Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Predictive Maintenance of Railway Point Machine

Using Machine Learning Algorithm
1 2
Tapash Paul Jay Saha
Reasearch & Developement Data Scientist
Bitcomm Technologies Pvt Ltd Bitcomm Technologies Pvt Ltd
Greater Noida, India Greater Noida, India

Abstract:- Indian Railways is one of the largest railway II. LITERATURE REVIEW
networks in the world, and the efficient functioning of its
infrastructure is essential for safe and timely Several research studies have explored the application
transportation. The point machine is one of the most of predictive maintenance techniques to railway systems
critical components of the railway system, which worldwide. An article by Ruiyun Qi et al. (2020) proposed a
controls the movement of trains at junctions. Regular machine learning-based approach for the early detection of
maintenance of the point machine is necessary to ensure point machine failures in high-speed railway systems. The
its proper functioning and prevent failures that can lead authors used a Support Vector Machine (SVM) algorithm to
to accidents and delays. Predictive maintenance is a analyze the data collected from the point machine sensors,
proactive approach that can significantly improve the and the results showed that the proposed approach could
reliability and availability of the point machine. This predict the point machine failure with an accuracy of 90%.
research paper aims to explore the application of
predictive maintenance techniques to Indian Railway Another study by Huang et al. (2018) proposed a deep
Point Machine and analyze its benefits. learning-based approach for the early detection of point
machine failures in railway systems. The authors used a
Keywords:- IoT, Predictive, Point Machine, Machine Convolutional Neural Network (CNN) algorithm to analyze
Learning, Artificial Intelligence, K-means, Clustering, the data collected from the point machine sensors, and the
Indian Railways. results showed that the proposed approach could predict the
point machine failure with an accuracy of 93%.
I. INTRODUCTION
III. METHODOLOGY
The Indian Railways network is spread across 67,415
km, covering around 7,357 stations and transporting The research methodology adopted for this study
millions of passengers and freight every day. The point involves a case study approach, where data related to the
machine is an essential component of the railway system, maintenance of point machines on Indian Railways will be
which controls the movement of trains at junctions. The analyzed. The data will be collected from various sources,
point machine operates through the activation of electric such as maintenance logs, sensor data, and inspection
motors, which align the rail tracks and divert the train onto reports. The collected data will be pre-processed to remove
the desired track. Any malfunction or failure of the point any inconsistencies and noise. Various machine learning
machine can lead to accidents, delays, and disruptions in the algorithms, such as K-Means and DBSCAN, will be used to
train service. analyze the data and predict the potential failures of the
point machine components.
Traditionally, the maintenance of point machines was
done through preventive maintenance, where the  Data Cleaning and Labeling Using Unsupervised
components were replaced at regular intervals, irrespective Learning
of their condition. However, preventive maintenance is not
very efficient, as it may lead to unnecessary replacements of
components that are still functional, resulting in higher
costs. Predictive maintenance is a proactive approach that
uses real-time data analysis to predict the potential failure of
a component and schedule maintenance accordingly. This
approach can significantly reduce maintenance costs,
downtime, and improve system reliability.

Fig 1 Sensor Data

IJISRT23APR178 www.ijisrt.com 1
Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
As our base logic for normal and abnormal data is
based on the reading value, it makes sense that all normal
values will have similar range data while abnormal will have
a different range of data.

In K means we give the model number of cluster (K) as

input with our unlabeled data and the model will create K
clusters and divide all data points into those clusters

 Working:

 Initially the model will randomly choose K points and

assign them as cluster centroids
 Then assign every point to each cluster and then
recalculate the centroids
 Repeat the process until we get a stable centroids where
Fig 2 Operating Current VS Operating Time it’s no more updating.

Important is we need to give the value of K. And that

can be calculated using elbow curve.

Fig 3 Peak Current VS Operating Current

Fig 4 K-Means Model

IJISRT23APR178 www.ijisrt.com 2
Volume 8, Issue 4, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
From our analysis K is coming to be 5.. ACKNOWLEDGMENT

I put K in range from 3 to 5 and created different K- The first author would like to thank Indian Railway for
Means model and analyse the results. the contribution on globally rising technology IoT and AI-
ML for the Indian railway.
IV. RESULTS AND DISCUSSION
 Abbreviations

 AI – Artificial Intelligence
 ML- Machine Learning
 IoT – Internet of Things

REFERENCES

[1]. Qi, R., Wang, Y., Zhao, C., Zhao, Y., Wang, J., &
Zhang, J. (2020). A Machine Learning-Based
Approach for Early Detection of Point Machine
Failures in High-Speed Railway Systems. IEEE
Transactions on Intelligent Transportation Systems,
21(12), 5303-5313. doi: 10.1109/TITS.2020.2985684.
[2]. Huang, H., Wu, Y., Lai, X., & He, Q. (2018). Deep
learning-based approach for early detection of point
machine failures in railway systems. Transportation
Research Part C: Emerging Technologies, 93, 232-
248. doi: 10.1016/j.trc.2018.05.008

Fig 5 K Means Output: Classification of Data

The results of the study showed that predictive

maintenance can significantly improve the reliability and
availability of the point machine on Indian Railways. The
analysis of the maintenance data showed that the failure of
the point machine components could be predicted accurately
using machine learning algorithms. The predictive
maintenance approach can help Indian Railways to schedule
maintenance activities proactively, reducing the downtime of
the point machine and improving the system's availability.

 Roadmap and Future Challenges

 Supervised Learning with more input to the system for

prescriptive Maintenance.
 Availability of Sensors and devices to Point Machine.

V. CONCLUSION

Predictive maintenance is a proactive approach that

can significantly improve the reliability and availability of
the point machine on Indian Railways. The study showed
that machine learning-based algorithms can predict the
potential failures of the point machine components
accurately. The predictive maintenance approach can help
Indian Railways to schedule maintenance activities
proactively, reducing the downtime of the point machine
and improving the system's availability. The study
recommends the adoption of predictive maintenance
techniques for Indian Railways to improve the reliability.

IJISRT23APR178 www.ijisrt.com 3