Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 21

Machine Learning Based Approach for selection of Ball

Bearing
A Project Report on

Machine Learning Based


Approach for Selection of Ball
Bearing.
By
Vasant Phirke (22110602)
Aaditya Naik (22220040)
Rohan Nirwal (22110605)
Omkar Pawar (22110634)

Guide
Dr. Ashok R. Mache

Department of Mechanical Engineering


Vishwakarma Institute of Information Technology, Pune
[2023-24]
Machine Learning Based Approach for selection of Ball
Bearing
Vishwakarma Institute of Information Technology, Pune

CERTIFICATE

This is to certify that Mr. Vasant Phirke, Aaditya Naik, Rohan Nirwal
, Omkar Pawar, have successfully completed the project work entitled “Machine
Learning Based Approach for Selection of Ball Bearing” under my
supervision, in the partial fulfilment of Third Year BTech – Dept. Mechanical
Engineering of Vishwakarma Institute of Information Technology, Pune.

Date: 4th December, 2023


Place: VIIT, Pune

Dr. Ashok Mache Prof. Prashant Anerao


Guide Examiner

Dr. Pravin P Hujare Dr. Sandeep S Kore


Project Coordinator Head of Department
Machine Learning Based Approach for selection of Ball
Bearing

ABSTRACT

In contemporary mechanical systems like automobiles the importance of selecting the


right machine parts has grown significantly due to the increasing number of standard
components. Among these components, ball bearings play a pivotal role within
automotive gearboxes, ensuring seamless motion and exerting a substantial influence on
vehicle performance. Choosing the appropriate bearings is paramount for enhancing
gearbox durability, long-term functionality, and minimizing the risk of unexpected
failures [1]. To assess the performance of ball bearings comprehensively, Finite Element
simulations are conducted under diverse loading conditions, encompassing radial and
axial forces, as well as various angular speeds. These simulations facilitate the evaluation
of contact stress, steel ball deformation, and frictional stress. [4] The data derived from
these simulations is utilized to develop a machine learning-based model that predicts
bearing life and aids in selecting the right bearings for specific design requirements.
Numerical simulation and analytical approach found to be time consuming. The
integration of this machine learning based model streamlines the bearing selection
process, offering a more efficient approach for designers during the initial stages of
design. [4] It reduces the need for extensive simulations and intricate calculations. A
practical case study is presented, focusing on the selection of bearings for automobile
gearbox applications. This approach proves to be a valuable tool in the realm of
automotive engineering, simplifying the process of choosing bearings for gearbox
assemblies. [3]

Vishwakarma Institute of Information Technology, Pune B. Tech. (Mechanical Engineering)


Machine Learning Based Approach for selection of Ball
Bearing

TABLE OF CONTENTS

Description Page No

ABSTRACT i

TABLE OF CONTENTS ii

1. INTRODUCTION 1

1.1 LITERATURE REVIEW 2

2. PROBLEM DEFINITION 3

2.1 Need of Project: 3

2.2 Problem Statement: 3

2.3 Objectives: 3

2.4 Future Scope: 4

3. DESIGN OVERVIEW 5

4. Methodology 6

5. Results 9

6. CONCLUSIONS 11

7. REFERENCES 12
8. Appendix 13

Vishwakarma Institute of Information Technology, Pune B. Tech. (Mechanical Engineering)


Machine Learning Based Approach for selection of Ball
Bearing
LIST OF FIGURES
Description Page No.
1. (A) 6003 Bearing CAD 5
(B) 6900 Bearing CAD
(C) 6800 Bearing CAD
(D) 6901 Bearing CAD

2. 6003 Transient Structural Analysis 7


3. Flow Chart of Methodology 8
4. (A) Showing Bearing No. 6218 10
(B) Showing Bearing No. 6303

Vishwakarma Institute of Information Technology, Pune B. Tech. (Mechanical Engineering)


Machine Learning Based Approach for selection of Ball
Bearing

1. INTRODUCTION
Ball bearings play a crucial role in the seamless operation of various machines and
devices, ranging from automobiles to wind turbines[1]. Accurately predicting their
performance across diverse conditions is imperative for enhancing overall efficiency and
reliability. Tackling this intricate challenge requires the integration of advanced
technology and specialized expertise[2]. Leveraging machine learning, engineers can
construct predictive models that offer heightened accuracy, enabling them to anticipate
potential issues and mitigate the risk of errors and failures[3]. This innovative approach
revolutionizes the selection process for ball bearings, especially in complex systems like
automobile gearboxes[4]. By incorporating machine learning algorithms, engineers can
optimize decision-making, considering intricate factors such as load capacities, rotational
speeds, material characteristics, and operating conditions[5]. This transformative shift
from traditional methods to predictive modelling holds the promise of significantly
improving the performance and longevity of machines, thereby elevating overall
engineering standards[6].

Purpose and Focus of the Project:


Objective:
The objective of this project is to develop a machine learning-based model aimed at
predicting and facilitating the selection of optimal bearings for specific design
requirements, with a particular focus on ball bearings. The goal is to minimize the
reliance on extensive simulations and intricate calculations traditionally associated with
bearing selection processes. Through a comprehensive performance assessment, Finite
Element simulations will be conducted under diverse loading conditions, taking into
account radial and axial forces, as well as varying angular speeds. The ultimate aim is to
streamline the bearing selection process, allowing for quicker decision-making in the
design phase by providing a practical and efficient alternative to time-consuming
numerical simulations and complex calculations.
Significance:
This project holds significant importance as it seeks to develop a machine learning-based
model for efficient bearing selection in mechanical design. The traditional reliance on
extensive simulations and complex calculations is addressed by leveraging machine
learning, particularly focusing on ball bearings. The inclusion of Finite Element
simulations under diverse loading conditions enhances the model's sophistication,
ensuring a comprehensive performance assessment. The primary objective is to
streamline the bearing selection process, reducing dependence on time-consuming
numerical analyses and facilitating quicker decision-making in the design phase. This
approach has broad implications for industries where rapid design iterations are crucial,
promising a more efficient and intelligent method for optimal bearing selection in mech
1
Machine Learning Based Approach for selection of Ball
Bearing
Overview of the Report:
The report provides a comprehensive overview of the methodology, considerations, and
progress made in the development of the machine learning model and the CAD models. It
details the research, analysis, and design phases undertaken to create a practical solution
aimed developing a machine learning model for predicting and selecting bearings based on
specific design requirements is the goal. By employing advanced algorithms and data
analysis, the model aims to discern patterns in diverse datasets to provide accurate
predictions. This approach enhances efficiency in selecting optimal bearings, fostering data-
driven decision- making and improving overall engineering design process.

1.1 LITERATURE REVIEW

The literature review underscores the importance of ball bearings in various


machinery, emphasizing the need for accurate performance prediction through the integration
of advanced technology and specialized expertise. The role of machine learning in
constructing predictive models is highlighted, showcasing its potential to anticipate issues
and enhance decision- making in complex systems, such as automobile gearboxes. The
project's objective is to minimize reliance on traditional, time-consuming simulations and
calculations by developing a machine learning-based model for efficient bearing selection.
Finite Element simulations under diverse loading conditions add sophistication to the model,
enabling a comprehensive performance assessment. The project's significance lies in its
potential to revolutionize mechanical design by offering a more efficient and intelligent
alternative to traditional methods[7]. The overview of the report provides insights into the
methodology, considerations, and progress made in both the machine learning model and
CAD models[8]. It outlines the comprehensive research, analysis, and design phases aimed at
creating a practical solution for predicting and selecting bearings based on specific design
requirements. The emphasis on data- driven decision-making and the reduction of
dependence on time-consuming numerical analyses aligns with the broader implications for
industries requiring swift design iterations[9]. As the project advances, it is expected to
contribute not only to the efficiency and reliability of machines but also to the broader field
of engineering standards[10]. The incorporation of machine learning in bearing selection
processes holds the promise of not only streamlining design phases but also fostering a
paradigm shift in how engineers approach and optimize complex mechanical systems. The
report, as it unfolds, will likely provide valuable insights and guidelines for future endeavors
in the intersection of machine learning and mechanical design[11].

2
Machine Learning Based Approach for selection of Ball
Bearing

2. PROBLEM DEFINITION

2.1 Need of Project:


The utilization of computer-aided engineering (CAE) and machine learning to
predict ball bearing performance, as outlined in the project, holds immense potential
and relevance in contemporary industries. Accurately foreseeing the behaviour of
ball bearings through these advanced technologies brings about substantial
advantages. The ability to predict performance not only aids in minimizing
downtime and maintenance costs but also contributes to heightened efficiency and
improved safety standards. The integration of machine learning in the design and
manufacturing processes can revolutionize the industry, leading to the creation of
superior products and enhancing global competitiveness. The future implications of
this technology are both vast and exciting. Envisioning a world where machines
possess the capability to self-diagnose and repair, or autonomous vehicles navigate
with unparalleled precision and safety, illustrates the transformative power of CAE
and machine learning. As the project progresses, it aligns with the broader vision of
unlocking the full potential of these technologies. Continued research and
development in this direction promise a future where the seamless integration of
CAE and machine learning not only optimizes current industrial processes but also
shapes a brighter, more efficient landscape across various sectors. The outcomes of
this project have the potential to drive innovation, ushering in a new era of
intelligent and high-performing machinery with far-reaching implications for
technological progress.

2.2 Problem Statement:

The conventional methods of predicting ball bearing performance in industrial applications


often rely on time-consuming simulations and calculations, leading to increased downtime,
maintenance costs, and safety concerns. This poses a significant challenge for industries
striving for efficiency and global competitiveness. To address this, the project aims to
leverage computer-aided engineering (CAE) and machine learning to develop a predictive
model for optimal ball bearing selection. The objective is to streamline the design process,
reduce reliance on intricate simulations, and enhance decision-making efficiency. By
integrating advanced technologies, the project seeks to revolutionize the industry, offering a
practical alternative that not only predicts ball bearing performance accurately but also
contributes to improved product quality, increased safety standards, and overall
competitiveness in the global market.

2.3 Objectives:

• To develop a machine learning-based model that predict and aids in selecting the
right bearings for specific design requirements.

3
Machine Learning Based Approach for selection of Ball
• Bearing
To reduce the need for extensive simulations and intricate calculations in the
bearing selection process.

4
Machine Learning Based Approach for selection of Ball
Bearing
• To present a practical case study focusing on the selection of ball bearings.

• Comprehensive Performance Assessment: Conduct Finite Element simulations under


diverse loading conditions to comprehensively assess ball bearing performance,
considering radial and axial forces, as well as different angular speeds.

• The primary objective is to streamline the bearing selection process, reducing the
reliance on time-consuming numerical simulations and intricate calculations. This
efficiency facilitates quicker decision-making in the design phase.

2.4 Future Scope:

The future scope of the project extends beyond its immediate objectives, offering
transformative possibilities for various industries. Once successfully implemented, the
predictive model developed through computer-aided engineering (CAE) and machine
learning for ball bearing performance could pave the way for broader applications and
innovations. Industries could witness a substantial reduction in downtime and
maintenance costs, leading to increased operational efficiency and enhanced safety
measures. [7] As the technology matures, it holds the promise of revolutionizing the entire
design and manufacturing process for machines. Looking ahead, the potential applications
are both expansive and groundbreaking. The incorporation of self-diagnosing and self-
repairing capabilities in machines could become a reality, ushering in a new era of
autonomous and adaptive systems. Imagine a future where machines possess the
intelligence to identify and address issues proactively, reducing the need for human
intervention and minimizing disruptions. In the realm of transportation, the application of
this technology could lead to autonomous vehicles navigating with unparalleled precision
and safety, fundamentally changing the landscape of transportation systems. Continued
research and development in the field of CAE and machine learning can unlock the full
potential of these technologies, creating a brighter and more efficient future. The project's
success could act as a catalyst

for advancements in predictive modeling across various engineering domains, fostering a


paradigm shift in how industries approach design, manufacturing, and maintenance.
Ultimately, the future scope of this project lies in its potential to contribute to a world
where intelligent, data-driven systems enhance efficiency, safety, and competitiveness
across diverse sectors.

5
Machine Learning Based Approach for selection of Ball
Bearing

3. DESIGN OVERVIEW

The project introduces ten distinct CAD models of ball bearings, each meticulously
crafted to align with specific design requirements and performance considerations. These
models showcase a diverse range of configurations, incorporating variations in load
capacities, rotational speeds, material characteristics, and operating conditions. The
intricacies of each design have been carefully addressed, demonstrating adaptability and
versatility in response to different project needs. The CAD models serve as a
comprehensive representation of the machine learning-based predictive model, offering a
tangible and visual aspect to the theoretical advancements achieved in the project. These
designs not only underscore the flexibility of the developed system but also act as
practical tools for engineers and designers seeking optimal ball bearing solutions in
varied mechanical applications.

Figure 1(A) 6003 Bearing CAD Figure 2(B) 6900 Bearing CAD

Figure 4(C) 6800 Bearing CAD Figure 3(D) 6901 Bearing CAD

6
Machine Learning Based Approach for selection of Ball
Bearing

4. Methodology

1. Data Collection and Preprocessing:

- Initiated the testing procedure by creating a comprehensive set of CAD models


representing diverse ball bearing configurations. These models were designed with
careful consideration of various parameters, such as load capacities, rotational speeds,
and material characteristics, to ensure a representative and varied dataset for subsequent
testing phases. Conducted a thorough collection of historical data related to ball bearing
performance. Key parameters, including load, speed, temperature, lubrication, and
instances of failure, were systematically gathered from various sources. This diverse
dataset formed the foundation for subsequent analysis and model development.

2. Computer Aided Engineering (CAE)


Simulation:
Implemented CAD models into ANSYS, a powerful finite element analysis (FEA)
software, to conduct a series of rigorous simulations. These simulations aimed to evaluate
the performance of each ball bearing design under various operating conditions, providing
insights into factors like stress distribution, deformation, and thermal effects. Multiple
tests were performed to assess the robustness and reliability of the CAD models across a
spectrum of real-world scenarios.

 Utilization of CAE Software:

Employed state-of-the-art CAE software to create virtual models of ball bearings. This
process involved translating real-world specifications into digital representations,
ensuring accuracy and fidelity in the virtual environment.

 Comprehensive Simulation:

Executed simulations under a myriad of operating conditions to comprehensively capture


the behavior of ball bearings. This involved varying loads, speeds, and environmental
factors to create a robust dataset that reflects the diverse conditions these components
encounter in real-world applications.

 Recording Relevant Metrics:

During Finite Element Analysis (FEA) simulations, systematically recorded relevant


performance metrics and environmental variables. This included stress distribution,
deformation, and thermal characteristics, providing a detailed understanding of how ball
bearings respond to different stimuli.
7
Machine Learning Based Approach for selection of Ball
Bearing

3. ANSYS Calculations and Output Analysis:

- Leveraged ANSYS calculations to quantify and analyze the performance metrics of the
ball bearing designs. Evaluated factors such as load-bearing capacities, stress tolerances,
and thermal stability to ensure alignment with project specifications. The results from
these simulations were systematically recorded and further analyzed to draw correlations
between design parameters and performance outcomes.

Figure 2. 6003 Transient Structural Analysis

4. Data Compilation and Storage:

Compiled the output data from ANSYS simulations and systematically stored the results
in Excel spreadsheets. This organized data repository serves as a valuable resource for
further analysis, enabling comparisons between different ball bearing designs and
facilitating the extraction of key insights. The structured data format ensures clarity and
accessibility for future reference and collaborative analysis.

5. Machine Learning Model Development:

 Algorithm Selection:

Explored a spectrum of machine learning algorithms, including regression, neural


networks, and decision trees. The goal was to identify the most suitable algorithms
that align with the complexity and nature of the ball bearing performance dataset.

8
Machine Learning Based Approach for selection of Ball
Bearing

Figure 3. Flow Chart of Methodology

 Evaluation of Model Performance:

Evaluated the performance of the machine learning models using separate validation and test
datasets. This step assessed the models' generalization capability, ensuring their effectiveness
in making accurate predictions on new and unseen data.

 Optimization of Model Parameters:

Systematically optimized model parameters to enhance accuracy and robustness. This


involved adjusting hyperparameters and fine-tuning the model architecture, ensuring optimal
performance in real-world applications. The iterative optimization process aimed to achieve
the most reliable and efficient predictive models.

9
Machine Learning Based Approach for selection of Ball
Bearing

5. Results

 The machine learning-based model developed for this project has shown promising
results in predicting and aiding in the selection of the right bearings for specific
design requirements. The model has effectively reduced the need for extensive
simulations and intricate calculations in the bearing selection process, thereby
streamlining it.

 A practical case study focusing on the selection of ball bearings was presented.
The model was trained with data from Finite Element simulations conducted
under diverse loading conditions. These simulations comprehensively assessed
ball bearing performance, considering radial and axial forces, as well as different
angular speeds.

 The primary objective of streamlining the bearing selection process was achieved.
The reliance on time-consuming numerical simulations and intricate calculations
was significantly reduced. This efficiency has facilitated quicker decision-making
in the design phase, proving the effectiveness of the machine learning approach in
this context.

 In conclusion, the machine learning model has successfully met the objectives of the
project. It has demonstrated its capability to predict and aid in the selection of the
right bearings, thereby reducing the need for extensive simulations and intricate
calculations. This marks a significant advancement in the field of mechanical
engineering and opens up new possibilities for further research and development.

10
Machine Learning Based Approach for selection of Ball
Bearing

 GUI:

Figure 4. (A) Showing Bearing No. 6218

Figure 4. (B) Showing Bearing No. 6303

11
Machine Learning Based Approach for selection of Ball
Bearing

6. CONCLUSIONS

In conclusion, the developed machine learning-based model for ball bearing selection
represents a significant advancement in the realm of engineering design. The project,
encompassing data collection, preprocessing, CAE simulations, and machine learning model
development, has yielded a system that offers substantial benefits to the engineering
community. The model's contribution to efficiency is evident through its ability to swiftly
determine optimal configurations, eliminating the need for manual calculations and reducing
iterative processes. This efficiency boost translates into valuable time savings within project
timelines. Moreover, the model addresses the issue of human errors inherent in traditional
methods by considering various parameters, thereby minimizing the likelihood of mistakes.
This not only enhances the overall reliability of bearing selection but also ensures more
accurate outcomes in product performance. The reduction in errors contributes to improved
product quality and reliability. One of the standout features of the model is its design
adaptability. By considering multiple initial conditions and allowing designers to explore
diverse scenarios, the model grants a level of flexibility that goes beyond conventional
approaches. This adaptability ensures that the selected bearings not only meet theoretical
expectations but also excel in real-world applications. Such adaptability contributes
significantly to overall product success, as the chosen bearings can be optimized to meet
specific project requirements. In essence, the project has successfully combined advanced
technologies, including computer-aided engineering and machine learning, to streamline and
enhance the ball bearing selection process. The outcomes hold the potential to revolutionize
how engineers approach design, offering a more efficient, reliable, and adaptable solution
with far-reaching implications for the field of mechanical engineering.

12
Machine Learning Based Approach for selection of Ball
Bearing

7. REFERENCES

1. M. Kordestani et al., "New Condition-Based Monitoring and Fusion Approaches


With a Bounded Uncertainty for Bearing Lifetime Prediction," in IEEE Sensors
Journal, vol. 22, no. 9, pp. 9078-9086, 1 May1, 2022, doi:
10.1109/JSEN.2022.3159624.

2. Gao, C.; Lv, K.; Wu, T.; Si, J.; Hu, Y. Method for Determining Starting Point of
Rolling Bearing Life Prediction Based on Linear Regression. Electronics 2019, 8,
923. https://doi.org/10.3390/electronics8090923

3. Yanwei Xu, Weiwei Cai, Tancheng Xie, Pengfei Zhao, "Residual Life Prediction of
Metro Traction Motor Bearing Based on Convolutional Neural Network", Shock and
Vibration, vol. 2021,

4. Ma Hailong and Li Zhen 2021 IOP Conf. Ser.: Mater. Sci. Eng. 1043 032065

5. Zhang R, Zeng Z, Li Y, Liu J, Wang Z. Research on Remaining Useful Life


Prediction Method of Rolling Bearing Based on Digital Twin. Entropy. 2022;

6. Schwarz, S., Grillenberger, H., Graf-Goller, O., Bartz, M., Tremmel, S., & Wartzack,
S. (2022). Using Machine Learning Methods for Predicting Cage Performance
Criteria in an Angular Contact Ball Bearing. Lubricants, 10(2), 25.

7. Kahr, M., Kovács, G., Loinig, M., & Brückl, H. (2022). Condition monitoring of ball
bearings based on machine learning with synthetically generated data. Sensors, 22(7),
2490.

8. Toma, R. N., Providing, A. E., & Kim, J. M. (2020). Bearing fault diagnosis of
induction motors using a genetic algorithm and machine learning classifiers. Sensors,
20(7), 1884.

9. Yang, F., Zhang, W., Tao, L., & Ma, J. (2020). Transfer learning strategies for deep
learning-based PHM algorithms. Applied Sciences, 10(7), 2361.

10. Nishat Toma, R., & Kim, J. M. (2020). Bearing fault classification of induction
motors using discrete wavelet transform and ensemble machine learning algorithms.
Applied Sciences, 10(15), 5251.

11. Ren, L., Sun, Y., Cui, J., & Zhang, L. (2018). Bearing remaining useful life prediction
based on deep autoencoder and deep neural networks. Journal of Manufacturing
Systems, 48, 71-77.

13
Machine Learning Based Approach for selection of Ball
Bearing

8. Appendix
Program for Ball Bearing Selection
Tool:

import pandas as pd
import numpy as np
from sklearn.model_selection import
train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score,
mean_absolute_error # for Error Determination

vasant_data =
pd.read_csv('/content/Vasant_Project.csv')
#Importing CSV file

columns_with_null = ['MIN_Pressure',
'MAX_Pressure', 'AVG_Pressure']
vasant_data.dropna(subset=columns_with_null,
inplace=True)

selected_columns = ['Shaft_Diameter',
'Limitng_speed', 'Bearing_No']
vasant_data = vasant_data[selected_columns]
# Select a column of input nad value to predict

X = vasant_data.drop(['Bearing_No'], axis=1)
y = vasant_data['Bearing_No']
# Predicting Data

X_train, X_test, y_train, y_test =


train_test_split(X, y, test_size=0.2,
random_state=42)
# training of the data
random_model =

14
Machine Learning Based Approach for selection of Ball
Bearing
RandomForestClassifier(n_estimators=100,
random_state=42)

15
Machine Learning Based Approach for selection of Ball
Bearing

y_train_pred = random_model.predict(X_train)

# predicting accuracy of the model


accuracy = accuracy_score(y_train, y_train_pred)
print(f'Accuracy on training data: {accuracy}')

# Determination of mean error Value


mae = mean_absolute_error(y_train, y_train_pred)
print(f'Mean Absolute Error on training data:
{mae}')

# Giving Input Paramaters for predicting Value


Shaft_Diameter = float(input("Enter The Diameter:
"))
Limitng_speed = float(input("Enter Limiting Speed:
"))

user_input = np.array([[Shaft_Diameter,
Limitng_speed]])
prediction = random_model.predict(user_input)

# Predicing the Value


print(f"The predicted Bearing No based on the input
is: {prediction}")

16

You might also like