B7 Final Presentation Sem 5

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 27

MINI PROJECT 2-A

A Data-driven Machine Learning Approach for


the Identification of a Combination of Gear and
Bearing Defects in Rotary Machines.
Team Members
Group - B7
Ananya Sastry 2021226
Shashank Singh 2021237
Amey Yewale 2021242
Team Guide
Mrs. Pallavi Khaire
History and Background

[1] Source: www.researchgate.net

04-11-2023 2
History and Background

• The research states that 82% of companies have experienced


unplanned downtime over the last three years.

• Another research states that 64% of unplanned downtime is related


to equipment failures arising from improper maintenance,
overrunning machines, and a lack of efficient condition tracking.

• It is thus important to analyze the machine’s faults in earlier stages


to avoid catastrophic failure.

04-11-2023 3
Aim
Utilizing a vibration-based machine
learning approach to detect defects in
rotary machines.

Objective
• Develop a machine learning program to identify defects
• Identify suitable algorithm
• Enough computing power for sophisticated analytics.
• Ability to use data from many machines.
• Reduce intervention to shop floor processes.
04-11-2023 4
Literature Review
PUBLICATION/
Sr. No. AUTHORS TITLE CONCLUSION
JOURNAL

In this work, a comprehensive methodology is proposed to


Identification achieve fault identification with minimal error, even with
G.N.D.S. of unbalance fewer measured vibrations. Two different approaches,
Journal of Sound and equivalent loads minimization, and vibration
1. Sudhakar and in a rotor minimization methods, were applied successfully to
Vibration
A.S. Sekhar[1] bearing identify unbalance faults in rotor systems by measuring
system. transverse vibrations at only one location.

The research paper investigates the effectiveness of


Fault detection ensemble machine learning techniques for detecting faults
of anti-friction in anti-friction bearings. The study explores the
bearing using application of various ensemble algorithms to enhance the
International Journal
2. S. Patil and ensemble accuracy and reliability of bearing fault detection.
of Engineering Through rigorous experimentation and analysis, the paper
V. Phalle[2] machine
learning demonstrates the potential of ensemble methods in
methods. improving the early detection of faults, contributing to
more robust machinery maintenance and reliability.

04-11-2023 5
Sr. AUTHOR PUBLICATION/
TITLE CONCLUSION
No. S JOURNAL

Experimental fault
This study focused on the fault detection and classification
diagnosis for
of centrifugal pumps (CPs) to ensure their consistent and
known and unseen
reliable operation in various industries, this research
J.S. Rapur operating
contributes to enhancing the reliability and maintenance
3. and Rajiv conditions of Measurement
practices of CPs by providing accurate fault detection and
Tiwari [3] centrifugal pumps
classification techniques, ultimately ensuring the
using MSVM and
uninterrupted workflow and optimal performance of
WPT based
industrial plants.
analyses.

This study focuses on gear application with tooth defect. ML


Gear fault algorithms are trained using simulated vibration data,
V. C. detection using validated with experimental data, and achieve over 87%
International
Handikher machine learning accuracy across Support Vector Machine, Random Forest,
4. Journal of
kar and V. techniques-a and Multi-Layer Perceptron models. These results
Engineering
M. Phalle simulation-driven demonstrate the promising applicability of this approach for
[4]
approach gear fault detection in industrial settings.
04-11-2023 6
Sr. PUBLICATION/
AUTHORS TITLE CONCLUSION
No. JOURNAL

The study applies three machine learning


techniques, namely support vector machine
(SVM), artificial neural network (ANN), and
Fault diagnosis of ball
self-organizing maps (SOM), for fault
bearings using Applied Soft
5. Kankar et al.[5] classification. The results demonstrate the
continuous wavelet Computing
efficacy of the methodology, with SVM
transform.
outperforming ANN and SOM in accurately
identifying fault categories of rolling element
bearings.

Development of In this paper, a data-driven intelligent fault


intelligent fault diagnosis technique was developed for early
diagnosis technique of fault detection. An experimental setup was
6. Saha et al.[6] rotary machine Sensors fabricated considering all the design criteria to
element wearing: a produce data similar to real-life industry data.
machine learning The time wave data were captured in image
approach. format
04-11-2023 7
Sr. AUTHO PUBLICATION/
TITLE CONCLUSION
No. RS JOURNAL

An experiment was conducted on a high-pressure


centrifugal pump with single-stage impellers to analyze
the impact of different impeller faults at varying flow
Diagnosis of rates. The study focused on pressure fluctuations and
Centrifugal Pump pump performance. Results revealed that vibration levels
Albraik
7. Faults Using Journal of Physics increased with higher flow rates due to distinct defects
et al.[7]
Vibration Methods on each impeller. The interaction between impellers and
the volute was noted as crucial for designing high-energy
pumps. The study concluded that impellers with varying
gaps exhibited different data patterns, attributed to
unique faults.

04-11-2023 8
Summary of literature review

These research papers collectively contribute to the advancement of fault diagnosis in mechanical
systems through various machine learning and signal processing techniques. They address specific
issues such as rotor bearing unbalance, anti-friction bearing faults, gear fault detection, ball bearing
diagnosis using wavelet transforms, intelligent fault diagnosis for rotary machine elements through
vibration analysis. The papers employ innovative approaches to improve the reliability and
performance of mechanical systems by detecting and diagnosing faults effectively, making them
valuable contributions to the field of predictive maintenance and machinery health.

04-11-2023 9
Problem definition

The traditional methods for defect identification often rely on


post-failure analysis, leading to costly downtimes and decreased
productivity. Thus the problem statement proposes

“To develop a data-driven approach for real-


time defect detection in rotary machinery
minimizing downtime and enhancing
operational efficiency.”

04-11-2023 10
Project Scope
• Acquiring vibration data from rotating machines
• Convert the vibration data into frequency domain
vibration plot.
• Extraction of features using machine learning
• Reduction of dominating features using machine
learning
• Application of machine learning algorithms (e.g.,
ANN, SVM, Random Forest ) for training the
vibration data
• Test the accuracy of above mentioned algorithms
• Selection of most accurate algorithm for web [2] Source: www.sonotec.eu
application
• Development of web application

04-11-2023 11
METHODOLOGY

• Data Collection and Preprocessing


• Feature Reduction
• Machine Learning Model Selection
• Model Training
• Performance Evaluation

04-11-2023 12
FEATURE Reduction
The task of feature reduction and selection methods is to obtain the most relevant
information from the original data and represent that information in lower dimensionality
space.

The goal is to build using the available features, those that will perform better.

Common techniques of Feature Reduction:

1) Principal Component Analysis

2) High Correlation Method


[3] Source: freecontent.manning.com
3) Backward Feature Elimination

4) Forward Feature Selection

04-11-2023 13
PRINICIPAL COMPONENT ANALYSIS(PCA)
The primary goal of PCA is to reduce the dimensionality of a
dataset while preserving as much variance as possible.
It achieves this by identifying orthogonal axes called
principal components, where the data varies the most.

Steps in Principal Component Analysis:

Data Standardization

Covariance Matrix

Eigenvalue Decomposition

Selecting Principal Components

Projection

[4] Source: twitter.com

04-11-2023 14
High correlation method
The "high correlation method" is used to address highly
correlated features in a dataset. High correlation can lead to
issues like multicollinearity. To apply this method:
1. Calculate the correlation matrix to measure relationships
between features.
2. Identify highly correlated pairs of features using a
predefined threshold.
3. Decide how to handle high correlation, which can involve
feature removal, combination, dimensionality reduction, or
grouping.
4. Evaluate the impact on model performance before and after
applying the method. [5] Source: www.scribbr.com
5. Iterate as needed to balance information preservation and
redundancy reduction.
6. This method is valuable for improving model performance,
interpretability, and efficiency.

04-11-2023 15
Backward feature elimination

The process involves the following steps

1. Build an Initial Model


2. Rank Features
3. Remove the Least Important Feature
4. Rebuild the Model
5. Evaluate Model Performance
6. Repeat

[6] Source: www.hindawi.com

04-11-2023 16
Forward feature selection

The process typically proceeds as follows


1. Start with a empty model
2. Evaluate feature candidates
3. Select the best Feature
4. Build model
5. Evaluate the model performance
6. Repeat
7. Stopping Criterion
[7] Source: www.researchgate.net

04-11-2023 17
Support vector machines (SVM)

• Support Vector Machines (SVMs) are a type of supervised


machine learning algorithm used for classification and
regression tasks.

• They are particularly effective for problems involving


complex decision boundaries and high-dimensional data due
to their ability to handle complex data distributions and
produce robust classifiers.

• However, they might not perform as well on very large


datasets or when the number of features significantly [8] Source: www.ris-ai.com
outweighs the number of samples, as training SVMs in such
cases can be computationally intensive.

04-11-2023 18
Support vector machines (SVM)

• The primary objective of SVMs is to find a hyperplane that


best separates different classes in the dataset.

• The main objective of SVMs is to find the hyperplane that


maximizes the margin between the two classes.

• SVMs assume that the data is linearly separable, meaning that


it can be divided into two classes using a straight line (in 2D),
a plane (in 3D), or a hyperplane (in higher dimensions).

• Kernel functions enable SVMs to handle nonlinear data by


[9] Source: editor.analyticsvidhya.com
transforming the original features into a higher-dimensional
space where the data becomes linearly separable.

04-11-2023 19
RANDOM FOREST
• Random Forest is an ensemble learning technique that
combines the predictions from multiple machine learning
models to make more accurate and robust predictions.

• The base model of a Random Forest is the decision tree.

• For classification tasks, each tree in the Random Forest


"votes" for a class, and the class with the most votes is
considered the final prediction.

• For regression tasks, the predictions of all trees are


averaged to produce the final result.

• Random Forests can be computationally expensive, [10] Source: www.researchgate.net


especially when there are a large number of trees in the
ensemble.

04-11-2023 20
RANDOM FOREST
Applications of Random Forest

• Spam detection, image classification, and customer churn prediction.

• Regression: Stock price prediction, real estate price prediction, and more.

• Feature Selection: Identifying the most important features in a dataset.

• Anomaly Detection: Detecting anomalies in financial transactions, network


traffic, etc.

04-11-2023 21
ARTIFICIAL NUERAL NETWORK
• Artificial neural network (ANN) is an interconnected group of artificial
neurons that use a mathematical or computational model for information
processing.

• ANN is an adaptive system that changes its structure based on


information that flows through the network.

• The fundamental building blocks of a neural network are artificial


neurons, which are also known as nodes or units. These neurons are
organized into layers.

• Hidden Layers are intermediate layers between the input and output
layers responsible for processing the input data by applying various
mathematical transformations to it.
[11] Source: www.tutorialspoint.com
• Multiple hidden layers can be present, making the network "deep" (thus
called deep learning).

• The output layer produces the final result or prediction.

04-11-2023 22
ARTIFICIAL NUERAL NETWORK
Applications

• Image and speech recognition

• natural language processing

• recommendation systems

• autonomous vehicles

• medical diagnosis

[12] Source: miro.medium.com


04-11-2023 23
RESULT AND Conclusion
• Code have been created for distinct feature reduction techniques,
including Forward Feature Reduction, Backward Feature Reduction,
Principal Component Analysis, and the High Correlation Method.

• Code have been developed for various fault detection methods,


including Artificial Neural Network, Multi-Class Support Vector
Machine, and Random Forest

04-11-2023 24
04-11-2023 25
REFERENCES:
[1] Sudhakar, G. N. D. S., & Sekhar, A. S. (2011). Identification of unbalance in a rotor bearing
system. Journal of Sound and Vibration, 330(10), 2299-2313.
[2] Patil, S., & Phalle, V. (2018). Fault detection of anti-friction bearing using ensemble machine learning
methods. International Journal of Engineering, 31(11), 1972-1981.
[3] Rapur, J. S., & Tiwari, R. (2019). Experimental fault diagnosis for known and unseen
operating conditions of centrifugal pumps using MSVM and WPT based analyses. Measurement, 147,
106809.
[4] Handikherkar, V. C., & Phalle, V. M. (2021). Gear fault detection using machine learning
techniques-a simulation-driven approach. International Journal of Engineering, 34(1), 212- 223.
[5] Kankar, P. K., Sharma, S. C., & Harsha, S. P. (2011). Fault diagnosis of ball bearings using
continuous wavelet transform. Applied Soft Computing, 11(2), 2300-2312.
[6] Saha, D. K., Hoque, M. E., & Badihi, H. (2022). Development of intelligent fault diagnosis
technique of rotary machine element bearing: A machine learning approach. Sensors, 22(3), 1073.
[7] Albraik, A., Althobiani, F., Gu, F., & Ball, A. (2012, May). Diagnosis of centrifugal pump faults
using vibration methods. In Journal of Physics: Conference Series (Vol.364, No.1, p.
012139). IOP Publishing.

04-11-2023 26
Thanks!

04-11-2023 27

You might also like