point-to-point changes between the present harmonic amplitudes and historical values. However,
as mentioned in last Section, problems exist because HIAFs contacting various surfaces can
produce significantly different amplitudes of currents. As a result, if using amplitudes directly,
distinctions can hardly be found out between non-fault disturbances and the HIAFs with very
limited current.

It was proved that low order harmonics can make remarkable performances in reflecting
arc features. In this paper, 2nd-5th orders of harmonics are used to define a unified harmonic

energy:

where, Ek represents the harmonic energy of the k th cycle, and u Ek represents the unified
harmonic energy. ( ) k X i represents amplitude of the ith harmonic order. Superscript F marks
the variables after fault while N marks those at non-fault situations. kwindow represents the
length of detection window commonly set as 10~15 cycles. coeflim represents the limitation of
E1 in case of its extremely large values.

The comparisons of three different harmonic definitions from amplitudes to unified

harmonic energies by illustrating with the HIAFs in and two non-fault events that can affect the
zero sequence currents. One non-fault event is the capacitor switching, affecting the zero
sequence current due to its asynchronous operations. The other event is the motor start of wind
turbine which connects with the network through Yn-Yn wiring transformer for control
requirements so that the fluctuations caused by wind turbines can be transmitted across the
transformer and affect the zero sequence current at high-voltage side.

Based on the defined unified harmonic energy, a global randomness index RAND is
expressed as:
which reflects the general variation of unified harmonic energy within the detection window.

Fig 21 Comparisons of three harmonic definitions for HIAFs and non-fault events

4.14 WAVEFORM DISTORTION EXTRACTION

The global randomness index based on unified harmonic energy can perform well in
identifying the Type-2 HIAFs that begin with unstable arcing period, but obviously cannot
reliably settle the HIAFs coming into the stable arcing period early (Type-1 and some Type-3).
In this section, the discrete wavelet transform (DWT) is used to figure out this problem by
detecting HIAFs with waveform distortions. DWT is a tool which resolves signals into time and
frequency, making it convenient to analyze the local features with different frequency
components. For a sampling signal s[n] , its DWT is expressed as:
 where, ψ [n] is the wavelet base. The coefficient a and b are the scale and displacement
factors. A wavelet transformed signal in the selected frequency band [ f1, f 2] can be defined as:

The waveform distortion interval is generally named zero-off or zero-crossing period

(ZCP, normally two ZCPs per cycle). As Fig. shown, its wavelet transformed signal
reconstructed by band of 2k~3kHz presents the spike features during the ZCPs, which is
extracted and quantitatively described as follows:

where k represents the number of post-fault cycle, T N and 0 n respectively represent the
total number of sampling points in a cycle and the first point of each cycle. If the two
requirements in equation are met for several successive cycles (denoted as THRcount ,
commonly set as 4), the fault is believed to happen.

where THRR reflects the dominated ratio of amplitudes during ZCPs in a cycle, and
THRamp is to eliminate the cycles whose amplitudes during the ZCP is far below the average.
For a stable HIAF, its spikes of reconstructed signal are concentrated around the ZCPs, its k
RZCP during the stable cycles can thereby persist at higher values
 However, the effective description of this distortion index needs: 1) correct localization
of the distortion intervals, 2) good immunity against noise.

In numerical simulations, the relative positions of distortion intervals in a cycle are fixed
if using a certain arc model. Apparently, it is unrealistic for a real fault due to its varieties.
Therefore, a Distortion-Tracking method is proposed to locate the unpredictable distortion
intervals cycle by cycle. The method is illustrated with the zero sequence current and the ZCPs.
As shown in Fig. the relative position of each ZCP usually varies around the corresponding zero-
crossing point. As a result, the tracking range for each ZCP is with the length of 3T/4, and makes
zero-crossing point as its midpoint. In the tracking range, search for the interval with length of
T/4 which possesses the sharpest spikes (e.g. the maximal and this interval will be marked as the
ZCP of this zero-crossing point.

Noise immunity is significant for an algorithm applied in the distribution system.

Compared with the global randomness index that is based on the low-order harmonics, the
DWT-based extraction method for waveform distortion is usually more sensitive to the noise
even after filtering.

4.14.1 The Integrated Detection Flow

Based on the feature description methods, flow of the integrated detection algorithm for
HIAFs is shown in and the specific steps are instructed as follows

Step 1: Start. If a disturbance is detected, waveforms of zero sequence current will be recorded,
and filtered by wavelet filters with minimax thresholds mentioned

Step 2: Unified Harmonic Energy. Apply fast Fourier transform (FFT) to get harmonic
amplitudes of 2nd~5th orders and calculate the unified harmonic energy u Ek of each cycle
within the whole detection window according to equation
Step 3: Randomness Analysis (RA). Calculate the global randomness index as equation. Output
the result, which is ‘Y’ if the index reaches over the threshold, and otherwise, is ‘N’.

Step 4: Waveform Distortion Extraction (WDE). This step proceeds simultaneously with Step 2
after waveforms uploaded. Then, obtain the wavelet transformed signals, apply the noise
suppression and locate the distortion intervals. Describe the spike features as equation and if the
requirements in equation are satisfied for THR count cycles successively, a result of ‘Y’ is
output.

Step 5: Final Judgment. According to the result from two detection branches, the principle of
final judgment is also shown in Fig. HIAFs will be confirmed if the final results are ‘YY’, ‘YN’
or ‘NY’.

Fig.22 Flow of the integrated detection algorithm for HIAF.

4.1 POWER SUPPLY

BLOCK DIAGRAM

The ac voltage, classically 220V rms, is linked to a transformer, which steps with the
purpose of ac voltage down to the echelon of the desired dc output. A diode rectifier then
provides a full-wave rectified voltage that is initially filtered by a simple capacitor filter to
produce a dc voltage. This resulting dc voltage usually has some ripple or ac voltage variation.

A regulator circuit removes the ripples and also remains the same dc value even if the
input dc voltage varies, or the load connected to the output dc voltage changes. This voltage
regulation is usually obtained using one of the popular voltage regulator IC units.

TRANSFORMER RECTIFIER FILTER IC REGULATOR LOAD

Figure 2: Block diagram (Power supply)

4.1.1 WORKING PRINCIPLE

4.1.2 TRANSFORMER

The potential transformer will step down the power supply voltage (0-230V) to (0-6V)
level. Then the secondary of the potential transformer will be connected to the precision rectifier,
which is constructed with the help of op–amp. The advantages of using precision rectifier are it
will give peak voltage output as DC; rest of the circuits will give only RMS output.

4.1.3 BRIDGE RECTIFIER

When four diodes are connected as shown in figure, the circuit is called as bridge
rectifier. The input to the circuit is applied to the diagonally opposite corners of the network, and
the output is taken from the remaining two corners.

Let us assume that the transformer is working properly and there is a positive potential, at
point A and a negative potential at point B. the positive potential at point A will forward bias D3
and reverse bias D4.
 The negative potential at point B will forward bias D1 and reverse D2. At this time D3
and D1 are forward biased and will allow current flow to pass through them; D4 and D2 are
reverse biased and will block current flow.

The path for current flow is from point B through D1, up through RL, through D3,
through the secondary of the transformer back to point B. this path is indicated by the solid
arrows. Waveforms (1) and (2) can be observed across D1 and D3.

One-half cycle later the polarity across the secondary of the transformer reverse, forward
biasing D2 and D4 and reverse biasing D1 and D3. Current flow will now be from point A
through D4, up through RL, through D2, through the secondary of T1, and back to point A. This
path is indicated by the broken arrows. Waveforms (3) and (4) can be observed across D2 and
D4. The current flow through RL is always in the same direction. In flowing through RL this
current develops a voltage corresponding to that shown waveform. Since current flows through
the load during both half cycles of the applied voltage, this bridge rectifier is a full-wave
rectifier.

This may be shown by assigning values to some of the components shown in views A
and B. assume that the same transformer is used in both circuits. The peak voltage developed
between points X and y is 1000 volts in both circuits. In the conventional full-wave circuit
shown—in view A, the peak voltage from the center tap to either X or Y is 500 volts. Since only
one diode can conduct at any instant, the maximum voltage that can be rectified at any instant is
500 volts.

The maximum voltage that appears across the load resistor is nearly-but never exceeds-
500 v0lts, as result of the small voltage drop across the diode. In the bridge rectifier shown in
view B, the maximum voltage that can be rectified is the full secondary voltage, which is 1000
volts. Therefore, the peak output voltage across the load resistor is nearly 1000 volts. With both
circuits using the same transformer, the bridge rectifier circuit produces a higher output voltage
than the conventional full-wave rectifier circuit.

4.1.4 IC VOLTAGE REGULATORS

 Voltage regulators comprise a class of widely used ICs. Regulator IC units
contain the circuitry for reference source, comparator amplifier, control device, and overload
protection all in a single IC. IC units provide regulation of either a fixed positive voltage, a fixed
negative voltage, or an adjustably set voltage. The regulators can be selected for operation with
load currents from hundreds of milli amperes to tens of amperes, corresponding to power ratings

from milli watts to tens of watts.

Figure 3: Circuit diagram (Power supply)

A fixed three-terminal voltage regulator has an unregulated dc input voltage, Vi, applied
to one input terminal, a regulated dc output voltage, Vo, from a second terminal, with the third
terminal connected to ground.

The series 78 regulators provide fixed positive regulated voltages from 5 to 24 volts.
Similarly, the series 79 regulators provide fixed negative regulated voltages from 5 to 24 volts.

 For ICs, microcontroller, LCD --------- 5 volts

 For alarm circuit, op-amp, relay circuits ---------- 12 volts

DISCUSSION
 The +5 volt power supply is based on the commercial 7805 voltage regulator IC. This
simplifies the design and layout of the circuit considerably, because all of the regulating circuitry
as well as current limiters and overload protection are built into the IC. As a result, little is
needed in the way of support circuitry.
We do still need the external capacitors. One thing that is very difficult to achieve in ICs
is a capacitor of high capacitance value. Therefore, the electrolytic capacitors must be provided
to work with the IC. The disc ceramic capacitor must also be of a higher value than is readily
obtainable within an IC, so it, too, must be provided externally.
The 7805 voltage regulator IC is capable of handling load currents up to an ampere or so.
However, the IC will dissipate a fair amount of heat when the load current gets this high.
Without a heat sink, the IC will get hot and shut itself down at load currents above about 150
mA. If you add a heat sink for a TO-220 case (available at Radio Shack), this power supply can
easily deliver an ampere or more to its load

ADVANTAGE
 Small size and less weight.
 Less expensive
 High reliable and versatile
 Reduced man power
APPLICATION

It is typically used in electronic devices such as switched-mode power supply, linear

regulator, rectifier and inverter (electrical).

MICROCONTROLLER

 A microcontroller is a very small computer on a single integrated circuit (IC). All

computers have several things in common:
 • CPU (central processing unit) that executes programs.
 • RAM (random-access memory) where it can store variables.
 • Input and output devices for interaction.
 Desktop computers are “general purpose computers” that can run any of thousands of
programs. Microcontrollers are “special purpose computers.” Microcontrollers do one
thing well. Microcontrollers are often embedded onto a single printed circuit board.
This board provides all of the circuitry necessary for a useful control task. The
intention is that the board is immediately useful to an application developer, without
them needing to spend time and effort in developing the controller hardware.


 ARDUINO UNO
 The Arduino Uno is a microcontroller board based on the ATmega328 (datasheet). It
has 14 digital input/output pins (of which 6 can be used as PWM outputs), 6 analog
inputs, a 16 MHz ceramic resonator, a USB connection, a power jack, an ICSP
header, and a reset button. It contains everything needed to support the
microcontroller; simply connect it to a computer with a USB cable or power it with a
AC-to-DC adapter or battery to get started. The Uno differs from all preceding boards
in that it does not use the FTDI USB-to-serial driver chip. Instead, it features the
Atmega16U2 (Atmega8U2 up to version R2) programmed as a USB-to-serial
converter
 "Uno" means one in Italian and is named to mark the upcoming release of Arduino
1.0. The Uno and version 1.0 will be the reference versions of Arduino, moving
forward. The Uno is the latest in a series of USB Arduino boards, and the reference
model for the Arduino platform

 Features
  1.0 pinout: added SDA and SCL pins that are near to the AREF pin and two other
new pins placed near to the RESET pin, the IOREF that allow the shields to adapt to
the voltage provided from the board. In future, shields will be compatible both with
the board that use the AVR, which operate with 5V and with the Arduino Due that
operate with 3.3V. The second one is a not connected pin, that is reserved for future
purposes.
  Stronger RESET circuit.
  Atmega 16U2 replace the 8U2.
 AURDINO PIN DESCRIPTION


 Power
 The Arduino Uno can be powered via the USB connection or with an external power
supply. The power source is selected automatically. External (non-USB) power can
come either from an AC-to-DC adapter (wall-wart) or battery. The adapter can be
connected by plugging a 2.1mm centre-positive plug into the board's power jack.
Leads from a battery can be inserted in the Gnd and Vin pin headers of the POWER
connector. The board can operate on an external supply of 6 to 20 volts. If supplied
with less than 7V, however, the 5V pin may supply less than five volts and the board
may be unstable. If using more than 12V, the voltage regulator may overheat and
damage the board. The recommended range is 7 to 12 volts. The power pins are as
follows:
  VIN. The input voltage to the Arduino board when it's using an external power
source (as opposed to 5 volts from the USB connection or other regulated power
 source). You can supply voltage through this pin, or, if supplying voltage via the
power jack, access it through this pin.
  5V.This pin outputs a regulated 5V from the regulator on the board. The board can
be supplied with power either from the DC power jack (7 - 12V), the USB connector
(5V), or the VIN pin of the board (7-12V). Supplying voltage via the 5V or 3.3V pins
bypasses the regulator, and can damage your board. We don't advise it.
  3V3. A 3.3 volt supply generated by the on-board regulator. Maximum current
draw is 50 mA.
  GND. Ground pins.


 AURDINO PIN LAYOUT


 Memory
 The ATmega328 has 32 KB (with 0.5 KB used for the bootloader). It also has 2 KB
of SRAM and 1 KB of EEPROM
 Input and Output
 Each of the 14 digital pins on the Uno can be used as an input or output, using
pinMode(), digitalWrite(), and digitalRead() functions. They operate at 5 volts. Each
pin can provide or receive a maximum of 40 mA and has an internal pull-up resistor
(disconnected by default) of 20-50 kOhms. In addition, some pins have specialized
functions:
  Serial: 0 (RX) and 1 (TX). Used to receive (RX) and transmit (TX) TTL serial
data. These pins are connected to the corresponding pins of the ATmega8U2 USB-to-
TTL Serial chip.
  External Interrupts: 2 and 3. These pins can be configured to trigger an interrupt on
a low value, a rising or falling edge, or a change in value. See the attachInterrupt()
function for details.
  PWM: 3, 5, 6, 9, 10, and 11. Provide 8-bit PWM output with the analogWrite()
function.
  SPI: 10 (SS), 11 (MOSI), 12 (MISO), 13 (SCK). These pins support SPI
communication using the SPI library.
  LED: 13. There is a built-in LED connected to digital pin 13. When the pin is
HIGH value, the LED is on, when the pin is LOW, it's off.
 The Uno has 6 analog inputs, labelled A0 through A5, each of which provide 10 bits
of resolution (i.e. 1024 different values). By default they measure from ground to 5
volts, though is it possible to change the upper end of their range using the AREF pin
and the analogReference() function. Additionally, some pins have specialized
functionality:
 TWI: A4 or SDA pin and A5 or SCL pin. Support TWI communication using the
Wire library
 There are a couple of other pins on the board:
 AREF. Reference voltage for the analog inputs. Used with analogReference().
 Reset. Bring this line LOW to reset the microcontroller. Typically used to add a reset
button to shields which block the one on the board.
 Communication
 The Arduino Uno has a number of facilities for communicating with a computer,
another Arduino, or other microcontrollers. The ATmega328 provides UART TTL
(5V) serial communication, which is available on digital pins 0 (RX) and 1 (TX). An
ATmega16U2 on the board channels this serial communication over USB and
appears as a virtual com port to software on the computer. The '16U2 firmware uses
the standard USB COM drivers, and no external driver is needed. However, on
Windows, a .inf file is required. The Arduino software includes a serial monitor
which allows simple textual data to be sent to and from the Arduino board. The RX
and TX LEDs on the board will flash when data is being transmitted via the USB-to-
serial chip and USB connection to the computer (but not for serial communication on
pins 0 and 1). A Software Serial library allows for serial communication on any of the
Uno's digital pins. The ATmega328 also supports I2C (TWI) and SPI communication
 Automatic (Software) Reset
 Rather than requiring a physical press of the reset button before an upload, the
Arduino Uno is designed in a way that allows it to be reset by software running on a
connected computer. One of the hardware flow control lines (DTR) of the
ATmega8U2/16U2 is connected to the reset line of the ATmega328 via a 100
nanofarad capacitor. When this line is asserted (taken low), the reset line drops long
enough to reset the chip. The Arduino software uses this capability to allow you to
upload code by simply pressing the upload button in the Arduino environment. This
means that the boot loader can have a shorter timeout, as the lowering of DTR can be
well-coordinated with the start of the upload. This setup has other implications. When
the Uno is connected to either a computer running Mac OS X or Linux, it resets each
time a connection is made to it from software (via USB). For the following half
second or so, the boot loader is running on the Uno. While it is programmed to ignore
malformed data (i.e. anything besides an upload of new code), it will intercept the
first few bytes of data sent to the board after a connection is opened. If a sketch
running on the board receives one-time configuration or other data when it first starts,
 make sure that the software with which it communicates waits a second after opening
the connection and before sending this data. The Uno contains a trace that can be cut
to disable the auto-reset. The pads on either side of the trace can be soldered together
to re-enable it. It's labelled "RESET-EN". You may also be able to disable the auto-
reset by connecting a 110 ohm resistor from 5V to the reset line; see this forum thread
for details.

USB Over current Protection

 The Arduino Uno has a resettable polyfuse that protects your computer's USB ports
from shorts and over current. Although most computers provide their own internal
protection, the fuse provides an extra layer of protection. If more than 500 mA is
applied to the USB port, the fuse will automatically break the connection until the
short or overload is removed.

.3.1 Contact Sensors

Contact temperature sensors measure the temperature of the object to which the sensor is in
contact by assuming or knowing that the two (sensor and the object) are in thermal equilibrium,
in other words, there is no heat flow between them.

 Thermocouples
 Resistance Temperature Detectors (RTDs)
 Full System Thermometers
 Bimetallic Thermometers

4.3.2 Noncontact Sensors

Most commercial and scientific noncontact temperature sensors measure the thermal radiant
power of the Infrared or Optical radiation received from a known or calculated area on its surface
or volume within it.
An example of noncontact temperature sensors is a pyrometer, which is described into further
detail at the bottom of this section.

4.3.3 Humidity Sensor

Thermometers are the most common temperature sensors encountered in simple, everyday
measurements of temperature. Two examples of thermometers are the Filled System and Bimetal
thermometers.

Filled System Thermometer

The familiar liquid thermometer consists of a liquid enclosed in a tube. The volume of the fluid
changes as a function of temperature. Increased molecular movement with increasing
temperature causes the fluid to expand and move along calibrated markings on the side of the
tube. The fluid should have a relatively large thermal expansion coefficient so that small changes
in temperature will result in detectable changes in volume. A common tube material is glass and
a common fluid is alcohol. Mercury used to be a more common fluid until its toxicity was
realized. Although the filled-system thermometer is the simplest and cheapest way to measure
temperature, its accuracy is limited by the calibration marks along the tube length. Because filled
system thermometers are read visually and don’t produce electrical signals, it is difficult to
implement them in process controls that rely heavily on electrical and computerized control.

4.3.4 RTD Structure

A schematic diagram of a typical RTD is shown in Figure 1.

Figure 8 Schematic Diagram of Resistance Temperature Structure

As shown in Figure 1, the RTD contains an outer sheath to prevent contamination from the
surrounding medium. Ideally, this sheath is composed of material that efficiently conducts heat
to the resistor, but resists degradation from heat or the surrounding medium.

The resistance sensor itself is responsible for the temperature measurement, as shown in the
diagram. Sensors are most commonly composed of metals, such as platinum, nickel, or copper.
The material chosen for the sensor determines the range of temperatures in which the RTD could
be used. For example, platinum sensors, the most common type of resistor, have a range of
approximately -200°C – 800°C. (A sample of the temperature ranges and resistances for the most
common resistor metals is shown in Table 1). Connected to the sensor are two insulated
connection leads. These leads continue to complete the resistor circuit.

Table 1. Common Metal Temperature and Resistance Ranges

Element Temperature Base

TCR(Ω/Ω/°C)
Metal Range Resistance

Copper -100 – 260 °C 10 Ω at 0 °C 0.00427

Nickel -100 – 260 °C 120 Ω at 0 °C 0.00672

Platinum -260 – 800 °C 100 Ω at 0 °C 0.003916

There are 4 major categories of RTD sensors. There are carbon resistors, film thermometers,
wire-wound thermometers and coil elements. Carbon resisters are the most commonly used.
They are inexpensive and are accurate for low temperatures. They also are not affected by
hysteresis or strain gauge effects. They are commonly used by researchers. Film thermometers
have a very thin layer of metal, often platinum, on a plate. This layer is very small, on the
micrometer scale. These thermometers have different strain gauge effects based on what the
metal and plates are composed of. There is also stability problems that are dependent on the
components used. In wire-wound thermometers the coil gives stability to the measurement. A
larger diameter of the coil adds stability, but it also increases the amount the wire can expand
which increases strain and drift. They have very good accuracy over a large temperature range.
Coil elements are similar to wire-wound thermometers and have generally replaced them in all
industrial applications. The coil is allowed to expand over large temperature ranges while still
giving support. This allows for a large temperature range while decreasing the drift.

4.3.5 Thermocouples

Another temperature sensor often used in industry is the thermocouple. Among the various
temperature sensors available, the thermocouple is the most widely used sensor. Similar to the
RTD, the thermocouple provides an electrical measurement of temperature.

4.3.6 Thermocouple Operation

The main principle upon which the thermocouple function is based on is the difference in the
conductivities of the two wire materials that the thermocouple is made of, at a given temperature.
This conductivity difference increases at higher temperatures and conversely, the conductivity
difference decreases at lower temperatures. This disparity results in the thermocouples being
more efficient and useful at higher temperatures. Since the conductivity difference is small at
lower temperatures and thus more difficult to detect, they are inefficient and highly unreliable at
low temperatures.

The conductivity difference between the two wires, along with a temperature
difference between the two junctions, creates an electrical current that flows through the
thermocouple. The first junction point, which is the point at which the two wires are connected,
is placed within the medium whose temperature is being measured. The second junction point is
constantly held at a known reference temperature. When the temperature of the medium differs
from the reference temperature, a current flows through the circuit. The strength of this current is
based upon the temperature of the medium, the reference temperature, and the materials of the
metal wires. Since the reference temperature and materials are known, the temperature of the
medium can be determined from the current strength.

Error associated with the thermocouple occurs at lower temperatures due to the
difficulty in detecting a difference in conductivities. Therefore, thermocouples are more
commonly used at higher temperatures (above -125°C) because it is easier to detect differences
in conductivities. Thermocouples are operable over a wide range of temperatures, from -200°C
to 2320°C, which indicates its robustness and vast applications. Thermocouples operate over this
wide range of temperatures, without needing a battery as a power source. It should be noted that,
the wire insulation might wear out over time by heavy use, thus requiring periodical checks and
maintenance to preserve the accuracy of the thermocouple.

ESP8266 WIFI DEVICE

In this modern world, the Wi-Fi technology has removed the distances, in which the electronic
devices are used, which are connected to the wireless Lan (WLAN) network, by using at 2.4
gigahertz or 5 gigahertz radio frequency bands. The Lan network provides the facility of
password protection and also allows the device to use the network which is within the range of
that device. In this advance technology different types of modules are used, which have different
features. Here we would explain the ESP8266 WI-FI module is details but before explaining
these types of module, first of all we would explain the brief history of Wi-Fi.

The history of WI-FI was started in 1985, when the united states FCC used the 900 MHZ
frequency band without license. Because the frequency band of this technology was very low
therefore in beginning, this technology was only assumed for home appliances such as
microwave oven. After that, in 1991 the NCR and AT&T corporation invented a first wireless
product that’s name was Wave Lan, which was used for cashier system. In this fragmented
market the research was continued, finally the first WI- FI system was introduced in 1997, which
was the prototype communication system.

The ESP8266 WI-FI module is basically, a complete WI-FI solution, which have self-contained
operating system and integrated TCP/IP protocol stack that can be easily connect to the
microcontroller for gaining the access to any Wi-Fi network This module has the capability of
either to give or gain the applications and functions from other module or processor, mean’s that
this also have the facility of hosting or uploading the functions and applications. For connecting
this module to any WI-FI network, just up load the pre-programmed program which is set in this
module as firmware. This is very cost effective module with an ever growing and huge
communication community.

This module has powerful storage and data processing capability therefore it can easily integrate
with sensors and others processing units. This is high degree on chip integration device which
has a very low PCB circuitry area. It is supported with IPSD for output applications and also
interfaced with Bluetooth. This module does not require any external RF signal because it have
self-contained calibrated signal. The ESP 8266 WI-FI module is shown in figure below

ESP8266 WI-FI Module

The ESP8266 WI-FI module is basically, a complete WI-FI solution, which have self-contained
operating system and integrated TCP/IP protocol stack that can be easily connect to the
microcontroller for gaining the access to any Wi-Fi network This module has the capability of
either to give or gain the applications and functions from other module or processor, mean’s that
this also have the facility of hosting or uploading the functions and applications. For connecting
this module to any WI-FI network, just up load the pre-programmed program which is set in this
module as firmware. This is very cost effective module with an ever growing and huge
communication community.

This module has powerful storage and data processing capability therefore it can easily integrate
with sensors and others processing units. This is high degree on chip integration device which
has a very low PCB circuitry area. It is supported with IPSD for output applications and also
interfaced with Bluetooth. This module does not require any external RF signal because it have
self-contained calibrated signal. The ESP 8266 WI-FI module is shown in figure below esp8266
wifi module

Pin Configuration of ESP 8266 WI-FI Module

The ESP8266 WI-FI module consists of two rows of eight pins. The pin configuration diagram is
shown in figure below. According to the figure 2 the pin 1, which is GND the ground pin which
is directly connected to the ground for power on this module. The pins 2 and 3, which are the
GPIO 2 and GPIO 0, these pins decided in which mode the module would be start-up, in other
words these are mode selected pins. The pins 4 and 5, which are RX and TX, these pin are used
for communication purposes and program the module. The pin 6 which is CH_PD, it is called
chip power down pin. The pin 7 which is RST pin and this pin is used for reset the module. The
pin 8 which is VCC pin and is used to power on the module. The operating voltage of this
module is 3.3 volts, which are directly connected to the ground pin (GND) and VCC pin.

ESP8266-01 Module Pin Description

Pin Configuration of ESP 8266 WI-FI Module

The ESP8266 WI-FI module consist of two rows of eight pins. The pin configuration diagram is
shown in figure belowesp8266 pinout

According to the figure 2 the pin 1, which is GND the ground pin which is directly connected to
the ground for power on this module. The pins 2 and 3, which are the GPIO 2 and GPIO 0, these
pins decided in which mode the module would be start-up, in other words these are mode
selected pins. The pins 4 and 5, which are RX and TX, these pin are used for communication
purposes and program the module. The pin 6 which is CH_PD, it is called chip power down pin.
The pin 7 which is RST pin and this pin is used for reset the module. The pin 8 which is VCC
pin and is used to power on the module. The operating voltage of this module is 3.3 volts, which
are directly connected to the ground pin(GND) and VCC pin.

Instruction Commands of ESP 8266 Module

The wireless WI-FI ESP 8266 module is drive through serial interfacing with Arduino or any
type of modules. The standards AT commands are used for these purposes, which are almost
twelve.

(1) AT: This type of command is used to test the startup function of WI-FI module. The response
would be ok, against this command if everything is ok.

(2) AT+RST: This type of command is used for reset the WI-FI module when it is in working
condition. The response would be ok, when reset the module.

(3)AT+GMR: This type of AT command is used to check the version of AT command and we
used SDK version of AT command in this type of WI-FI module.

(4) AT+GSLP: This type of AT command is used to invoke the module from sleep mode, in
other words this command is used for wake -up the module.

(5) ATE:This type of command is used for echo sound, means that, when this command is
entered then the echo send back to the sender.
(6) AT+RESTORE: This type of command is used for restore factory setting means, when this
command is entered then all the parameters are reset automatically.

(7) AT+UART: This type of command is used for set the UART configuration and writes the
new configuration in flash means, it is used for setting the default parameters.

(8) AT+UART_CUR: This type of command is used for set the current UART configuration and
does not write to the flash.

4.6 RELAY

Fig 14 Relay

A relay is an electrically operated switch. Many relays use an electromagnet to operate a

switching mechanism mechanically, but other operating principles are also used. Relays are used
where it is necessary to control a circuit by a low-power signal with complete electrical isolation
between control and controlled circuits or where several circuits must be controlled by one
signal. The first relays were used in long distance telegraph circuits, repeating the signal coming
in from one circuit and re-transmitting it to another. Relays were used extensively in telephone
exchanges and early computers to perform logical operations. A type of relay that can handle the
high power required to directly control an electric motor or other loads is called a contractor.
Solid-state relays control power circuits with no moving parts, instead using a semiconductor
device to perform switching. Relays with calibrated operating characteristics and sometimes
multiple operating coils are used to protect electrical circuits from overload or faults; in modern
electric power systems these functions are performed by digital instruments still called
"protective relays".
4.6.1 BASIC DESIGN AND OPERATION

Fig 15 Relay circuit diagram

A simple electromagnetic relay consists of a coil of wire wrapped around a soft iron core, an iron
yoke which provides a low reluctance path for magnetic flux, a movable iron armature, and one
or more sets of contacts. The armature is hinged to the yoke and mechanically linked to one or
more sets of moving contacts. It is held in place by a spring so that when the relay is de-
energized there is an air gap in the magnetic circuit. In this condition, one of the two sets of
contacts in the relay pictured is closed, and the other set is open. Other relays may have more or
fewer sets of contacts depending on their function. The relay in the picture also has a wire
connecting the armature to the yoke. This ensures continuity of the circuit between the moving
contacts on the armature, and the circuit track on the printed circuit board via the yoke, which is
soldered to the PCB.

When an electric current is passed through the coil it generates a magnetic field that activates the
armature, and the consequent movement of the movable contact(s) either makes or breaks a
connection with a fixed contact. If the set of contacts was closed when the relay was de-
energized, then the movement opens the contacts and breaks the connection, and vice versa if the
contacts were open. When the current to the coil is switched off, the armature is returned by a
force, approximately half as strong as the magnetic force, to its relaxed position. Usually this
force is provided by a spring, but gravity is also used commonly in industrial motor starters.
Most relays are manufactured to operate quickly. In a low-voltage application this reduces noise;
in a high voltage or current application it reduces arcing.
4.6.2 APPLICATION

Relays are used for:

 Amplifying a digital signal, switching a large amount of power with a small operating
power. Some special cases are:
o A telegraph relay, repeating a weak signal received at the end of a long wire
o Controlling a high-voltage circuit with a low-voltage signal, as in some types of
modems or audio amplifiers,
o Controlling a high-current circuit with a low-current signal, as in the starter
solenoid of an automobile,
 Detecting and isolating faults on transmission and distribution lines by opening and
closing circuit breakers

4.7 ALARM

An alarm device or system of alarm devices gives an audible, visual or

other form of alarm signal about a problem or condition.

Fig 16 Alarm circuit diagram

 Burglar alarms designed to warn of burglaries; this is often a silent alarm: the police or
guards are warned without indication to the burglar, which increases the chances of catching him
or her.

 Alarm clocks can produce an alarm at a given time

 Distributed control manufacturing systems or DCSs, found in nuclear power plants, refineries
and chemical facilities also generate alarms to direct the operator's attention to an important
event that he or she needs to address.

 Alarms in an operation and maintenance (O&M) monitoring system, which informs the bad
working state of the system under monitoring.

Fig 17 Buzzer

4.8 LCD

A liquid crystal display is a flat panel display, electronic visual display, or video display that uses
the light modulating properties of liquid crystals. Liquid crystals do not emit light directly. LCDs
are available to display arbitrary images or fixed images which can be displayed or hidden, such
as preset words, digits, and 7-segment displays as in a digital clock. They use the same basic
technology, except that arbitrary images are made up of a large number of small pixels, while
other displays have larger elements. LCDs are used in a wide range of applications
including computer monitors, televisions, instrument panels, aircraft cockpit displays, and
signage. They are common in consumer devices such as video players, gaming
devices, clocks, watches, calculators, and telephones, and have replaced cathode ray
tube displays in most applications. They are available in a wider range of screen sizes than CRT
and plasma displays, and since they do not use phosphors, they do not suffer image burn-in.
LCDs are, however, susceptible to image persistence.
 Fig 18 LCD display unit

4.8.1 SPECIFICATION:

Important factors to consider when evaluating an LCD:

RESOLUTION VERSUS RANGE

Fundamentally resolution is the granularity with which a performance feature of the display is
divided. Resolution is often confused with range or the total end-to-end output of the display.
Each of the major features of a display has both a resolution and a range that are tied to each
other but very different. Frequently the range is an inherent limitation of the display while the
resolution is a function of the electronics that make the display work.

SPATIAL PERFORMANCE

LCDs come in only one size for a variety of applications and a variety of resolutions within each
of those applications. LCD spatial performance is also sometimes described in terms of a "dot
pitch". The size of an LCD is always described in terms of the diagonal distance from one corner
to its opposite. This is an historical remnant from the early days of CRT television when CRT
screens were manufactured on the bottoms of glass bottles, a direct extension of cathode ray
tubes used in oscilloscopes. The diameter of the bottle determined the size of the screen. Later,
when televisions went to a squarer format, the square screens were measured diagonally to
compare with the older round screen.
CONCLUSION

An improved arc model is proposed, which can realize the description of randomness and
intermittence during the unstable arcing period of HIAFs. The suggested parameters of the arc
model are presented. With the classification and feature analyses for different HIAFs, an
integrated detection algorithm is proposed. In the algorithm, the defined unified harmonic energy
can unify the quantitative descriptions for HIAFs under different situations, and enlarge the
difference between faults and non-fault events. Subsequently, the global randomness index based
on the unified harmonic energy and the feature extraction method based on DWT can
respectively perform well in detecting different types of HIAFs, while their combination can
ensure the greater reliability and security at the same time as the verification by comparing with
other algorithms shows. The practical application of the proposed algorithm in a D-PMU based
system is also introduced.
REFERENCE

[1] Fang, W.; Xue, F.; Ding, Y.; Xiong, N.; Leung, V.C. EdgeKE: An on-demand deep learning
IoT system for cognitive big data on industrial edge devices. IEEE Trans. Ind. Inform. 2020, 17,
6144–6152.

[2] Fazio, M.; Ranjan, R.; Girolami, M.; Taheri, J.; Dustdar, S.; Villari, M. A note on the
convergence of IoT, edge, and cloud computing in smart cities. IEEE Cloud Comput. 2018, 5,
22–24.

[3] Javaid, A.; Javaid, N.; Wadud, Z.; Saba, T.; Sheta, O.E.; Saleem, M.Q.; Alzahrani, M.E.
Machine learning algorithms and fault detection for improved belief function based decision
fusion in wireless sensor networks. Sensors 2019, 19, 1334.

[4] Santo, Y.; Dalmazo, B.L.; Immich, R.; Riker, A. On the Performance of Machine Learning at
the Network Edge to Detect Industrial IoT Faults. In Proceedings of the 2022 IEEE 21st
International Symposium on Network Computing and Applications (NCA), Boston, MA, USA,
14–16 December 2022; Volume 21, pp. 291–295.

[5] Jan, S.U.; Lee, Y.D.; Koo, I.S. A distributed sensor-fault detection and diagnosis framework
using machine learning. Inf. Sci. 2021, 547, 777–796.

[6] Jan, S.U.; Lee, Y.D.; Shin, J.; Koo, I. Sensor fault classification based on support vector
machine and statistical time-domain features. IEEE Access 2017, 5, 8682–8690.

[7] Saeed, U.; Jan, S.U.; Lee, Y.D.; Koo, I. Machine learning-based real-time sensor drift fault
detection using Raspberry PI. In Proceedings of the 2020 International Conference on
Electronics, Information, and Communication (ICEIC), Barcelona, Spain, 19–22 January 2020;
pp. 1–7.

[8] Zidi, S.; Moulahi, T.; Alaya, B. Fault detection in wireless sensor networks through SVM
classifier. IEEE Sens. J. 2017, 18, 340–347.

[9] Suthaharan, S.; Alzahrani, M.; Rajasegarar, S.; Leckie, C.; Palaniswami, M. Labelled data
collection for anomaly detection in wireless sensor networks. In Proceedings of the 2010 Sixth
International Conference on Intelligent Sensors, Sensor Networks and Information Processing,
Brisbane, QLD, Australia, 7–10 December 2010; pp. 269–274.

[10] Naimi, A.; Deng, J.; Shimjith, S.; Arul, A.J. Fault Detection and Isolation of a Pressurized
Water Reactor Based on Neural Network and K-Nearest Neighbor. IEEE Access 2022, 10,
17113–17121.

[11] Khodabakhsh, A.; Ari, I.; Bakir, M.; Ercan, A.O. Multivariate sensor data analysis for oil
refineries and multi-mode identification of system behavior in real-time. IEEE Access 2018, 6,
64389–64405.

P12[ Tupras Refineries Dataset. Available online: https://www.openml.org/d/41170 (accessed on

1 October 2022).

[13] Dofe, J.; Nguyen, A.; Nguyen, A. Unified Countermeasures against Physical Attacks in
Internet of Things-A survey. In Proceedings of the 2021 IEEE International Symposium on
Smart Electronic Systems (iSES) (Formerly iNiS), Jaipur, India, 18–22 December 2021; pp.
194–199.

[14] Mustafa, M.A.; Konios, A.; Garcia-Constantino, M. IoT-Based Activities of Daily Living
for Abnormal Behavior Detection: Privacy Issues and Potential Countermeasures. IEEE Internet
Things Mag. 2021, 4, 90–95.

[15] Li, Y.; Dong, M.; Kothari, R. Classifiability-based omnivariate decision trees. IEEE Trans.
Neural Netw. 2005, 16, 1547–1560.

[16] D. B. Russell. "Expert system for detecting high impedance faults." U.S. Patent No.
5,550,751. 27 Aug. 1996.

[17] M. Michalik, M. Lukowicz, W. Rebizant, et al., "New ANN-Based Algorithms for

Detecting HIFs in Multigrounded MV Networks," IEEE Trans. Power Del., vol. 23, no. 1, pp.
58-66, Jan. 2008.
[18] P. H. Schavemaker and L. van der Slui, "An improved Mayr-type arc model based on
current-zero measurements [circuit breakers]," IEEE Trans. Power Del., vol. 15, no. 2, pp. 580-
584, Apr 2000.