Distribution Transformer Health Monitoring 1

First A. Author, Fellow, IEEE, Second B. Author, and Third C. Author, Jr., Member, IEEE
Design of IoT based Transformer Health
Monitoring System (IoT - THMS) for a
Distribution Transformer
processes [2]. The paper explores power transformer
Abstract—The transformers are pivotal components of the insulation system reactions like oxidation, hydrolysis,
electrical grid, responsible for voltage transformation, and pyrolysis, partial discharge, and arcing. Moreover, it analyzed
thus play a crucial role in efficient energy distribution. their mechanisms, conditions, and relationships and discussed
However, like all machinery, they are susceptible to state-of-the-art methods for monitoring byproducts,
potential failures that can have severe consequences. The overcoming cost and time constraints. Thereby, it evaluated
monitoring and management of distribution the challenges and prospects for improving efficiency [3].
transformers face challenges with manual data entry and A high-performance piezoelectric vibration energy
handwritten documentation, impacting efficiency. This harvester (PVEH) for wireless monitoring in smart grids is
paper proposed an IoT based Transformer Health discussed in [4]. Also, it achieved 0.7 V output, a 33.3%
Monitoring System (IoT - THMS) for a distribution reduction in charging time, and 50% increased efficiency,
transformer. Transitioning to digital data collection aiming for commercialization. Therefore, it aimed to
methods, like mobile applications, can mitigate errors. accelerate the commercialization of PHEV with the grid
Establishing a digital database addresses the lack of system [4]. The transformer fault diagnosis method, using an
historical data, enabling trend analysis. Real-time IoT-based monitoring system and ensemble machine learning
monitoring systems replace delayed issue detection with is proposed in [5], which effectively collects vibration signals
immediate alerts for timely responses. Investing in and improves the RVM (Relevance vector machine) classifier
predictive maintenance technologies transforms the compared to other classifiers. The paper represents the
difficulty of predicting failures into proactive maintenance development of monitoring systems and diagnostic approaches
opportunities. Introducing data logging systems turns the to increase the lifespan of power transformers and reduce
absence of recorded critical parameters into valuable catastrophic failures. It discussed the use of machine learning
resources for analyzing transformer health. The as a solution, highlighting its advantages and disadvantages. In
overarching research domain involves enhancing power addition, the study presents intelligent fault diagnosis methods
distribution networks through the incorporation of
and algorithms to evaluate transformer conditions [6].
onboard sensors intended for monitoring status of the grid
A deep learning neural network (DLNN) for
in real time. The primary focus is on developing
detecting the health of power transformers is introduced in [7].
sophisticated sensors capable of estimating the health of
Furthermore, the network used echo state networks, IDRSNs,
transformers, thereby contributing to the overall
and a one-dimension convolution neural network to extract
intelligence and reliability of the grid infrastructure and
significant features. Thus, the method outperforms the latest
aim to enable effective strategies for mitigating aging
neural networks and power transformer health assessment
related challenges in power distribution transformers
methods [7]. Thus, predictive maintenance using machine
learning (ML) could minimize downtime and losses by
Index Terms—Internet of Things based Transformer
monitoring machinery condition. However, the effectiveness
Health Monitoring System, Message Queuing Telemetry
Transport
depends on selecting the right data processing method and ML
model. This survey is highlighted the importance of
considering all factors and evaluating the trade-off between
task constraints and diagnostic techniques for optimal
I. INTRODUCTION
solutions [8].

T ransformer is the crucial equipment which transmits

electrical power in different voltage levels. Normally, it
faces many faults due to non-maintenance of various parts,
Another study is introduced a novel deep learning
architecture to forecast temperature for transformer oil, target
to improve efficiency and prevent high temperatures.
which leads to breakdown and life time reduction. The Additionally, the architecture used a Smooth Residual Block
monitoring of transformer condition and investigation of to balance inductive bias and applied to Electricity
alternative dielectric fluids, claiming excellent insulation Transformer datasets where the results showed a 13%
lifecycle, better compatibility, and improved performance improvement in mean square error (MSE) and a 57%
compared to traditional mineral oil base fluids are discussed in improvement in performance compared to the best current
[1]. The study developed a fault-sensitive smart asset architectures [9]. A novel machine learning approach using
management system for power transformers, using LabVIEW neural networks is presented and ideal structures for BLEVE
for compatibility and achieving high accuracy in diagnosing loads are predicted. The approach examined conventional
intentional faults through online and offline verification models, multi-layer perceptron (MLP) by 20%. Therefore, the
Distribution Transformer Health Monitoring 2

model predicts pressure and impulse with less than 14% module to transmit the
relative errors, and MLP by 20% [10]. readings. monitoring.
The transformer model is demonstrated as an The system was
effective tool for blast loading prediction, outperformed the GSM-based system
not tested in
integrated with PIC
multi-layer perceptron approach in BLEVE pressure Transformer microcontroller to
operational
prediction with a relative error of 3.5% and R 2 0.997, and scenarios and
[ 20] health monitor load currents,
does not utilize
highlighting its potential in engineering applications [11]. monitoring over voltage, transformer
machine
Moreover, AnoFed is proposed as a novel framework that oil level and oil
learning for
temperature
combines transformer-based AE and VAE with Support fault detection
Vector Data Description in a federated setting. This Observes temperature,
voltage, and current of
framework enhanced privacy protection, improves outcomes, Transformer
the transformer. PIC
Does not utilize
and supports adaptive anomaly detection in healthcare health machine
[ 21] microcontroller to
monitoring learning for
applications. Thereby, it showed that the AnoFed is effective using IoT
interface with sensors
fault detection.
and efficient and made suitable for resource-constrained edge and data sent through
GSM-enabled network.
devices [12]. Monitors current,
The absence of a structured dataset indeed impacts voltage, temperature,
the efficiency and lifespan of the transformers. Handwritten and humidity data. The
documents can be prone to errors during data entry. By IoT-based sensors were interfaced Does not utilize
transformer with Node MCU machine
implementing digital data collection methods will reduce [ 22]
monitoring and microcontroller. learning for
transcription errors. This can involve using mobile protection Buzzers, LEDs, and web fault detection.
applications or other digital tools for meter reading [13]. application provide cost-
Without a dataset, it's challenging to analyze historical effective, user-friendly,
and remote monitoring
performance trends. By maintaining a digital database to store Node MCU
historical data, allowing for trend analysis, predictive Thermal Reported delays
microcontroller was used
monitoring and during testing
maintenance, and performance optimization lead increase of protection
to interface temperature,
and does not
transformer life. Relying on manual processes might result in [23] humidity, and current
system for utilize machine
sensor. Cooling system
delayed detection of potential problems. The real time transformers
and tripping mechanism
learning for
monitoring systems can receive immediate alerts about using IoT fault detection
were introduced.
anomalies or malfunctions, enabling quick response and PIC microcontroller was
preventive measures [14]. The data logging systems to record used to interface
The system was
temperature and oil level
important parameters like temperature, load, voltage, and sensors. Energy meter
not tested in
current is introduced in [15]. In [16], the possibility of Remote operational
provided transformer
monitoring scenarios and
integrating smart meters, IoT devices, and other advanced [24]
system for
loading information and
does not utilize
technologies are discussed to streamline data collection and a microphone was used
transformers machine
to measure humming
management. The data collection methods comply with noise.
learning for
regulatory standards and guidelines in the energy sector is fault detection.
given in [17]. The summary of the recent literature focusing
on smart transformer monitoring is given in Table I. Monitors voltage, load The system was
IoT-based real- current, temperature, and not tested in
The rest of the article is planned as follows. The time monitoring oil level by interfacing operational
objectives of the proposal are presented in section II. The and sensors to ATMEGA328 scenarios and
[25]
methodology and strategies are elaborated in section III. The maintenance of microcontroller. The does not utilize
salient research achievements are highlighted in section IV. distribution system uses MQTT for machine
transformers energy-efficient and learning for
The goals are stated in section V. The dataset used to obtain faster communication. fault detection
results are listed in section VI. The Data are analyzed in
section VII. The obtained results are elaborated in section
VIII. This article conclusion is given in Section IX and the
II. TRANSFORMER HEALTH INDEX (HI)
reference is listed in section X.
TABLE I
SUMMARY OF RECENT LITERATURE FOCUSING ON SMART The health index (HI) for a Distribution Transformer
TRANSFORMER MONITORING AND MANAGEMENT
Monitoring System involves integrating various critical
parameters obtained from the sensors. Health Index formula
Source Application Implementation Limitations that encapsulates essential aspects affecting transformer
Measure voltage, health, namely voltage levels, thermal stress, and oil quality
current, power, and Does not utilize
Transformer temperature. IoT and
[ 18]
monitoring Communicates with a machine (HI )=α∗H I volt + β∗H I temp +γ∗H I oil (1)
GSM modem to a learning.
remote database server.
Where, H I volt : Represents the contribution of voltage
[ 19] Transformer PIC microcontroller to Does not
monitoring interface with the provide real- measurement anomalies to the overall health score. It may be
using IOT sensors and a GPRS time calculated as follows.
Distribution Transformer Health Monitoring 3

transformer’s health index, it is crucial to delve into the

( W i∗|v actual−v set|

)
detailed equivalent circuit model of the transformer. This
∑ model, as depicted in the Fig. 1, includes various elements that
v range account for the imperfections inherent in the transformer’s
H I volt = (2) operation. When combined, these elements enable us to treat
n
the transformer as an ideal one for analysis purposes. In this
Where, Wi denotes weight assigned to each voltage leg; v actual equivalent circuit, 𝑅1 represents the resistance of the primary
winding, while 𝑅 denotes the resistance of the secondary
refers to measured voltage value; v set corresponds to
winding. The inductive reactance components 𝑋1 and 𝑋2
prescribed set points; v range indicates permissible range; and n correspond to the leakage reactance of the primary and
equals the count of voltage legs monitored. secondary windings, respectively.
H I temp : Captures the effect of elevated temperature readings
on the Health Index. We can compute it using the following
expression

H I temp=
∑ ( K v∗t ) (3)
j

¿
¿

m
where Ki denotes penalty coefficient attributed to jth
excessive temperature indicator; t_exceeded_j signifies
duration for which jth temperature limit violation persists; and
m equals the summation of active temperature sensors Fig. 1 Transformer Equivalent Circuit
deployed.
Additionally, the parallel network comprising 𝑅n and 𝑋0
H I oil: Quantifies the repercussions of poor oil quality on the
models the no-load equivalent circuit of the transformer. The
transformer health. An illustrative computation might involve: transformer’s voltage and current transformations are
governed by the ratio 𝐾, expressed as in equation (5);
H I oil=δ∗OL I status + ε∗OT I status + ζ∗OT I A (4) status

E2 N2 I'2
where δ, ε, and ζ denote weights accorded to K= = = (5)
E1 N1 I2
respective oil parameters; OL I statusreflects whether adequate
oil levels are maintained; OT I status evaluates the presence of When a load impedance 𝑍 is connected across the secondary
undue moisture content; and OT I A status flags unusually high winding, the induced electromotive force 𝐸2 causes a
secondary current 𝐼2. This current results in voltage drops
oil temperatures necessitating immediate attention. Balancing
across 𝐼2𝑅2 and 𝐼2𝑋2, thereby reducing the load voltage 𝑉2
the contributions of individual subcomponents towards the
below E2 as mentioned in equation (6):
ultimate Health Index requires careful fine-tuning of
coefficients α, β, and γ, guided by expert judgment and
historical experience. Ideally, these scaling factors ought to V 2=E 2−I 2 ( R 2+ jX 2 )=E 2−I 2 Z 2(6)
satisfy the constraint: α + β + γ = 1
The total primary current 𝐼1 shown in equation (7) drawn from
TABLE II the supply is the vector sum of the no-load current 𝐼0 and the
HEALTH INDEX OF TRANSFORMER current 𝐼′2, which is reflected on the primary side to supply
the secondary load current.
HI% Health State Requirements
85-100 Good Normal Maintenance '
I 1=I 0+ I 2(7)
65-80 Fair Increase Periodic maintenance
50-65 Poor Replacement required
The primary voltage 𝑉1 is calculated by adding the voltage
0-50 Very Poor Risk of any time failure
drops 𝐼1𝑅1 and 𝐼1𝑋1 to the electromotive force 𝐸1 in equation
(8):
Calculating the Health Index periodically enables trend
analysis, threshold breach detection, and informed decision- V 1=E 1+ I 1 ( R 1+ jX 1 ) =E 1+ I 1 Z 1 (8)
making concerning preventive maintenance schedules and
corrective actions required to preserve optimal transformer In this formulation, the negative sign of 𝐸1 indicates that 𝐸1is
functioning. Regular reassessment of the Health Index 180 degrees out of phase with 𝑉1. By incorporating all these
promotes sustained reliability enhancement efforts and ensures imperfections through the various circuit elements, we can
continual alignment with evolving industry standards and effectively analyze the transformer as an ideal entity,
operating requirements. Following the assessment of the
Distribution Transformer Health Monitoring 4

providing a comprehensive understanding of its operational commensurable increases in heat generation. Drawing
dynamics. attention to a 100 kVA transformer again, a 5% variation in
rated current entails approximately ±1000-watt alterations in
active power losses—translating to noticeable modifications in
operating temperatures. Effective current management
III. PROPOSED IoT THMS remains instrumental in perpetuating optimum transformer
performance and staving off untoward degeneration.
In order to effectively monitor the health of the Thirdly, mounting oil temperatures jeopardize the insulating
distribution transformer, a methodology is formulated using fluid's dielectric strengths and spur quicker solid insulation
IoT based Transformer Health Monitoring System (IoT deterioration. Should oil temperatures chronically eclipse
THMS) and it is shown in Fig. 2. The line diagram of the accepted thresholds, partial discharge phenomena, arcing
proposed IoT THMS system is shown in Fig. 3. occurrences, or severe failure modes emerge. Adequate
supervision of oil temperature beneath prescribed limit values
guards the purity of insulating liquids and keeps pace with
favorable working settings.
Fourthly, ambient temperatures impose external thermal
challenges on the transformer, compelling concerted
endeavors to discard internally created warmth and secure
tenable winding and oil temperatures. Implanting competent
cooling architectures and siting transformers in befitting
climatic zones buttresses preferred working milieus and
nurtures sustained productivity.
Analogously, climbing winding temperatures elicit
comparable ramifications relating to insulation decay.
Curtailing thermally motivated duress demands strategic
oversight of current densities, deploying effectual cooling
schemes, and executing ancillary defensive maneuvers against
harmful effects of inflated winding temperatures.
Lastly, diligent observation of oil level indicators guarantees
Fig. 2. IoT based Transformer Health Monitoring System (IoT THMS) suitable submersion degrees for vulnerable parts,
predominantly those prone to oxidation or direct contact
In the initial phase, a strategic deployment of sensors harms. Satisfactory oil levels additionally foster efficient heat
is undertaken to monitor crucial physical and electrical exchange amongst inner constituents, helping manage
parameters within a distribution network. For physical operating temperatures. Intermittently assessing and restoring
parameters, a specialized sensor array is selected to capture oil levels pursuant to manufacturers' suggestions and locale
nuanced information. This includes temperature sensors for meteorological trends strengthens steadfast and reliable
Oil Temperature (OT) and Winding Temperature (WT), level transformer execution.
sensors for Oil Level (OL), and additional sensors for Collectively, astute cognizance of voltage, current, oil
monitoring Ambient Temperature (AT) and Humidity. These temperature, ambient temperature, winding temperature, and
sensors, chosen for their precision and suitability for the oil level indicator metrics galvanizes transformer persistence
specific parameters, collectively contribute to a and boosts confidence in electric power conveyance
comprehensive dataset that forms the basis for real time enterprises. Adhering to recommended operational margins,
monitoring and analysis. coupled with periodical servicing routines, empowers
Each factor exhibits distinct yet interrelated implications stakeholders to fully exploit their assets whilst simultaneously
for the transformer's overall health and functionality. Firstly, defending societal welfare prospects.
voltage substantially affects the transformer's core losses and
eddy current losses. For every 1% rise in voltage, core losses
increase by roughly 0.4%, assuming a linear relationship.
Subsequently, this generates more heat within the transformer,
expediting insulation aging rates. To illustrate, a mere 5%
increase in nominal voltage corresponds to nearly 2% growth
in core losses, equating to additional annual heat dissipation of
~7200 watts for a 100 kVA transformer. Thus, precise voltage
control proves indispensable in preserving the transformer's
structural integrity and functional endurance.
Secondly, fluctuating current magnitudes introduce
proportional changes in Joule heating within the transformer
windings. Every 1% upturn in current intensity triggers a
corresponding 1% amplification in I²R losses, producing
Distribution Transformer Health Monitoring 5

Fig.3. Line diagram of the proposed IoT THMS system

Fig.4. Architecture of proposed IoT THMS system

Simultaneously, multifunctional energy meters are
strategically employed to measure essential electrical An extensive Exploratory Data Analysis (EDA) unveils
parameters. These energy meters encompass Current patterns and correlations within the dataset, guiding the
Transformers (CTs) and Voltage Transformers (VTs), development of machine learning models. These models,
providing phase wise measurements such as Voltage (VL1, trained on the integrated data are deployed for real-time
VL2, VL3), Current (IL1, IL2, IL3), Power Factor (PF), monitoring. An alert mechanism, integrated with the model,
Active Power (KW), Apparent Power (KVA), Reactive Power activates when anomalies are detected, and alerts are
P=1Nn=1Nv(n)in (KVAR), System Voltage (System V), dispatched through communication channels like email, SMS,
System Current (System I), System Power Factor (System or system notifications. In the context of utilizing Digital
PF), Total KVA hours (KVAh), and Total KVAR hours Ocean for hosting the Ocean server, you'll start by creating a
(KVARh). The utilization of multifunctional energy meters Droplet, which is a virtual private server (VPS) offered by
ensures a comprehensive and accurate assessment of the Digital Ocean. This Droplet will serve as the Ocean server
electrical parameters critical for distribution transformer where the MQTT broker is deployed and the applications for
health. handling incoming sensor data is configured in the operating
To ensure seamless data transmission for real-time monitoring system, securing access to the server, and installing the
and predictive analysis, robust network architecture with Wi- necessary software packages. Digital Ocean offers a variety of
Fi connectivity is established. RS485 Modbus communication operating system options, including popular choices like
facilitates efficient data transmission across various sensors. Ubuntu. Once the basic configuration is complete, you'll
The collected data is then transmitted to a cloud-based proceed to install and configure the MQTT broker software,
platform using Wi-Fi, which is designed for receiving, such as Mosquito, on the Droplet. This involves setting up
processing, and securely storing information. Data processing topics, access controls, and security measures to ensure the
pipelines cleanse and preprocess the data, ensuring its quality confidentiality, integrity, and availability of the sensor data
for subsequent analysis. The architecture of the proposed being transmitted. Additionally, you may want to set up a
system is shown in Fig 4. database on the Droplet to store the incoming sensor data. This
could involve installing a database management system like
MySQL and configuring it to receive and store data from the
MQTT broker.
Distribution Transformer Health Monitoring 6

Fig 6 Sequence Diagram of the System

The Fig 6 depicts the sequence diagram and the flow of data
from the transformers to the user application. The transformer
collects the real time sensor data and transmit the data using
Rs485 to the microprocessor (Raspberry pi) and the data is
published to MQTT in the cloud and thus the sensor data is
published to topics, update the subscribers of the recent data
and connect to storage to store the subscribed data this data is
retrieved in form as status, location of the transformer,
monitoring health and sending alerts in Grafana

A. CLOUD-BASED DATA ANALYSIS AND MACHINE

LEARNING
The cloud platform receives data from multiple
transformers, ensuring secure and scalable storage of the
incoming data. Data processing pipelines are implemented to
cleanse and preprocess the data, ensuring data quality and
consistency. Within the cloud environment, the trained
Fig 5 Flow Diagram of the proposed IoT THMS system machine learning model, which is capable of real-time
prediction of mechanical and dielectric failures based on the
The cloud platform's user interface, accessible to incoming sensor data, is deployed. The dataset offers valuable
stakeholders, displays intuitive dashboards with distribution insights into distribution transformer health. The flow chart of
transformer health status, voltage and current trends, Training is shown in Fig. 7.
temperature profiles, and alerts. Historical data analysis Exploratory Data Analysis (EDA) is conducted.
informs proactive maintenance strategies, contributing to grid EDA will unveil patterns and correlations in the data, allowing
optimization. The Flow Diagram of the proposed IoT THMS for informed decisions during the analysis and model training
system is shown in Fig. 5. A comprehensive literature review stages. For this project, a highly-trained machine learning
was conducted covering over 100 national and international model has been carefully selected for its exceptional
high-impact factor journal and conference papers related to performance in handling complex datasets. The dataset is
transformer health monitoring, IoT deployment, and machine divided into training and testing sets to train and evaluate
learning. This extensive review of existing research allowed using the modified cluster based improved isolation Forest
for identification of significant research accomplishments as Algorithm. In CIF, the dataset is first partitioned into clusters
well as notable gaps. Leveraging these findings, specific using a clustering algorithm such as K-means.
research objectives were established based on the outlined Each cluster is then processed separately using the I
shortcomings within current knowledge. Forest algorithm to detect anomalies within that cluster. This
approach allows CIF to capture anomalies that may be
obscured within dense clusters or exhibit different
characteristics from the majority of the data. By leveraging
clustering, CIF can better identify anomalies that occur within
specific clusters or regions of the dataset, providing more
accurate anomaly detection compared to traditional I Forest
methods.
CIF is particularly effective in scenarios where
anomalies are not uniformly distributed throughout the dataset
and exhibit clustered patterns First by using a pre-selection
mechanism we can improve the shortcomings of Isolation
Forest which are the unstable detection results and the low
Distribution Transformer Health Monitoring 7

efficiency caused by randomly dividing features of the dataset The selection matrix S is the set of data boundaries
by selecting a suitable cluster based on the data distribution of and means of the selection cluster Cs in each attribute, which
the dataset, and preferentially select the boundary and centre reflects the distribution characteristics of the selection cluster.
of that cluster as the split values. Then by proceed by The CIIF algorithm implements the selection mechanism for
implementing k-means algorithm to obtain the distribution of split value selected by the selection matrix when constructing
the dataset, which is used to construct a selection matrix to the forest. In this process, the selection of the split values will
implement a pre-selection mechanism. be prioritised from the optional points of the selection matrix
S in that attribute. Then we use the parameter selection degree
I to control the influence of the pre-selection mechanism on
the method and avoid overfitting.
The selection degree I is defined as the maximum
number of times that split value can be selected by the
selection matrix S in each attribute. The degree of selection is
a parameter that controls the randomness of the algorithm and
is determined artificially. The larger the value of I, the more
the forest is influenced by the selection cluster Cs and the
lower the randomness; the smaller the value of I, the less the
forest is influenced by the selection cluster Cs, the greater the
randomness, and the closer it is to the original Isolation Forest
algorithm; when the selection degree I is 0, the algorithm is
the original Isolation Forest algorithm at this time. Define the
discriminant matrix J as the record of the number of split
values decided by each dimension according to the selection
matrix S during the construction of the Isolation Forest by
CIIF:

J (d)=i(12)
where d represents the dimension and i is the record
value. Equation (12) indicates that CIIF performs i times split
value selection for d dimensions in constructing the Isolation
Forest.
After the training phase, the proposed method will
Fig 7 Model Training Flow Chart calculate the outlier scores of all data points in the isolation
forest with the following outlier score calculation formula:
The CIIF algorithm is an unsupervised outlier
detection algorithm that analyses the distribution of the dataset −E (h( x))
through the k-means clustering algorithm, divides the dataset c(n)
into k clusters, and selects the appropriate cluster as the Score ( x , n )=2 (13)
selection cluster Cs. Let the dataset X be divided into k
clusters C1, C2, ……, Ck by the k-means clustering method, where in Equation (13) represents that h(x) is the path
and each cluster is scored as in Equation (9): length of sample x from the root node to the leaf node where it
mi is located, E(h(x)) is the expectation of path length h(x) in an
Score ( c i )=∑ dist ¿¿ ¿ isolated forest, and c(n) is the average of the path lengths of all
j=1 data points, calculated as in Equation (14):
is the Euclidean distance, n1, n2, . . . nk are the amount of
data contained in each cluster, and c1, c2, . . . ck are the 2 ( n−1 )
cluster centres of each cluster. c ( n )=2 H ( n−1 )− (10)
n
Let the dimension of the dataset be ‘d’, and define the
selection matrix as mentioned in Equation (10):
where H(i) is the Harmonic series, which can be
calculated as ln(i) + γ, and γ is the Euler’s constant, which is
S={S1 , S 2 , ·· · S d }(10) approximately equal to 0. 5772156649.While comparing the
anomaly detection performance Matrix for the best methods.
Define the maximum value of the selection cluster Cs give in With the other algorithm One Class Support Vector Machine,
Equation (11) in dimension d as Max (cd), the minimum value Recurrent Neural Network (RNN). It is observed that the CIF
as Min(cd), and the average value as Mean(cd), then: model has with the trained model in place, the cloud-based
platform offers real-time monitoring and predictive analysis.
Sd ={ Max ( c ) , Min ( c ) , Mean ( c ) } (11) In the realm of transformer fault analysis, the adaptation of the
Cluster-based Improved Isolation Forest (CBIF) algorithm
Distribution Transformer Health Monitoring 8

stands as a promising avenue for predictive maintenance. This conditions. Moreover, we utilize a Classification Report to
approach harnesses a sequence of steps tailored to the unique delve deeper into the nuances of the model's performance.
parameters governing transformer health, including voltage, This report furnishes detailed statistics relevant to each
current, oil temperature, ambient temperature, winding category of True positive rate, True negative rate, False
temperature, and oil level indicator readings Beginning with positive rate, False negative rate.
data normalization, raw dataset values undergo standardization TABLE II
ANOMALY DETECTION PERFORMANCE METRICS
within a [0, 1] range using methods like min-max scaling or z-
score scaling. This normalization ensures uniform scales
across all dimensions, facilitating effective analysis. Model TPR TNR FPR FNR
Modified
Cluster based
100.00 98.15 0.00 1.84
Isolation
Forest
One-Class
87.50 97.15 2.84 12.39
SVM

Local Outlier
62.45 81.30 36.89 17.45
Factor

These metrics collectively offer insights into the

success rates and areas needing improvement for distinct types
of transformer malfunctions. By employing such rigorous
evaluation techniques, we ensure utmost confidence in our
predictive capabilities. This, in turn, facilitates timely
intervention, mitigating the risk of costly downtime or
significant damage to transformer assets.

B. REAL TIME MONITORING AND PREDICTIVE

ANALYSIS
A field visit was subsequently undertaken to a
distribution transformer located at Madurai Manikampatti,
providing valuable firsthand insight into practical applications
Fig 8 Inference Flow Chart of framework and considerations surrounding transformers. Following this,
data collection commenced with metering data logs,
After identifying the key indicators, perform preventative maintenance records, and breakdown
dimensionality reduction, if necessary, followed by maintenance reports all being gathered from relevant sources.
partitioning records into homogeneous clusters. Then use Additionally, supplementary information was sourced from
algorithms to group similar transformer operating states. For online datasets available through platforms such as Kaggle.
instance, N=1000 records may split into C=3 clusters after Building upon these foundational steps, an innovative solution
tuning hyperparameters. Utilize the k-means clustering - the system was proposed. Key components of this novel
algorithm to segment the dataset into clusters. Each cluster approach include development of an IoT gateway, creation of
represents a distinct operational behavior or state of the a digital ocean server droplet, setup of the server infrastructure
transformer. The CIIF algorithm can then analyze the including database installation, and implementation of Grafana
distribution within these clusters to identify anomalies. Train for visualization purposes. Through integration of these
the CIIF model using the preprocessed and clustered dataset elements, we aim to advance the state-of-the-art in transformer
and define the threshold value based on the anomaly score health monitoring using cutting-edge technology.
observed during training. Compare the anomaly score of the
new data points against the predefined threshold, based on the With the trained model in place, the cloud-based
predictions, trigger appropriate action plans such as platform offers real-time monitoring and predictive analysis.
maintenance, repairs, or replacements to mitigate potential The model continuously assesses the incoming sensor data
faults and ensure the reliable operation of transformers. from distribution transformers. When anomalies or potential
The model continuously assesses the incoming sensor issues are detected, the model generates alerts. These alerts
data from distribution transformers. When anomalies or can be delivered through various means, such as email, SMS,
potential issues are detected, the model generates alerts. These or system notifications, ensuring timely responses to
alerts can be delivered through various means, such as email, impending issues. Once the model is trained, it can be used for
SMS, the best performance in identifying both anomalous inference on new data. The inference process involves the
readings and normal readings with high TPR and TNR as following system notifications, ensuring timely responses
shown in table I. Each outcome represents specific scenarios impending issues. Once the model is trained, it can be used for
where predictions either align or diverge from actual events, inference on new data. The inference process involves the
enabling us to evaluate the model's reliability across different following
Distribution Transformer Health Monitoring 9

steps: Data is continuously collected from each transformer's lifecycle. The machine learning model
distribution transformers via IoT devices, including energy enables predictive Transformer health management,
meters and sensors. The collected data is processed and optimizing preventive checks scheduling for efficient
analyzed in real-time to monitor transformer health. The maintenance. SMS and email alerts for threshold limits such
collected data is processed and analyzed in real-time to as overload, underload, unbalance, low/empty oil level, high
monitor transformer health. The trained model is utilized to temperature (hot spots), power on/off, low power factor,
predict the likelihood of mechanical and dielectric failures in ensuring proactive response to potential issues. The system
distribution transformers. Real-time alerts are triggered if the contributes to effective demand management, aiding in the
model predicts an impending failure. Alerts can be sent via optimization of power distribution.
various means, such as email, SMS, or system notifications, Predictive maintenance and efficient scheduling
depending on the application's requirements and the contribute to a reduced overall cost of maintenance. Proactive
mechanism flow is represented in Fig 8. health monitoring and timely interventions lead to an extended
Smart transformers efficiently manage industrial, service life for distribution transformers. The platform extends
agricultural, and domestic loads by regulating voltage and beyond transformer health, encompassing power quality,
optimizing energy distribution. In industrial settings, they power theft detection, and critical parameter monitoring at an
ensure stable power supply, minimizing downtime and enterprise level. The data for this project is collected from
enhancing productivity. For agriculture, smart transformers distribution transformers via a combination of energy meters
adapt to varying power demands, supporting sustainable and various sensors. A range of sensors are meticulously
farming practices and renewable energy integration. In deployed to collect essential data for monitoring and
households, these transformers enable demand response and predicting the health of distribution transformers.
energy optimization, reducing electricity bills and enhancing
comfort. Their advanced monitoring and control features
enhance grid stability and efficiency across diverse load types.

Fig 9 Smart Transformer with different types of Loads

Fig 10 Heat Map for different parameters influencing the parameters
Smart transformers represent a transformative
approach to energy management, aligning with the demands of At the heart of our transformer health monitoring system lie
modern industries, farms, and households. Overall, they play a two crucial sets of sensors: Voltage Transformers (VTs) and
crucial role in fostering energy efficiency, reliability, and Environmental Sensors. Starting with VTs, they play a pivotal
sustainability in electrical systems. The different type of loads role in gauging the voltage levels not only across individual
powered by the transformer is depicted in the Fig 9. phases but also between them. Specifically, VTs measure
This transformative project pioneer’s innovation by parameters such as VL1, VL2, VL3, VL12, VL23, and VL31
seamlessly integrating cutting-edge technologies, such as a which contribute significantly towards comprehending the
diverse sensor array, cloud based predictive analytics, and electrical behavior of the transformer. By doing so, potential
advanced machine learning models. The deployment of problems can be proactively detected while ensuring safe
environmental sensors, coupled with RS485 Modbus operability. Complementing the functionality provided by VTs
communication, represents a groundbreaking approach to real- is a carefully curated selection of environmental sensors
time data collection and transmission. The utilization of a designed to keep track of diverse aspects related to both
comprehensive machine learning model not only facilitates internal operations and external influences acting on the
predictive maintenance but also encompasses future life transformer.
predictions, embodying a forward-thinking strategy for
ensuring the longevity of distribution transformers. The In the context of transformer analysis, a heat map is
project introduces the following key enhancements. Realtime instrumental in detecting anomalies or patterns indicative of
monitoring enables swift responses, minimizing downtime and potential issues is represented in Fig. 10. Elevated readings on
optimizing asset availability. A centralized platform maintains the Oil Temperature Indicator (OTI) and Winding
an online asset register, offering a comprehensive history of Temperature Indicator (WTI) can signify overloading
Distribution Transformer Health Monitoring 10

conditions or failures in the cooling system. Increased neutral serving as an early warning mechanism for impending
current (INUT) may point to load imbalances or internal faults complications. Simultaneously, OTI_T acts as a fail-safe
within the transformer windings. Persistently high or mechanism, instigating a controlled shutdown or trip
unbalanced phase currents (IL1, IL2, IL3) and phase voltages procedure should the oil temperature reach dangerously high
(VL1, VL2, VL3) are often indicative of load imbalances or levels, thus mitigating the risk of catastrophic failure.
issues related to power supply quality. Together, these intricate layers of sensory feedback work
Moreover, the heat map elucidates the interrelationships harmoniously to ensure reliable and robust transformer health
between various parameters. For instance, a correlation monitoring.
between rising Ambient Temperature Indicator (ATI) and
increased OTI might suggest that external environmental IV. RESULTS AND DICUSSIONS
conditions are adversely affecting the transformer's operating
parameters. The Oil Temperature Indicator Alarm (OTI_A) Multifunction Energy Meter connected to a
and Oil Temperature Indicator Trip (OTI_T) are critical for Raspberry pi via current transformer as shown in Fig. 11.
understanding the operational thresholds and the transformer's gives information on consumption of real power, reactive
response to high-temperature conditions, providing clear power, power factor, voltage, current values. The Raspberry
visual cues when these thresholds are breached. Utilizing a Pi/USB to Modbus converter features a robust hardware
heat map enables engineers and technicians to rapidly assess configuration, housing the Broadcom BCM2711 Quad-core
the overall health and performance of a distribution Cortex A72 (ARM v8) 64-bit SoC running at 1.8GHz. With
transformer. This facilitates the prioritization of maintenance 8GB LPDDR43200 SDRAM (on model), it ensures efficient
activities, the prediction of potential failures, and the and responsive performance. The device supports versatile
assurance of efficient and reliable operation. The color-coded wireless connectivity, offering 2.4 GHz and 5.0 GHz IEEE
representation enhances the ability to communicate findings 802.11ac wireless, Bluetooth 5.0, and BLE options. For high-
and insights effectively to stakeholders, ensuring that complex speed wired connections, it includes Gigabit Ethernet,
data is both accessible and comprehensible, even to those complemented by 2 USB 3.0 and 2 USB 2.0 ports for diverse
without a technical background. peripheral connectivity.
TABLE III
By leveraging a heat map, engineers can identify trends and PARAMETERS OF DISTRIBUTION TRANSFORMER
patterns over time, allowing for predictive maintenance. For
example, consistent monitoring and analysis of OTI and WTI PARAMETER RATING
can reveal gradual increases in temperature, indicating KVA Rating 100KVA
insulation degradation or other emerging issues. This Voltage Range 10/0.4 KV
proactive approach helps schedule maintenance activities Connection Type Δ/Y
before a failure occurs, thereby avoiding unplanned outages
Turns 1819/42
and extending the transformer's operational lifespan. Among
Core Material M5
these, the Oil Temperature Indicator (OTI) stands out as it
continually monitors the temperature of the insulating oil, Lamination Thickness 0.30mm
thereby flagging any deviations indicative of underlying Winding Material Aluminum
concerns such as overloading or internal faults. Another vital Core Volume 0.05498 m3
component within this sensor array is the Winding Pcoreloss 400 + 10% W
Temperature Indicator (WTI), entrusted with tracking the Pcu 2000 + 10% W
temperature fluctuations within the transformer's windings. Impedance Voltage 4%
Its continuous vigilance helps identify anomalous rises in
temperatures signaling possible overloads or deteriorating
A MicroSD card slot facilitates loading the operating system
insulation quality before they escalate into major issues. In
and serves as a data storage solution. Powering the converter
addition to OTI and WTI, the Ambient Temperature Indicator
is achieved through a 5V DC supply via a USB C connector,
(ATI) assumes importance due to its ability to record the
with a minimum requirement of 3 A*. Operating reliably in
prevailing temperature around the transformer unit. This
ambient temperatures ranging from 0 to 50°C, this converter
feature provides essential context regarding the external
provides a versatile and capable solution for Modbus
environment's influence on overall performance and condition.
communication requirements across various applications.
Furthermore, the Oil Level Indicator (OLI) serves a critical
purpose by consistently monitoring the quantity of oil present
inside the transformer. Maintaining appropriate oil levels is
paramount for averting overheating risks and upholding
optimal insulation standards.
Lastly, two additional safety features complete the
ensemble of environmental sensors: the Oil Temperature
Indicator Alarm (OTI_A) and the Oil Temperature Indicator
Trip (OTI_T). When activated, OTI_A sends forth alarms
whenever the oil temperature surpasses predefined thresholds,
Distribution Transformer Health Monitoring 11

It publishes the measured voltage, power and tariff to the

cloud/server. Then, RPI 4B displays transformer health
monitoring to the cloud. Then, the cloud stores the data sent
by RPI 4B to the web app as a web notification for every
abnormal condition. Message Queuing Telemetry Transport
(MQTT) is a communication protocol where a publisher,
acting as the data source, transmits messages to a broker via
specific topics. Subscribers, serving as data receivers, connect
to the broker using the same topic, facilitating efficient and
organized data exchange within the MQTT framework.
MQTT Box enables you to create MQTT clients to publish or
subscribe to topics, create MQTT virtual devices, load test
Fig 11 Current Transformer, Raspberry -Pi/USB to Modbus Converter, MQTT devices or brokers and much more. It is used for data
Physical connection of MFEM with current transformer implementation in
house exchange between constrained devices and server applications.
The data being subscribed using the topic is shown in Fig. 12.
Modbus operates as a request-response protocol, The cloud-based platform provides a user interface for data
employing a master-slave relationship. In this configuration, a visualization and reporting. Stakeholders can access
single master device, exemplified by the Raspberry Pi, can dashboards that display the current status of distribution
communicate with a maximum of 247 slave devices. The transformers and provide insights into future predictions.
communication process involves the master, in this case, the Visualizations include voltage and current trends, temperature
Raspberry Pi, sending a request to a specific slave device, profiles, and alerts generated by the trained model.
such as a Multifunction meter. This request prompts a
response from the designated slave device, forming the basis
of the master-slave communication exchange. Central to this
communication is the concept of a unique slave ID, which
distinguishes each slave device in the network. Parameters
such as slave ID, baud rate, start bit, stop bit, number of data
bits, parity bit, register type, and register address play crucial
roles in ensuring accurate and reliable data exchange between
the master and slave devices. These parameters collectively
define the communication characteristics and specifications,
allowing for seamless integration and effective information Fig 13 Real, reactive power and the availability of power with min, max,
transfer within the Modbus protocol framework. The signal average and range in Grafana Visualization
conditioning block collects the required information for every
time interval, and is an analog signal. These signals, after Cloud platforms offer robust security features to protect
some limitations, corrections and filtering process, become sensitive data, including the model itself. Additionally, the
standard data and reach the Analog to Digital Converter cloud infrastructure can easily scale to accommodate
(ADC), where it is converted to digital data. Those collected additional transformers and data streams as the power
samples are again sampled and stored in a buffer for short distribution network expands. Complementing the metrics, the
duration to improve accuracy in monitoring. The data is then availability of power serves as a vital indicator of the
shared with a nearby Raspberry Pi module by means of transformer's operational health. Implementing a gauge
Message Queuing Telemetry Transport (MQTT) protocol to visualization within Grafana allows you to depict availability
cloud for further communication to the customer Web percentages or relevant metrics, enabling at-a-glance
application. By connecting Raspberry Pi (RPI) 4 B to 5 V, 2 A assessment of the transformer's reliability. By configuring
supply, and parses the JavaScript Object Notation (Json) and it appropriate thresholds, you can quickly identify deviations
is saved as Comma-Separated Values (CSV) file. from optimal performance and take timely corrective actions.
To provide a comprehensive overview, a table visualization
can display additional metrics such as minimum, maximum,
average, and range for real and reactive power. By calculating
these statistics using functions like MIN (), MAX (), AVG (),
and RANGE () in your queries, Grafana facilitates the
presentation of essential performance indicators in a structured
format. Additionally, you can include metrics like power loss,
commonly referred to as "units lost," to highlight
inefficiencies in energy transmission and distribution.
Visualization of data in form of dashboard is done using
Grafana and the power details is mentioned in Fig 13.
Fig 2 MQTT Box with Publisher and Subscribe with user defined Client
name, MQTT/TCP protocol
Distribution Transformer Health Monitoring 12

geographical locations, aiding in efficient asset management

and maintenance planning. Alert logs promptly notify users of
any parameter exceeding predefined thresholds, enabling swift
intervention to minimize downtime and optimize performance.
A pie chart visualization offers a concise overview of
individual transformer loading conditions, categorizing them
into loading, overloading, underloading, and no power states.
This visual representation facilitates quick assessment and
proactive load management strategies.

Fig 14 Log of parameters (electrical physical), timestamp for transformer in

Grafana Visualization

We can create a versatile visualization setup for monitoring

distribution transformer health by logging electrical physical
parameters with timestamps. This setup enables users to select
specific transformers and parameters, whether electrical,
physical, or a combination, and view them in a tabular format
with timestamps for analysis. Through Grafana's interface,
users can dynamically choose the transformer of interest and Fig 16 Plot of parameter against time in the given time period with min, max,
select parameters such as oil temperature, ambient range and mean.
temperature, winding temperature, oil level, line voltage, line
current, and more. This flexibility allows for targeted Complementing the pie chart, tabulated data presents detailed
monitoring tailored to the unique characteristics and information on transformer status and loading conditions for
requirements of each transformer. By organizing the data in a each area, empowering users to make informed decisions and
table format with timestamps, users can easily track the prioritize maintenance efforts effectively. The transformer
variation of selected parameters over time, facilitating trend stats can be monitored based on location and the condition of
analysis and identification of patterns or anomalies. loading as mentioned in Fig 15. Grafana’s versatile
Timestamps provide crucial context, enabling users to visualization tools allow users to select parameters of interest
correlate parameter readings with specific events or and generate dynamic plots that display the minimum,
operational conditions. This visualization setup not only aids maximum, range, and mean values over the chosen time span.
in real-time monitoring but also supports historical analysis Whether monitoring parameters such as voltage, current,
and performance evaluation. Users can identify trends, detect temperature, or other physical attributes, users can gain
abnormalities, and make informed decisions to optimize valuable insights into transformer performance and health
transformer operation, prevent failures, and ensure reliability. trends. The interactive nature of Grafana's interface enables
The log of data with electrical and physical parameter users to analyse data trends, identify outliers, and make data-
(ambient temperature, winding temperature, oil level, current, driven decisions to optimize transformer operation. By
voltage, power) in Fig 14. visualizing parameter data over time with statistical metrics,
stakeholders can effectively monitor distribution transformer
health, detect anomalies, and implement proactive
maintenance strategies The data log of time series of various
parameter against time can be plotted and analyzed as
mentioned in Fig 16.

Fig 15 Transformer Status with location, alert logs, and loading condition of
individual Transformer in Grafana Visualization

In a distribution transformer health monitoring system

powered by Grafana, users benefit from a streamlined
Fig 17 Phase Current of Phase A, B, C in respect to feeders, overload and
visualization interface that provides crucial insights into
fault threshold in Grafana Visualization
transformer status, location, loading conditions, and alert logs.
Through interactive dashboards, stakeholders can effortlessly The phase Current of the three phases A, B, C can be viewed
track the operational status of transformers across various with respect to the faulty threshold and the overload threshold
Distribution Transformer Health Monitoring 13

as mentioned in Fig 17. Grafana facilitates the visualization of anomalies in the power supply that might affect the
phase currents for the three phases A, B, and C, relative to transformer's efficiency. The phase currents corresponding to
predefined faulty and overload thresholds. Through time- each phase—R, Y, and B—are also under constant
series graphs specifically designed for feeder analysis, surveillance. Monitoring these currents helps in identifying
stakeholders can monitor the fluctuations in phase currents load variations and potential overcurrent scenarios, which
over time, identifying deviations from normal operating could lead to overheating and subsequent failure if not
conditions. By overlaying the thresholds for fault and overload addressed promptly. The ability to track these currents in real
on the graphs, users gain immediate visibility into potential time enables maintenance teams to perform load management
issues or abnormalities in current levels. and predictive maintenance effectively. The dashboard is
equipped to display the maximum recorded power and energy,
providing insights into the peak operational thresholds reached
by the transformer. The maximum recorded power indicates
the highest power load the transformer has handled, which is
critical for understanding its capacity and performance limits.
Similarly, the maximum recorded energy value sheds light on
the total energy consumption over a given period, helping in
energy auditing and efficiency assessments.
Beyond electrical parameters, the dashboard
presents essential machine specifications and operational
status indicators. These include the operational state of the
transformer (e.g., status off/on), the manufacturer details,
Fig 18 Remaining useful Lifetime of Transformer based on prediction of effect
of parameters on the lifetime of sensor device code, and operational mode. These specifications are
To visualize the remaining useful lifetime (RUL) of a vital for inventory management, warranty tracking, and
distribution transformer in Grafana, several parameters are ensuring that the transformer operates in the correct mode as
considered to plot in Fig. 18. First, the "Time vs. Faults per the designed specifications. The power factor and
Encountered" graph shows the number of faults over time, frequency are displayed using a gauge system, providing an
highlighting periods with higher fault frequencies. The "Time intuitive and immediate understanding of these critical
vs. Resultant Life Due to Faults" plot demonstrates the impact parameters. The power factor is a measure of how effectively
of these faults on the transformer's lifespan, with each fault the transformer is using the electrical power, with values
reducing the remaining life, shown as a decreasing trend. The closer to 1 indicating more efficient utilization. Frequency
"Time vs. Expected Lifetime of Transformer" line graph monitoring ensures that the transformer operates within the
represents the projected lifetime of the transformer, starting standard frequency range, essential for maintaining
from installation and decreasing over time due to faults and synchronization with the grid and preventing operational
operational stresses. Finally, the "End of Life Prediction" issues.
marks when the cumulative effect of all parameters and faults A timeline graph of energy consumption per day is
predicts the transformer's end of life. incorporated into the dashboard to provide a visual
representation of the energy usage trends over time. This
graph is instrumental in identifying patterns, peak usage times,
and potential inefficiencies. By analyzing these trends,
operators can implement energy-saving measures and
optimize the transformer's performance. An advanced alert
system is integrated into the dashboard, designed to trigger
notifications in case of abnormal conditions or parameter
breaches. These alerts are dynamically updated in the alert list,
ensuring that any deviation from the norm is promptly flagged
for investigation and resolution. This proactive approach to
fault detection minimizes downtime and prevents minor issues
from escalating into major failures
Fig 19 Different Electrical Parameters of Transformer with Specifications in Combining these plots in Grafana provides a
a Dashboard comprehensive view of how faults and operational conditions
impact the transformer's expected life, offering a clear
The dashboard depicted in Fig 19 meticulously tracks the visualization of its RUL. Grafana's customizable visualization
voltage levels of the transformer across different phases, features allow for the adjustment of threshold values and the
designated as R, Y, and B. Each phase's voltage is monitored presentation of data in a clear and intuitive manner. The cloud
to ensure they remain within the prescribed limits, preventing platform also enables historical data analysis, allowing for the
potential damage due to over or under-voltage conditions. identification of long-term trends and insights that can aid in
Additionally, the system measures the line-to-line voltages— proactive maintenance and grid optimization. The trained
RY (between R and Y phases), YB (between Y and B phases), model continues to play a crucial role in providing valuable
and BR (between B and R phases). These parameters are insights into transformer health based on historical data with
crucial for diagnosing and rectifying any imbalance or
Distribution Transformer Health Monitoring 14

sending alerts on reaching threshold by reducing the data from generation computer systems, 108, pp.533-545.
time series and whenever the value exceeds it triggers the 6)Esmaeili Nezhad, A. and Samimi, M.H., 2022. A review of
firing and sending alerts to the customer. the applications of machine learning in the condition
monitoring of transformers. Energy Systems, pp.1-31.
V. CONCLUSION 7)Xing, Z., He, Y., Chen, J., Wang, X. and Du, B., 2023.
Health evaluation of power transformer using deep learning
The "Distribution Transformer Health Monitoring" project is a neural network. Electric Power Systems Research, 215,
pivotal step in ensuring the reliability of distribution p.109016.
transformers in the power distribution system. By employing 8)Surucu, O., Gadsden, S.A. and Yawney, J., 2023. Condition
data analysis, machine learning, and real-time monitoring, this Monitoring using Machine Learning: A Review of Theory,
project enhances the overall management of distribution Applications, and Recent Advances. Expert Systems with
transformers, leading to cost savings and increased system Applications, 221, p.119738.
reliability. It represents a significant advancement in the field 9)Jiménez-Navarro, M.J., Martínez-Ballesteros, M., Martínez-
of power distribution, addressing critical needs in maintaining Álvarez, F. and Asencio-Cortés, G., 2023. A new deep
the integrity of the electrical grid. The integrated sensors and learning architecture with inductive bias balance for
RS-485 Modbus communication are integral to the success of transformer oil temperature forecasting. Journal of Big Data,
this transformative project. 10(1), p.80.
In the long term, the project envisions revolutionizing 10)Li, Q., Wang, Y., Li, L., Hao, H., Wang, R. and Li, J.,
power distribution by establishing a foundation for smart 2023. Prediction of BLEVE loads on structures using machine
grids. The integration of predictive analytics and machine learning and CFD. Process Safety and Environmental
learning lays the groundwork for proactive grid management, Protection, 171, pp.914-925.
facilitating a transition towards self-healing networks. This 11)Li, Q., Wang, Y., Shao, Y., Li, L. and Hao, H., 2023. A
transformative approach promises sustained reliability, comparative study on the most effective machine learning
reduced maintenance costs, and the evolution of resilient model for blast loading prediction: From GBDT to
power infrastructure tailored to future demands. In the Transformer. Engineering Structures, 276, p.115310.
immediate future, the project's application potential unfolds as 12)Raza, A., Tran, K.P., Koehl, L. and Li, S., 2023. AnoFed:
a Realtime solution for stakeholders. By providing Adaptive anomaly detection for digital health using
instantaneous insights into distribution transformer health, the transformer-based federated learning and support vector data
system empowers utilities with timely decision-making description. Engineering Applications of Artificial
capabilities. Immediate benefits include a drastic reduction in Intelligence, 121, p.106051.
downtime through early fault detection, efficient resource 13)Sadiq Ur Rehman and Adeel Khan, “Integrating IoT
allocation for maintenance, and the optimization of grid Technology for Improved Distribution Transformer
performance. The project's immediate impact lies in enhancing Monitoring and Protection,” Electrical, Control and
the operational efficiency and reliability of existing power Communication Engineering, vol. 19, no. 1, pp. 22–28,2023.
distribution networks 14)R. Rajesh Sharma, “Design of distribution transformer
health management system using IoT sensors,” Journal of Soft
REFERENCES Computing Paradigm, vol. 3, no. 3, pp. 192–204, 2021.
15)C. Bethalsha, S. Karthik, and M. S. Sreejavijay, “Real-time
1)Soni, R. and Mehta, B., 2023. A review on transformer transformer health monitoring using IoT,” International
condition monitoring with critical investigation of mineral oil Journal for Research in Applied Science & Engineering
and alternate dielectric fluids. Electric Power Systems Technology (IJRASET), vol. 8, no. IX, 2020.
Research, 214, p.108954. 16)Yamuna R., Geetha R., Gowdhamkumar S., and
2)Biçen, Y. and Aras, F., 2023. Smart asset management Jambulingam S., “Smart Distribution Transformer Monitoring
system for power transformers coupled with online and offline and Controlling using IoT,” International Research Journal of
monitoring technologies. Engineering Failure Analysis, 154, Multidisciplinary Technovation, vol. 1, pp. 111–115, 2019.
p.107674. 17)A. Swain, E. Abdellatif, A. Mousa, and P.W.T. Pong,
3)Thiviyanathan, V.A., Ker, P.J., Leong, Y.S., Abdullah, F., “Sensor Technologies for Transmission and Distribution
Ismail, A. and Jamaludin, M.Z., 2022. Power transformer Systems: A Review of the Latest Developments,” Energies,
insulation system: A review on the reactions, fault detection, vol. 15, p. 7339, 2022.
challenges and future prospects. Alexandria Engineering 18)Wornpuen, A.; Charbkaew, N.; Bunyagul, T. Advanced
Journal, 61(10), pp.7697-7713. Distribution Transformer Load Monitoring. In Proceedings of
4)Wang, F., Zhou, M., Wu, P., Gao, L., Chen, X. and Mu, X., the 2012 9th International Conference on Electrical
2023. Self-powered transformer intelligent wireless Engineering/Electronics, Computer, Telecommunications and
temperature monitoring system based on an ultra-low Information Technology, Phetchaburi, Thailand, 16–18 May
acceleration piezoelectric vibration energy harvester. Nano 2012; pp. 1–4.
Energy, 114, p.108662. 19)Pawar, R.R.; Wagh, P.A.; Deosarkar, S.B. Distribution
5)Zhang, C., He, Y., Du, B., Yuan, L., Li, B. and Jiang, S., Transformer Monitoring System Using Internet of Things
2020. Transformer fault diagnosis method using IoT based (IoT). In Proceedings of the 2017 International Conference on
monitoring system and ensemble machine learning. Future Computational Intelligence in Data Science (ICCIDS),
Distribution Transformer Health Monitoring 15

Chennai, India, 2–3 June 2017; pp. 1–4.

20)Rahman, S.; Dey, S.K.; Bhawmick, B.K.; Das, N.K.
Design and Implementation of Real Time Transformer Health
Monitoring System Using GSM Technology. In Proceedings
of the 2017 International Conference on Electrical, Computer
and Communication Engineering (ECCE), Cox’s Bazar,
Bangladesh, 6–18 February 2017; pp. 258–261.
21)Srivastava, D.; Tripathi, M.M. Transformer Health
Monitoring System Using Internet of Things. In Proceedings
of the 2018 2nd IEEE International Conference on Power
Electronics, Intelligent Control and Energy Systems
(ICPEICES), Delhi, India, 22–24 October 2018; pp. 903–908.
22) Hasan, W.K.A.; Alraddad, A.; Ashour, A.; Ran, Y.;
Alkelsh, M.A.; Ajele, R.A.M. Design and Implementation
Smart Transformer Based on IoT. In Proceedings of the 2019
International Conference on Computing, Electronics &
Communications Engineering (iCCECE), London, UK, 22–23
August 2019; pp. 16–21.
23) Jamal, H.; Nadeem Khan, M.F.; Anjum, A.; Janjua, M.K.
Thermal Monitoring and Protection for Distribution
Transformer Under Residential Loading Using Internet of
Things. In Proceedings of the 2018 IEEE Global Conference
on Internet of Things (GCIoT), Alexandria, Egypt, 5–7
December 2018; pp. 1–6.
24)Nelson, A.A.; Jaiswal, G.C.; Ballal, M.S.; Tutakne, D.R.
Remote Condition Monitoring System for Distribution
Transformer. In Proceedings of the 2014 Eighteenth National
Power Systems Conference (NPSC), Guwahati, India, 18–20
December 2014; pp. 1–5
25)Roy, T.K.; Roy, T.K. Implementation of IoT: Smart
Maintenance for Distribution Transformer Using MQTT. In
Proceedings of the 2018 International Conference on
Computer, Communication, Chemical, Material and
Electronic Engineering (IC4ME2), Rajshahi, Bangladesh, 8–9
February 2018; pp. 1–4.