Fault Detection and Classification Using

11 IV April 2023
https://doi.org/10.22214/ijraset.2023.50867
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue IV Apr 2023- Available at www.ijraset.com
Fault Detection and Classification using

Machine Learning
Dr. B. Suresh Kumar1, V. Abhinav Varma2, S. Devika Rani3, A. Nishanth4
Chaitanya Bharathi Institute of Technology
Abstract: Machine learning plays a crucial role in predicting and classifying faults in electrical power systems. The complexity
and dynamic nature of these systems make them vulnerable to disturbances and electrical faults. Detecting faults in the circuit
can aid in maintaining the system by preventing potential damage from occurring. It is particularly important to be able to locate
faults in transmission lines to minimize power and revenue losses. In this study, MATLAB software will be utilized to simulate
and locate faults in transmission lines. A transmission line model will be designed, and a fault toolbox will be employed to cre-
ate various faults. These faults will be saved and used to train machine learning models to identify the best algorithm with high
precision and accuracy.
Keywords: Machine learning· Fault classification· MATLAB· Power sys-tems.
I. INTRODUCTION
The task of fault detection and classification (FDC) holds significant importance in a range of industries, including transportation,
energy, and manufacturing. The primary objective of FDC is to recognize any abnormal conditions present in a given system or
process and categorize them into distinct fault categories. This process can aid in the prevention of system failures, reduction of
down- time, and improvement of overall productivity. With the emergence of machine learning (ML), FDC has become more
efficient as it can learn patterns and re- lationships automatically from large amounts of data. ML algorithms can be trained on data
collected from MATLAB Simulink models to accurately detect and classify faults. Additionally, ML-based FDC has the capability
to adapt to changing conditions and learn from past experiences, making it more effective than conventional rule-based methods.
II. RELATED WORK

Goh et al.[1] emphasize the importance of fault location in electric power systems, which can be caused by various factors such as
natural events, human error, and equipment malfunctions. Effective fault location is critical to ensure continuous system operation,
minimize revenue loss, and prevent harm to the environment and equipment. The authors suggest that a traveling-based method is
superior for detecting faults in transmission lines as it is not affected by load variation, series capacitor banks, or high grounding
resistance. However, this approach may require more expensive devices than other methods, even though it is more efficient.
In order for this study to be successful, it is crucial to develop a reliable model that can simulate a three-phase power system
network. Without an accurate simulation environment, any algorithm tested will not be considered effective. Tharani et al. [2]
demonstrated the effectiveness of a MATLAB-based simulation model for three-phase power system networks. This study
highlights the conve- nience and reliability of MATLAB software for analyzing and simulating faults in transmission lines within a
three-phase power system network.
Karekar, Thakur, and Manju [3] conducted a study on analyzing and detecting faults in transmission lines through a novel
approach. The study involved testing L-G fault, LL-G fault, and LLL-G fault using MATLAB Sim- ulation with the Simulink and
sim-power toolbox. The researchers found that the use of MATLAB software was efficient for this purpose. The system used in the
study consisted of 33Kv generators, three-phase simulators, and distributed parameter lines. The three-phase simulators were
utilized to simulate faults at the mid-position, while the distributed parameter lines represented faults on the transmission line.
Approximately 80 percent of transmission line disturbances are caused by undetected faults. As faults can arise due to various
environmental factors, such as natural disasters, it is essential to implement techniques that ensure a continuous power supply
within the system. A continuous power supply is crucial because unaccounted faults can lead to the gradual degradation of power
quality across transmission lines. Therefore, it is imperative to have a fast and accurate fault detection process in place to ensure
stability, performance, and efficiency[4]. Since transmission networks have multiple lines, fault detection using fault detectors can
be prohibitively expensive.
© IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 3063
Therefore, it is important to explore alter- native methods such as impedance-based techniques and traveling-wave-based fault
location and integrate them into algorithms and online simulations to find an efficient means of detecting faults.
With the rapid advancements and constant changes in technology, the de- mand for power continues to rise steadily. As a result, it is
imperative to ensure the safe, efficient, and maximum delivery of power. Rajput and Chaturvedi conducted a MATLAB simulation
which demonstrated that the software is a proficient tool for executing algorithms to detect faults in transmission lines [5]. Out of
the various simulations and techniques tested, online simulations proved to be a viable option, as they are cost-effective and easily
modifiable.
Learning to Optimize" is a research paper presented at the International Conference on Machine Learning (ICML) in 2018[6]. The
paper proposes a new approach to optimization that learns to optimize using deep neural networks. The traditional optimization
methods used in machine learning involve finding the best values for a set of parameters in a given function. However, these
methods can be slow and inefficient for complex functions, and they often require man- ual tuning. Overall, "Learning to Optimize"
presents an innovative approach to optimization that has the potential to improve the efficiency and effectiveness of machine
learning algorithms. The paper provides valuable insights into the application of deep learning to optimization problems and has
implications for the development of more efficient and effective machine learning systems.
III. STATEMENT OF THE PROBLEM

The objective of this study is to comprehend the distinct behaviors exhibited by different fault types in a three-phase transmission
line and to detect and classify faults with machine learning. The researchers aim to simulate the potential outcomes of various
types of faults, including line-to-ground, line-to-line, double-line-to-ground, and triple-line-to-ground. By comprehending the
patterns and behaviors of these fault types, it will be easier to identify and resolve them in real-world scenarios. Initiating a
transmission line simulation in Matlab and Simulink can be chal- lenging, particularly for beginners. The required control blocks
and libraries may not be immediately clear, and the correct settings and parameters require some explanation. This documentation
may not fully convey the functions of certain block parameters, particularly in real-world equivalents. Consequently, this study
provides an overview of how to create a transmission line system and generatefaults within it to understand faults and their behaviors
and to be able to collect data from MATLAB and detect and classify faults with various machine learning algorithms by using data
collected.
IV. DESCRIPTION OF THE MATLAB SYSTEM

In order to simulate faults in transmission lines effectively, a MATLAB Simulink transmission line circuit was created. This model
was based on previous models that had been studied. The circuit consists of subsystems that utilize 5 cascaded blocks of the Three-
Phase Pi Section Lines Subsystem with connection ports at each end. On the left side of the circuit, a Three-Phase source acts as a
voltage source, while a load is created by adding another three-phase source with an offset of -30 degrees on the opposite end. To
measure voltage and current at each line, a Three-Phase V-I Measurement is included. Additionally, a Three- Phase Fault is added to
the system to simulate faults and compare stable systems with those that have faults. To ensure accurate results and decreased
simulation time, a graphical user interface, a Powergui block with a sample time of 3e-5, was added to the circuit. The Simulink
model can be viewed in Figure 1.
Fig. 1. Simulink block model of the Transmission Line System featuring a connected fault block.
64
V. METHODOLOGY
A. Methodology for creating MATLAB Simulink Model
1) Model Preparation
Created a Simulink model as described in figure 1, and set parameters as described above. we began by constructing a transmission
line block system in Simulink. The specific blocks used were the following: 1. Three-Phase Pi Section Line 2. Connec- tion Port 3.
Three-Phase Fault 4. Three-Phase V-I Measurement 5. Three-Phase Source 6. Powergui 7. Scope 8. From 9. To Workspace
We initiated the creation of a Transmission Line system by generating a Three-Phase Pi Line subsystem. This involved importing
five Three-Phase Pi Line blocks and six connection port blocks into the workspace. In Matlab R2013, these blocks should be found
in Simscape > Electrical > Specialized Power Systems > Fundamental Blocks > Elements.
As part of the theoretical assumptions of an ideal transmission line system,we did not modify the default settings of the Three-Phase
Pi Section Line blocks. These blocks were then connected in the manner depicted below. The blocks were then connected in the
manner shown below.
After setting up the Three-Phase Pi Section Line blocks as described above, we selected all the blocks and used the Ctrl+G keyboard
shortcut to create a subsystem block.
The Three-Phase Source block will act as the voltage source of the transmis-sion line circuit.
The Voltage (Vabc) and Current (Iabc) measurements will be taken on this block.
Fig. 2. The Three-Phase Pi Section Line subsystem block.
Fig. 3. Parameters on first Three-Phase V-I Measurement block.
The second Three-Phase V-I Measurement block did not perform any measurements. Instead, it functioned as a relay bus for the
circuit.In this block select voltage and current measurements as ’no’.
To decrease simulation computation time, the Powergui block was set to Discrete-time. The sample time was set to 3e-5, although
this value is arbitrary. Lower sample times will generate more accurate results at the expense of longer computation times. Prior to
running any simulations, no faults were enabled in the Fault block.
2) Signal Selection
Selected the signal’s voltage and current as parameters for data collection and evaluation from respective scopes.
Simulation: simulated Simulink model with the selected signals to generate data. The simulation can be run for a specific time
period, here we chose it around 0.5 or until a specific event occurs, such as a fault or a change in the system operation.
3) Data collection and Preprocessing

To fetch the data for an individual fault: choose runtime (0.5s) to observe fault graph > click on save > click on Run > select scope>
history > click on save data to workspace> variable name: I_xyz (for current scope), V_xyz( for voltage scope) > format: Structure
with time > click on apply and okay > select To workspace > save format: Time series > click on apply and okay > click on Matlab
home tab open variables > To collect current scope data: click on I_xyz > signals > values(Ia, Ib, Ic), To collect Voltage scope
data: click on V_xyz > signals > values(Va, Vb, Vc).
Here, select all values and copy them to the Excel sheet in the format: Ia, Ib, Ic, Va, Vb, Vc. Now, remove steady-state values from
the data fetched by observing the scope graph (i.e, remove data before and after the fault occurred) because we need only fault region
data for our machine learning analysis, here we included data from 0.02s to 0.08s ( because in this time range, the fault is observed
in both I, V scopes ), but for the NO-fault case, we have collected data in the entire 0.5s range. Repeat the same process for
fetching data of all individual faults and add the data collected to the Excel sheet in the format specified.
– columns of collected data are in the format: G, C, B, A, Ia, Ib, Ic, Va, Vb, Vc.
– binary code for [ G C B A ] is given as follows:
1- fault case, 0- no fault case.

– 0 0 0 0 - No-Fault,
– 1 0 0 1 - LG fault (Between Phase A and Gnd),
– 1 0 1 0 - LG fault (Between Phase B and Gnd),
– 1 1 0 0 - LG fault (Between Phase C and Gnd),
– 0 0 1 1 - LL fault (Between Phase A and Phase B),
– 0 1 1 0 - LL fault (Between Phase B and Phase C),
– 0 1 0 1 - LL fault (Between Phase C and Phase A),
– 1 0 1 1 - LLG Fault (Between Phases A, B and ground),
– 1 1 1 0 - LLG Fault (Between Phases B, C and ground),
– 1 1 0 1 - LLG Fault (Between Phases C, A and ground),
– 0 1 1 1 - LLL Fault(Between all three phases),
– 1 1 1 1 - LLLG fault( Three phase symmetrical fault).
Here’s a link to Data Set click here.
B. Methodology For Fault Detection And Classification Using Machine Learning In Power System Considered
1) Data Collection: Collected fault data from MATLAB workspace. This data includes signals relevant to the system’s operation,
such as voltages and currents only in the range where the fault occurred, and added 4 columns[G, C, B, A] to represent various
faults in binary format.
2) Data Labeling: Labeled the extracted features with the corresponding fault type as shown above in the data collection and
preprocessing section of Matlab methodologies.
3) Data Visualization: Data visualization plays a significant role in fault detection and classification using machine learning. It
helps in identifying patterns and anomalies that may indicate faults, understanding the relationship between different variables
and their impact on the system, and aiding in feature selection for building accurate machine learning models. Here, the data
visualization libraries used are Seaborn, and Matplotlib.
66
4) Model Training: Trained the selected machine learning model using the labeled data. This involves using a portion of the
labeled data to train the model(Train data: 80 percent) and another portion to validate the model’s performance.
5) Model Testing: Tested the trained model using additional data that was not used in the training phase( Test data: 20 percent).
This is necessary to evaluate the model’s performance on new, unseen data.
6) Data Encoding: Labeled the extracted features with the corresponding fault type. as shown in the figure 4.
7) Model selection: By performing all suitable machine learning models an optimal model was chosen for the task of fault
detection and classification. Which possesses the necessary capabilities to effectively handle the unique features of the collected
power system data. This model has been shown to achieve a higher degree of accuracy in accurately classifying different types
of faults.
This link guides you through Machine Learning Code in Github: click here.
Fig. 4. Label encoding
VI. RESULTS AND DISCUSSION

A. MATLAB-fault block settings and scope results
When no fault is selected the system is said to be in a steady state. Similarly, if any of the lines and ground are selected then it is line
to ground fault. Now to illustrate the scope view let’s select phase C and ground, block parameters, and scope graphs as shown in
Figs 5 to 7. Similarly, simulated all fault types.
B. Prediction Results From Different Machine Learning Algorithms

Machine learning algorithms and their obtained accuracies are:
1) Logistic regression’s model accuracy score: 43.18%
2) Decision Tree model accuracy score: 88.02%
3) Support Vector Machine’s accuracy score: 76.17%
4) Random Forest’s accuracy score: 89.06 %
After performing these algorithms we came to the conclusion that random forest and decision tree are working best for this type of
problem with greater accuracy. Comparatively Random forest is giving more accuracy than Decision tree.
VII. ANALYSIS OF DATA

The simulations conducted previously involve a fault where one phase is grounded.The results show that the voltage of the grounded
phase fluctuates, while the voltages of the other phases are only slightly affected. The current in the faulted
Fig. 5. Parameters on 3 phase fault block (ex: fault between line C and Ground.)
Fig. 6. voltage scope of fault between line C and Ground.
Fig. 7. current scope of fault between line C and Ground.)
68
Phase peaks abruptly at around 0.02 seconds, causing destabilization, whereas the currents in the other phases remain stable
throughout the simulation. The fault can occur when A and B, B and C, or A and C are connected to the ground. The voltage
experiences significant fluctuations during the fault, while the current in the faulted phase peaks at the beginning of the fault and
continues to fluctuate. The current in the non-faulted phase remains stable. The results indicate that the fault occurred at
approximately 0.02 seconds, causing the voltage to drop to 0 and the current in the faulted phase to peak. After approximately 1
second, both voltage and current stabilize. For line-to-line faults and beyond, the behavior is similar to that of multi-line-to-ground
faults.
VIII. CONCLUSION
MATLAB Simulink can be a potent tool to simulate transmission line faults. Before creating our own simulation model, this study
explored different transmission line fault detection techniques proposed in recent studies. They demonstrated how various
algorithms can be used to assess faults. The study then explained how to create a Transmission Line simulation model in Simulink,
including creating a cascaded Three-Phase Pi Line subsystem. The Three-Phase fault block was used for fault detection
simulations and connected across the circuit to simulate different fault scenarios. These scenarios showed different outcomes, with
voltage destabilization and lowering, and abrupt increases in current. The model was simulated in discrete time to reduce
computation time. The data collected was then fed into all suitable machine-learning models and we chose a model with greater
accuracy i.e Random forest. In summary, this study has successfully demonstrated the effectiveness of MATLAB and Machine
learning in simulating and detecting transmission line faults and classifying them with high accuracy.
REFERENCES
[1] H.H Goh et al., “Fault Location Techniques in Electrical Power System: A Review,” Indonesian Journal of Electrical Engineering and Computer Science, Vol.
8, No. 1, pp. 206 - 212, 2017.
[2] C. Tharani, M. Nandhini, R. Sundar, and K. Nithiyananthan, “MATLAB based Simulations model for three phases Power System Network,” International Journal
for Research in Applied Science Engineering Technology (IJRASET), vol. 4, p. 502-509, 2016.
[3] S. Karekar, V. Thakur and M., “A Novel Scheme of Transmission Line Faults Analysis and Detection by Using MATLAB Simulation,” International Journal
of Engineering Research and General Science, Vol. 4, No. 1, 2016.
[4] M. Salehi, et al. “Determining minimum number and placement of fault detectors in transmission network for fault location observability,”International Journal of
Electrical Power and Energy Systems, vol. 124, 2020.
[5] S. Rajput and K.T. Chaturvedi, “Transmission Line Fault Analysis by Us- ing Matlab Simulation,” IJREE - International Journal of Research in Electrical
Engineering, vol. 3, no. 4, 2016.
[6] "Learning to Optimize" by Yutian Chen, Matthew W. Hoffman, Sergio Gomez Colmenarejo, Misha Denil, Timothy P. Lillicrap, and David Duvenaud.
Conference: International Conference on Machine Learning (ICML), 2018.

Fault Detection and Classification Using

Uploaded by

Copyright:

Available Formats

Fault Detection and Classification Using

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fault Detection and Classification Using

Uploaded by

Copyright:

Available Formats

11 IV April 2023

Fault Detection and Classification using

II. RELATED WORK

III. STATEMENT OF THE PROBLEM

IV. DESCRIPTION OF THE MATLAB SYSTEM

Fig. 2. The Three-Phase Pi Section Line subsystem block.

Fig. 3. Parameters on first Three-Phase V-I Measurement block.

3) Data collection and Preprocessing

1- fault case, 0- no fault case.

Fig. 4. Label encoding

VI. RESULTS AND DISCUSSION

B. Prediction Results From Different Machine Learning Algorithms

VII. ANALYSIS OF DATA

Fig. 6. voltage scope of fault between line C and Ground.

Fig. 7. current scope of fault between line C and Ground.)

You might also like