PREDICTIVE MAINTENANCE ANALYSIS FOR VEHICLES

DISSERTATION

Submitted in partial fulfilment of the requirements of the

Degree : MTech in Data Sciences Engineering

By

Ankit Roy

2023DA04515

Under the supervision of

Ananya Roy, Project Manager

BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE

Pilani (Rajasthan) INDIA

May, 2025
 BIRLA INSTITUTE OF TECHNOLOGY & SCIENCE, PILANI

SECOND SEMESTER 2024-25

DSECLZG628T / AIMLCZG628T DISSERTATION

Dissertation Title: Predictive Maintenance Analysis for Vehicles

Name of Supervisor : Ananya Roy

Name of Student : Ankit Roy

ID No. of Student : 2023DA04515

Courses Relevant for the Project & Corresponding Semester :-

1. Introduction to Data Science - Semester 1

2. Mathematical Foundations for Data Science - Semester 1

3. Data Management for Machine Learning - Semester 2

4. Introduction to Statistical Methods - Semester 2

5. Machine Learning - Semester 2

6. Deep Learning - Semester 3

7. Big Data Systems - Semester 3

Abstract

Predictive Maintenance Analysis for Vehicles

1. Broad Area of Work. The broad area of this project is Predictive Analytics
and Machine Learning in Vehicle Maintenance Management. It falls under the
domain of Artificial Intelligence and Data Science and touches upon real-time
systems and the Internet of Things (IoT). The project incorporates techniques from
sub-fields such as time-series forecasting, anomaly detection, and edge computing.
Using tools such as Apache Spark and PySpark, along with deployment technology
 2

like Docker and cloud services like Amazon Web Services (AWS)/ Google Cloud
Platform (GCP), the system will demonstrate practical applicability in smart fleet
management and logistics industries.

2. Objectives. The objectives of the project are as follows:

(a) To develop a predictive maintenance system capable of determining

whether a vehicle requires maintenance using historical, time-series, and real-
time sensor data (via IoT).

(b) To estimate the timeframe (in terms of days or kilometers) before a

vehicle is likely to need maintenance using regression-based time-to-event
modelling.

(c) To identify and classify the probable issues or faulty components that
may arise in a vehicle using supervised and semi-supervised machine
learning.

(d) To create an interactive dashboard for real-time monitoring of vehicle

health, maintenance alerts, anomaly trends, and actionable insights using
tools like Streamlit or Grafana.

(e) To build a deployable system with containerisation (Docker) and real-

time Message Queuing Telemetry (MQTT) based alerts capabilities,
integrated with edge devices for smart fleet operations.

3. Scope of Work. The scope of this dissertation is to design and develop an

end-to-end predictive maintenance analysis system for vehicles with the following
capabilities:

(a) Ingest and process vehicle operation and maintenance data (including
telematics, On Board Diagnostics (OBD-II) sensor data, and service logs).

(b) Implement machine learning models for multi-task learning including

binary classification (maintenance required/not), regression (time/distance to
failure), and multi-label classification (probable fault categories).

(c) Integrate protocols like MQTT for real-time IoT communication and
implement anomaly detection logic for critical failure alerts.

(d) Develop a real-time dashboard using Streamlit for visualization and

 3

decision support, with the option to extend to Grafana for time-series

analytics.

(e) Ensure system scalability using cloud services (AWS/ GCP), and
maintainability via MLOps best practices.

4. Existing Process and Its Limitations. Traditional fleet maintenance follows

reactive or scheduled preventive approaches. These lack data-driven insights and
face limitations such as high operational cost (average breakdown costing $1,500+
(refer Para 10 (f)), unplanned downtimes, poor component failure prediction, manual
interventions, and no real-time updates (refer Para 10 (g)).

5. Justification for Selecting Methodology. This project applies supervised

Machine Learning (ML) (Random Forest, XGBoost), regression (Linear/ Ridge), and
deep learning (Long Short Term Memory (LSTM) for time series) due to their
superior performance on telemetry data. IoT protocols like MQTT enable timely
alerts. Multi-task models provide consolidated outputs with higher accuracy
(expected >85% accuracy in classification tasks). The visualization and deployment
stack ensures stakeholder usability and integration (refer Para 10 (h), (i), (j)).

6. Project Work Methodology.

(a) Data Acquisition – OBD logs and historical maintenance records.

(b) Feature Engineering – Outlier handling, rolling statistics, and

frequency encoding.

(c) Modelling – Binary classification (maintenance needed or not),

regression (predicting days/ km to failure/ maintenance required), and multi-
label classification (issues like brakes, engine, battery).

(d) Integration – MQTT-based real-time alerts.

(e) Dashboard – Built on Streamlit to show predictions, alerts, and

diagnostics.

(f) Deployment – Docker + cloud deployment for production-readiness.

7. Benefits Derivable from the Work.

(a) Reduced Downtime Optimized service schedules reduce

unscheduled failures.
 4

(b) Operational Cost Savings Up to 20–30% reduction in maintenance

cost using predictive maintenance over preventive maintenance (refer Para 10
(k)).

(c) Improved Safety Reduces risk of catastrophic vehicle failure.

(d) Data Utilization Converts raw telematics into actionable metrics.

(e) Scalable Usable by logistics companies, smart city fleets, and EV

service networks.

8. Any Other Details in Support of the Work. The system architecture allows
for integration with fleet Enterprise Resource Planning (ERP) software. The real-time
capabilities, anomaly alerts, and prediction metrics make it valuable for
transportation companies. Case studies by McKinsey & Co. show predictive
maintenance can reduce breakdowns by up to 50% and extend asset life by 20–40%
(refer Para 10 (l), (m), (n)).

9. Detailed Plan of Work (for 16 Weeks). The plan of work includes specific,
measurable tasks and milestones, organized weekly. The following table provides a
breakdown of each task, its timeline, and deliverables:-

Serial Tasks or Subtasks Start Date Duration Specific

- End Date (Weeks) Deliverable
(a) Requirement analysis, Week 1 - 2 Finalised
understanding business Week 2 project scope
problem, reviewing data and
sources and architecture architecture
planning diagram
(b) Data collection, cleaning, Week 3 - 2 Cleaned
integration of datasets Week 4 dataset ready
for EDA and
modeling
(c) Exploratory Data Week 5 1 Visual and
Analysis (EDA) and statistical
feature engineering summaries;
new feature
sets
(d) Binary classification Week 6 - 2 Trained model
model for maintenance Week 7 with evaluation
prediction metrics
 5

Serial Tasks or Subtasks Start Date Duration Specific

- End Date (Weeks) Deliverable
(e) Regression model to Week 8 - 2 Trained model
estimate time frame Week 9 with
before maintenance MAE/RMSE
results
(f) Multi-label model to Week 10 - 2 Trained fault-
predict probable vehicle Week 11 diagnosis
issues/components model
(g) Dashboard development Week 12 1 Functional
(Streamlit/ Grafana) dashboard with
real-time
predictions
(h) Integration of MQTT- Week 13 1 Live alert
based alert system system with
push
notifications
(i) Deployment of system Week 14 - 2 Deployed
(Docker/ Cloud) Week 15 system with
(Optional) documentation
(j) Final review, testing, Week 16 1 Complete
documentation project report
and
presentation

10. Literature References. The following are key journals, conference papers,
and books referred from the preliminary literature review:-

(a) Fensel, D., et al. (2020). *Semantic Web Technologies*. Springer.

(b) Carvalho, T.P., et al. (2019). Predictive Maintenance Using Machine

Learning: A Review. *Computers & Industrial Engineering*, 137, 106024.

(c) Singh, A., & Kapoor, M. (2021). IoT-based Predictive Maintenance for
Vehicle Health Monitoring. *IEEE Internet of Things Journal*, 8(4), 2295–
2303.

(d) Zhang, Y., & Wang, Y. (2020). Real-Time Fault Detection and
Diagnosis in Vehicle Systems. *Mechanical Systems and Signal Processing*,
141, 106712.
 6

(e) Zhang, J., et al. (2020). A Survey on Edge Computing for Predictive
Maintenance. *IEEE Access*, 8, 169752–169773.

(f) Fleet downtime cost estimation (AMC Group):

https://www.amcsgroup.com/blogs/controlling-fleet-costs-with-preventive-
maintenance

(g) Limitations of traditional ERP systems (Axestrack - Gartner Guide):

https://www.axestrack.com/pr/axestrack-recognized-in-the-2024-gartner-
market-guide-for-fleet-telematics-solution

(h) Effectiveness of ML in predictive maintenance (ScienceDirect):

https://www.sciencedirect.com/science/article/abs/pii/S0360835219304838

(i) IoT + MQTT for real-time alerts (Xyte Blog):

https://www.xyte.io/blog/iot-predictive-maintenance

(j) Multi-task learning for maintenance (ScienceDirect):

https://www.sciencedirect.com/science/article/pii/S0957417422017419

(k) Predictive fleet maintenance driving down costs in the US

(FleetPoint (2024)):
https://www.fleetpoint.org/fleet-industry-news/news-by-date/predictive-fleet-
maintenance-driving-down-costs-in-the-us/

(l) Industry 4.0 and AI relevance (World Economic Forum):

https://www.weforum.org/stories/2019/01/future-technologies-will-drive-
industry-4-0

(m) Volvo Trucks using AI for adaptive maintenance:

https://www.volvotrucks.us/news-and-stories/press-
releases/2024/october/volvo-trucks-taps-ai-for-adaptive-maintenance-uptime-
enhancements-as-part-of-blue-service-contract

(n) Customizable and scalable AI maintenance systems

(ScienceDirect): https://www.sciencedirect.com/journal/mechanical-systems-
and-signal-processing/vol/138/suppl/C
 7

Supervisor’s Rating of the Technical Quality of this Dissertation Outline

Good
EXCELLENT / GOOD / FAIR/ POOR (Please specify): ____________

Supervisor’s suggestions and remarks about the outline (if applicable).

21 May 2025
Date_____________ (Signature of Supervisor)

Name of the supervisor: Ananya Roy

Email Id of Supervisor : [email protected]

Mob # of supervisor: 8123595353

 BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE, PILANI

II SEMESTER 24-25

DSECLZG628T / AIMLCZG628T DISSERTATION

Dissertation Outline (Abstract)

BITS ID No. 2023DA04515 Name of Student: Ankit Roy

Name of Supervisor: Ananya Roy

Designation of Supervisor: Project Manager

Qualification and Experience: MBA, 6.5 Years

Official E- mail ID of Supervisor: [email protected]

Topic of Dissertation: Predictive Maintenance Analysis for Vehicles

Ankit Roy
(Signature of Student) (Signature of Supervisor)

Date: 21 May 2025 Date:--------------