JSS MAHAVIDYAPEETHA

JSS ACADEMY OF TECHNICAL EDUCATION, NOIDA

Approved by All India Council for Technical Education (AICTE), New Delhi.
UG programs are Accredited by National Board of Accreditation (NBA), New Delhi.
Affiliated to Dr APJ Abdul Kalam Technical University, Uttar Pradesh, Lucknow.
[Link]

PROJECT SYNOPSIS

GROUP NO:50

Predictive Maintenance System for Machinery with data analysis

Name of the group leader: SAGAR BAMAL

Email ID: 21ec158@[Link]

Group Members:

ROLL NO Name Mobile no. Email ID

210091031006 GAUTAM DADU 8630358895 21ec160@[Link]
3
210091031011 SAGAR BAMAL 9873000803 21ec158@[Link]
6
210091031011 SAGAR CHAUHAN 7037441609 21ec112@[Link]
7
 Problem statement:

In industrial settings, machinery downtime due to unexpected failures can lead to

significant financial losses, decreased productivity, and increased maintenance costs.
Traditional maintenance approaches, such as reactive maintenance or scheduled
preventive maintenance, often fail to optimize machine performance and can lead to
unnecessary service interruptions.

The objective of this project is to develop a Predictive Maintenance System that

leverages data analytics to forecast machinery failures before they occur. By utilizing
historical operational data, sensor readings, and machine learning algorithms, the
system will identify patterns and anomalies indicative of impending failures. This
proactive approach aims to enhance operational efficiency, reduce unplanned
downtime, and optimize maintenance schedules, ultimately leading to cost savings and
improved reliability in industrial operations.

By addressing the need for a more effective maintenance strategy, this project aims to
enhance operational efficiency and contribute to cost savings and improved reliability
in industries reliant on machinery.
 LITERATURE SURVEY:

1. Introduction to Predictive Maintenance:

- PdM shifts from reactive to proactive maintenance, minimizing downtime and costs.

2. Data Sources:

- High-quality data from sensors, maintenance records, and environmental factors is crucial.

- Feature engineering identifies key indicators of machine health (e.g., vibration, temperature).

3. Machine Learning Algorithms:

- Common techniques include decision trees, support vector machines, and neural networks.

- Ensemble methods, like Random Forests, improve prediction accuracy.

4. Anomaly Detection:

- Vital for identifying potential failures using techniques such as clustering and autoencoders.

5. Challenges and Future Directions:

- Challenges include data quality, real-time processing, and model interpretability

- Future research will focus on integrating advanced analytics and diverse data sources.
 Objectives:

1. To collect and preprocess at least 12 months of historical operational

data from selected machinery to ensure a comprehensive dataset for
analysis.

2. To identify at least five key features that significantly impact machine

health and performance through exploratory data analysis.

3. To develop and validate at least three machine learning models with a

prediction accuracy of 85% or higher for forecasting equipment failures.

4. To implement a real-time monitoring dashboard that provides alerts

and visualizations for machine status, ensuring a response time of under
five minutes for maintenance notifications.

5. To research and document the relationship between operational

parameters and failure incidents, resulting in a comprehensive report
outlining predictive insights.

6. To generate monthly maintenance prediction reports that detail

potential failures and recommended actions, facilitating proactive
maintenance scheduling.
 Methodology:

An overview of the implemented CPS architecture for PdM applications

 Sensors on the motor
Workflow

1. Data Collection
 Gather historical operational data from machinery, including sensor readings,
maintenance logs, and failure incidents.

2. Data Preprocessing
 Clean the data by handling missing values, removing outliers, and normalizing
numerical features.
 Split the data into training, validation, and test sets.

3. Data Analysis
 Exploratory Data Analysis (EDA):
o Visualize data distributions, trends, and correlations using techniques like
histograms, scatter plots, and heatmaps.
o Identify patterns in machine performance and failure occurrences.
 Statistical Analysis:
o Apply statistical tests to determine the significance of relationships between
features and failure events.
o Use time-series analysis to examine how operational parameters change over
time and correlate with failures.

4. Feature Engineering
 Identify and extract key features that influence machine health, such as vibration,
temperature, and usage patterns.
 Create additional features based on EDA findings (e.g., rolling averages, lagged
variables).

5. Model Development
 Select machine learning algorithms, including:
o Decision Trees
 o Random Forests
o Support Vector Machines
o Neural Networks
 Train multiple models using the training dataset and tune hyperparameters for optimal
performance.

6. Model Validation
 Evaluate model performance using the validation dataset by calculating metrics such as
accuracy, precision, recall, and F1 score.
 Conduct cross-validation to ensure model robustness.

7. Anomaly Detection
 Implement anomaly detection techniques (e.g., Isolation Forests, clustering algorithms)
to identify abnormal behaviour in machinery.

8. Dashboard Development
 Design and implement a user-friendly dashboard that provides real-time monitoring,
visualizations, and alerts for machine status.

9. Reporting and Insights

 Generate comprehensive reports detailing predictive maintenance insights, including
potential failures and recommended actions based on data analysis.

Algorithms

1. Machine Learning Algorithms

o Random Forest: For robust prediction through ensemble learning.
o Support Vector Machine (SVM): For classification tasks in failure prediction.
o Neural Networks: For capturing complex patterns in large datasets.

2. Anomaly Detection Algorithms

o Isolation Forest: For identifying anomalies in operational data.
o K-Means Clustering: To group similar operational states and detect outliers.

3. Evaluation Metrics
o Accuracy: To measure the proportion of correct predictions.
o Precision: To assess the quality of positive predictions.
o Recall: To evaluate the model’s ability to identify actual failures.
o F1 Score: To balance precision and recall.
 Hardware/Software Tools Required

Software required:

1. Data Collection and Storage

- Database: MySQL or PostgreSQL for data storage.
- ETL Tools: Apache NiFi or Talend for data integration.

2. Data Preprocessing and Analysis

- Python: For data manipulation and analysis.
- Libraries: Pandas, NumPy, Matplotlib, Seaborn.

3. Machine Learning Development

- Scikit-learn: For standard ML algorithms.
-TensorFlow/PyTorch: For neural networks.

4. Anomaly Detection
- Scikit-learn: For Isolation Forest and clustering.
- PyOD: For outlier detection.

5. Dashboard Development
- Flask/Django: For web framework.
- Dash: For interactive visualizations.

6. Version Control
- Git: For version control.
- GitHub/GitLab: For code repository.
7. Documentation
- Jupyter Notebook: For interactive documentation.

8. Deployment
- Docker: For containerization.
- Cloud Platforms: AWS or Google Cloud for hosting.

Hardware required:

1. Computing Hardware
- Personal Computer or Laptop: For development and data analysis.
-Server: For hosting the application and database, ideally with a multi-core CPU and sufficient
RAM (16GB or more).

2. Storage
- Solid State Drive (SSD): For faster data access and processing speeds.
- External Hard Drive: For backup and additional storage if needed.

3. Network Hardware
- Router: For internet connectivity and network access.
- Switch: If working with multiple machines in a local environment.

[Link] Acquisition Hardware

 Sensors: For collecting real-time operational data from machinery (e.g., vibration,
temperature sensors).
 Data Loggers: For storing data from sensors before transmission to the main system.

Responsibility of Each Member:

1. Gautam Dadu
 Documentation
 Software and hardware management.

2. Sagar Bamal
 Team management
  Data visualization
 Circuit planning
 Collecting data

3. Sagar Chauhan
 Report writing
 Information collection
 Data storage

Expected work to complete in next presentation:

In the upcoming presentation, we aim to finalize the data collection of the machinery parts
and visualize the data on the basis of their error limit. Our focus will be on conducting thorough
analysis and validation of the machine maintenance, ensuring it meets the desired
specifications for current needs. We will also explore various optimization techniques to
enhance the existing design, identifying potential improvements in the project.

Grant Chart

Tasks Aug Sept Oct Nov Dec Jan Feb Mar Apr

Literature
Survey and
Research
Collection of
data and
testing of
sensors
Implementing
on various
machines
Visualize the
 data on
dashboard
Error checking
and taking
Feedbacks
Report Writing

REFERENCES:
 Article A Predictive Maintenance System Design and Implementation for Intelligent
Manufacturing Eyup Cinar , Sena Kalay and Inci Saricicek
 Researchgate
 Predictive Maintenance Final Project Python, Machine Predictive Maintenance
Classification by [Link]
 Lee, J.; Jin, C.; Bagheri, B. Cyber physical systems for predictive production systems.
Prod. Eng. 2017, 11, 155–165. [Link]
 Su, C.-J.; Huang, S.-F. Real-time big data analytics for hard disk drive predictive
maintenance. Computer Electr. Eng. 2018,
[Link]
 Meesublak, K.; Klinsukont, T. A Cyber-Physical System Approach for Predictive
Maintenance. In Proceedings of the 2020 IEEE International Conference on Smart
Internet of Things (SmartIoT), Beijing, China, 14–16 August 2020; pp. 337–341

Signature of Students Signature of Supervisor