Prediction of endurance limit of steel for

high cycle fatigue

B.Tech Project Report submitted
in partial fulfillment of the requirements
for the degree of

Bachelor of Technology
in
Mechanical Engineering

by

Abhishek
(2001me02)
Adarsh Raj
(2001me05)
----
__
under the guidance of

Dr. Surajit Kumar Paul

----

DEPARTMENT OF MECHANICAL ENGINEERING

INDIAN INSTITUTE OF TECHNOLOGY PATNA
September, 2023
 Certificate

This is to certify that the work contained in this thesis titled “Prediction
of endurance limit of steel for high cycle fatigue” is a bonafide
research work of Abhishek and Adarsh Raj, carried out in the
Department of Mechanical Engineering, Indian Institute of Technology
Patna, under my supervision and that it has not been submitted
elsewhere for a degree.

Supervisor
Dr. Surajit Kumar Paul
Date
Place: IIT Patna

iii
 Declaration

Name of the Student(s): Abhishek and Adarsh Raj

Signature of the Student(s):

B.Tech. Project Title: Prediction of endurance limit of steel for high cycle
fatigue.

This is to certify that Mr Abhishek and Adarsh Raj

1. has/have sincerely worked on their project,
2. has/have contacted me regularly to update on the progress of the assigned
project,
3. has/have received my comments on the preliminary version of the report
and presentation and will address those prior to final
submission/presentation,
4. may be allowed to present/defend the project before the department.

Remarks (if any):

Name of the Project Guide(s) Signature of the Project Guide(s)

Dr. Surajit Kumar Paul

Date

iv
 Acknowledgements

We would like to acknowledge the sincere efforts and valuable time given by Professor Surajit
Kumar Paul (Department of Mechanical Engineering IIT Patna. His valuable guidance and
feedback helped us immensely during this project. We would like to acknowledge the
Mechanical Department of IIT Patna for providing us with an opportunity to broaden the
horizon of our knowledge.

Abhishek
2001me02
Adarsh Raj
2001me05

v
 Abstract

This thesis presents a novel approach to predict the endurance limit of steel by harnessing the
power of deep learning methods, specifically neural networks. The proposed model utilizes
critical material parameters, including cyclic yield strength, ultimate tensile strength, and
percentage area reduction, as well as Brinell hardness, as input features. These parameters are
indicative of a material's resistance to cyclic loading and deformation characteristics.

The research begins by preprocessing and curating a comprehensive dataset of steel samples
with a wide range of properties and characteristics. A neural network model with multiple
hidden layers is trained on this dataset to capture complex relationships between input
parameters and the endurance limit. Hyperparameter tuning and cross-validation techniques
are employed to optimize the model's performance.

Results obtained from the deep learning model are compared with traditional empirical models
and experimental data to evaluate its accuracy and generalizability. The outcomes demonstrate
the potential of deep learning in accurately predicting the endurance limit of steel, even for a
diverse range of steel grades and conditions.

vi
 Contents

List of Figures ................................................................................................................ vii

List of Tables ................................................................................................................ viii
Nomenclature ................................................................................................................. ix

1 Introduction .......................................................................................................................1
1.1 Intorduction ................................................................................................................1
1.1.1 Problem Statement ..............................................................................................1
1.1.2 Research Objectives ............................................................................................1

2 Literature Review ..............................................................................................................3

2.1 Introduction ................................................................................................................3
2.2 Historical Perspective ................................................................................................3

3 Data Collection and Preprocessing...................................................................................4

3.1 Data Collection ..........................................................................................................4
3.1.1 Data sources ........................................................................................................4
3.1.2 Data Variables .....................................................................................................4
3.1.3 Collected data ......................................................................................................5
3.2 Data Preprocessing ..................................................................................................10
3.2.1 Normalization and standardization ...................................................................10
3.2.2 Correlation Analysis .........................................................................................11
3.2.3 Correlation Analysis (Plots) ..............................................................................11

4 Methodology and Model Development ..........................................................................13

4.1 Introduction ..............................................................................................................13
4.2 Data Splitting ...........................................................................................................13
4.2.1 Training - Testing split......................................................................................13
4.3 Model Selection .......................................................................................................13
4.3.1 Classical Approach ...........................................................................................13
4.3.2 Neural Networks ...............................................................................................14
4.3.3 Model Architecture ...........................................................................................14
4.4 Model Training ........................................................................................................15
4.4.1 Training .............................................................................................................15
4.4.2 Loss Function ....................................................................................................15

vii
5 Result and Discussion ......................................................................................................16
5.1 Introduction ..............................................................................................................16
5.2 Model Performance..................................................................................................16

6 Conclusion ........................................................................................................................17
6.1 Conclusion ...............................................................................................................17

References ............................................................................................................................19

viii
List of Figures
Figure 5.1 Correlation Matrix Heatmap 11
Figure 5.2 (L) vs (u) 12
Figure 5.3 (L) vs (f) 12
Figure 5.4 (L) vs HB 12
Figure 4.1 Various regression model along wih their scores 14
Figure 6.1 MSE and RMSE score for neural network model 16
Figure 6.1 Predicted vs Original value using regression 17
Figure 6.2 Predicted vs Original value using neural networks 18

ix
List of Tables
Table 3.1 Monotonic deformation properties and fatigue limit of the selected 5
steels from the AISI Bar Steel Fatigue Database. 7
Table 3.2 Monotonic deformation properties and fatigue limit of the selected
steels from the ASM Handbook
Table 3.3 Monotonic deformation properties and fatigue limit of the selected steels 9
from SAE J1099

x
Nomenclature
• Elastic Modulus (E) in GigaPascals (GPa)
• Ultimate Tensile Strength (σu) in MegaPascals (MPa)
• Fracture Strength (σf) in MegaPascals (MPa)
• Percentage Area Reduction (RA) in percentage
• Total Strain at Fracture (εf)
• Brinell Hardness (HB)

Greek Symbols
α = Component angle, °

Acronyms
• AISI - American Iron and Steel Institute
• SAE - Society of Automotive Engineers
• ASM - American Society for Metals

xi
Chapter 1
Introduction

1.1 Introduction
In the field of mechanical engineering, the assessment of material properties and their
behavior under various conditions is of paramount importance. One crucial aspect in this regard
is the prediction of the endurance limit of steel, a mechanical property that plays a pivotal role
in the design and structural integrity of engineering components. The endurance limit
represents the maximum cyclic stress that a material can endure without failing, making it a
critical parameter in the evaluation of a material's fatigue resistance.

1.1.1 Problem Statement

Reliable life calculations of cyclically loaded structures require knowledge of a
material's cyclic deformation behavior. This study aimed to develop a method for estimating
the cyclic deformation properties of steels from typical tensile values and hardness as
monotonic tensile qualities are frequently available or easily obtained through a
straightforward tension test. Experimentally determining the fatigue limit is expensive given
the time and effort required. Numerous empirical relations based on monotonic tensile
characteristics and/or hardness have been established for calculating the fatigue limit in order
to get over this limitation.
Furthermore, in this study, explored the potential of Brinell hardness as an additional
predictor, aiming to enhance the accuracy and versatility of the predictive model. By
harnessing machine learning techniques, sought to overcome the limitations of traditional
empirical methods and create a more robust and generalizable approach to endurance limit
prediction.

1.1.2 Research Objectives

The primary objectives of this research project are as follows:

• To develop a deep learning model for predicting the endurance limit of steel.
• To analyze the significance of monotonic yield strength, ultimate tensile
strength, and percentage area reduction as predictors.
• To investigate the potential of Brinell hardness as an additional predictor.
• To assess the accuracy and generalizability of the developed model across a
diverse range of steel types and conditions.

2
Chapter 2
Literature Review

2.1 Introduction
In this chapter, a comprehensive review of the existing literature on the prediction of the
endurance limit of steel and the possible application of machine learning techniques in
materials science will be presented. This review is essential to provide a foundation for the
current research project, offering insights into the historical development of predictive
methods and the state-of-the-art approaches in the field.

2.2 Historical Perspective

Historically, the prediction of endurance limit was predominantly reliant on empirical
methods, rooted in extensive experimental testing and observations. These early methods
offered valuable insights into material behavior but were often limited in scope and
applicability, as these experiments were time consuming and expensive.

With the advent of computational techniques, such as finite element analysis (FEA),
researchers gained the ability to simulate and predict material behavior. FEA allowed for
the exploration of factors influencing fatigue limit, including stress concentrations,
material microstructures, and loading conditions. While FEA provided valuable insights,
its computational demands and complexity posed challenges.

Machine learning can be utilized as a powerful tool for predicting the endurance limit of
steel. Supervised learning algorithms, such as regression models, have been harnessed to
develop predictive models. These models are data-driven and capable of capturing
intricate relationships between material properties and endurance limit.
Despite these advancements, challenges persist in the field, including the need for
comprehensive datasets, model interpretability, and algorithm optimization. Future
research aims to enhance model accuracy and broaden the applicability of predictive
models for endurance limit.

3
Chapter 3
Data Collection and Preprocessing

3.1 Data Collection

3.1.1 Data Sources
The dataset utilized for this study was sourced from reputable references within the
field of materials science. The primary data sources include:
• SAE J1099 [8]: Monotonic deformation properties and fatigue limit data for selected
steels.
• ASM Handbook [7]: Monotonic deformation properties and fatigue limit data for
selected steels from the ASM Handbook.
• AISI Bar Steel Fatigue Database [6]: Monotonic deformation properties and fatigue
limit data for selected steels from the AISI Bar Steel Fatigue Database.
These sources are recognized for their comprehensive and authoritative data on the
mechanical properties of various steel grades, making them valuable for our research.

3.1.2 Data Variables

The dataset comprises several key variables, each of which plays a crucial role in
determining the endurance limit of steel. These variables include:

• Elastic Modulus (E) in GigaPascals (GPa)

• Ultimate Tensile Strength (σu) in MegaPascals (MPa)
• Fracture Strength (σf) in MegaPascals (MPa)
• Percentage Area Reduction (RA) in percentage
• Total Strain at Fracture (εf)
• Brinell Hardness (HB)
These variables serve as the foundation for our predictive model.

3.1.3 Collected Data

4
Table 3.1 Monotonic deformation properties and fatigue limit of the selected steels from
the AISI Bar Steel Fatigue Database.
Materials E (GPa) σu (MPa) σf (MPa) RA (%) εf HB σL (MPa)
1541 197 783.2 1207 55.1 0.8 180 260.0

1541 196 905.5 1247 41.7 0.54 195 311.7

1050M 211 821 1128 49.6 0.7 238 254.5

1050M 202 828.5 1065 34.1 0.416 220 360.2

1050M 220 2360 2115 14.7 0.16 536 796.1

1090 219 1090 966 14 0.15 259 307.1

1090M 217 1388 1427 25 0.29 357 404.1

1090 213 1147 1151 21.7 0.245 309 346.8

1090M 212 1252.3 1169 13.6 0.146 279 338.3

1090M 214 1124 840 37.6 0.47 272 376.0

1141AL 216 770.6 1207 57 0.85 223 286.0

1141AL 227 924.8 1405 59 0.88 277 432.0

1141Nb 220 695.2 1000 53 0.76 199 276.6

1141Nb 217 801.8 1228 54 0.77 241 341.8

1141V 214 725.2 1087 49 0.68 217 287.0

1141V 215 796.8 1243 58 0.88 252 332.2

1141V 220 789.4 1117 47 0.64 229 295.8

1038 201 582.2 898 54 0.77 163 222.6

1038 219 651.6 1051 53 0.76 185 240.8

1038 218 743 1292 69 1.16 245 317.6

10V45 213 764.6 1131 48 0.7 212 270.4

10V45 216 908.6 1197 42 0.5 269 374.3

1020 207 502.2 849 63.5 1.01 135 194.3

10B21 212 1104.8 1573 70.5 1.22 322 536.9

1022 200 604.1 1587 68.7 1.16 195 230.6

1045 209 746.6 1151 62 1 222 278.5

1151V 206 760.8 1319 50.6 0.7 205 266.9

15V24 209 877.6 1363 61.1 0.9 243 379.2

4130AL 213 1482.8 2055 44.3 0.6 442 527.6

5
4140 207 1042.9 1519 61.3 1 325 437.4

5140 218 1039.2 1445 52.7 0.8 305 496.8

5150 210 867.1 1382 55.8 0.8 245 294.5

51B60 200 1970 1968 21.6 0.23 450 771

9254V 206 2050 2118 35.1 0.4 536 710.2

1070 207 659 1360 36.2 0.5 280 339.3

C-70 201 964 837 16.8 0.197 241 340.2

15B35 219 939.9 1747 64.5 1.04 286 467.8

4620 208 997.8 1530 58.2 0.872 289 368.5

4320 202 994 1336 63 0.99 188 547.1

4620 211 964 1651.9 61.9 0.966 289 453.5

5120 214 1008 1287 58 0.87 252 471.8

9310 197 902 1418 70.8 1.23 260 464.1

8822 204 1131 1362 56 0.88 252 442.1

8620 203 1764 801.4 13.9 0.15 457 805.7

4320 201 1705.3 2197.8 55.5 0.811 445 774.5

4140 197 1390 1521.5 48.3 0.66 377 602.5

4140 200 1400.5 1771 48.2 0.657 380 607.5

4140 204 1537 1705.2 42.1 0.546 410 626.3

4140 205 1445.8 1754.5 52.1 0.736 390 697.3

4140 199 1355.1 1332.7 38.8 0.492 370 679

4140 201 1474.2 1786.2 52.5 0.745 396 730.4

8620 207 1677.2 1851.7 14.2 0.153 439 731.1

41B17M 213 872.3 1304 68 1.137 277 465.5

86B20 205 1034.1 869 63.7 1.013 294 498.6

4140 213 1924.3 1859 15.7 0.163 477 587.7

41B17M 216 904 1245.4 64.7 1.04 277 480.5

4140 206 1380.4 1384.5 23.6 0.269 394 554.81

86B20 206 1502 968.2 59.7 0.909 401 494.8

5160H 209 1761.2 1843.8 30.2 0.359 456 736.9

9254 206 2020 3644 51 0.71 508 664.6

15V41 193 1070.7 766.5 32.6 0.395 303 431.9

6
 4140 201 1514 2071 48 0.65 405 552.4

1552 196 964 1526 45 0.59 276 387.2

8620 208 1785 2206.4 39 0.49 461 614.6

4140 203 1247.3 1311.3 47.5 0.646 409 601.9

4140 207 1239.5 1537 47.2 0.64 417 575.2

4140 206 1818.3 2016.7 42.6 0.555 477 637.83

4140 208 1332.7 1517.6 44 0.581 375 574.2

8822 212 945.7 1170.2 67.2 1.115 297 511.2

8822 210 1334.7 1542.3 46.9 0.633 301 549.3

8822 194 1215 2690 59.4 0.903 338 608.5

8822 208 1723 3387 50 0.672 449 535.3

8620 212 991 1411.1 54 0.776 326 380.4

8620 210 1145.2 1585.7 50.8 0.711 352 430.2

8620 214 1310.8 1711.2 52 0.736 382 528.2

4320 194 905 1966 59.2 0.896 261 391.8

4320 194 1010 1984 57 0.846 288 466.3

4320 194 1238 2336 55.2 0.804 343 676.6

9310 196 883 1596 61.4 0.899 255 450

9310 196 1085 2155 57.5 0.857 306 531.5

9310 195 1201 2172 56.5 0.832 334 533.2

20MnCr5 194 1053 1991 56.5 0.832 299 489

20MnCr5 194 1337 2351 52 0.74 366 606.8

1538MV 193 973.4 1038.7 28.3 0.333 285 439.1

1538MV 194 943 1069.8 36.4 0.452 277 382.3

Table 3.2
Monotonic deformation properties and fatigue limit of the selected steels from the
ASM Handbook

7
Materials E (GPa) σu (MPa) σf (MPa) RA (%) εf HB σL (MPa)

A538A 185 1515 1896 67 1.1 405 644.5

A538B 185 1860 2137 56 0.82 460 762.1

A538C 180 2000 2241 55 0.81 480 811.3

AM-350 180 1905 2179 20 0.23 496 612.4

H-11 205 2585 3172 33 0.4 660 1037.2

RQC-100 205 940 1069 43 0.56 290 449.1

RQC-100 205 930 1331 67 1.02 290 449.1

10B62 195 1640 1779 38 0.89 430 673.4

1005–1009 205 360 717 73 1.3 90 157.2

1005–1009 205 470 745 66 1.09 125 218.8

1005–1009 200 415 641 64 1.02 125 187.2

1005–1009 200 345 848 80 1.6 90 111.1

1015 205 415 724 68 1.14 80 167.2

1020 205 440 710 62 0.96 108 156.9

1040 200 620 1048 60 0.93 225 202.0

1045 200 725 1227 65 1.04 225 308.7

1045 200 1450 1862 51 0.72 410 645.0

1045 205 1345 1862 59 0.89 390 541.7

1045 205 1585 2103 55 0.81 450 650.1

1045 205 1825 2275 51 0.71 500 712.7

1045 205 2240 2723 41 0.52 595 841.4

1144 195 930 1158 33 0.51 265 313.3

1144 200 1035 1517 25 0.29 305 429.5

1541F 205 950 1276 49 0.68 290 423.3

1541F 205 890 1276 60 0.93 260 455.1

4130 220 895 1420 67 1.12 258 382.4

4130 200 1425 1820 55 0.79 365 523.3

4140 200 1075 1524 60 0.69 310 571.7

4142 200 1060 1117 29 0.35 310 339.8

4142 205 1415 1827 48 0.66 380 571.7

8
 4142 200 1550 1896 47 0.63 400 513.5

4142 200 1760 1999 42 0.54 450 626.5

4142 200 2035 2068 20 0.22 475 629.9

4142 200 1930 2103 37 0.46 450 570.4

4142 205 1930 2172 35 0.43 475 670.0

4142 205 2240 2654 27 0.31 560 729.9

4340 195 825 1089 43 0.57 243 302.4

4340 200 1470 1558 38 0.48 409 534.1

4340 195 1240 1655 57 0.64 350 549.4

5150 195 1670 1931 42 0.87 430 688.9

52100 205 2015 2193 11 0.12 518 700.4

9262 205 925 1041 14 0.16 260 371.2

9262 195 1000 1220 33 0.41 280 423.0

9262 200 1565 1855 32 0.38 410 811.3

950C 205 565 931 64 1.03 159 205.1

950C 205 565 1000 69 1.19 150 196.6

950X 205 440 752 65 1.06 150 210.5

950X 205 530 1000 72 1.24 156 235.5

980X 195 695 1220 68 1.15 225 330.5

Table 3.3 Monotonic deformation properties and fatigue limit of the selected steels from
SAE J1099
Materials E (GPa) σu (MPa) σf (MPa) RA (%) εf HB σL (MPa)
1008 207 331 587.5 77.5 1.49 86 92.7

1008 203 365 649.3 77.9 1.51 90 100.3

1010 203 331 597.1 80.4 1.63 83 116.9

1015 207 414 695.5 68 1.14 80 146.3

1020 203 441 713.5 61.8 0.96 109 143.9

1020 186 393 644.5 64 1.02 108 129.5

1025 207 547 889.4 62.6 0.98 158 197.8

1045 207 752 1082.9 44 0.58 225 291.1

1045 207 1827 2758.8 51 0.71 500 690.3

9
 1045 207 2241 3159.8 41 0.53 595 728.5

1045 207 718 1062.6 48 0.65 192 227.9

1045 207 1956 2705.1 38.3 0.48 500 756.9

1045 207 1344 2137.0 59 0.89 390 512.6

10B21 197 1048 1756.4 67.6 1.13 318 482.7

10B21 203 834 NA NA NA 255 369.6

10B22 203 834 NA NA NA 255 450.7

15B27 203 847 1431.4 69 1.17 250 410.2

15B27 203 916 1525.1 66.5 1.09 264 451.2

4130 200 1427 2207.6 54.7 0.79 366 549.4

4130 221 896 1499.0 67.3 1.12 259 412.6

4140 207 938 NA NA NA 293 473.1

4140 200 2033 2439.6 20 0.22 475 663.5

4142 200 1551 2280.0 47 0.63 400 528.2

4142 200 1929 2642.7 37 0.46 450 597.0

4142 207 1413 2091.2 48 0.65 380 547.9

4142 207 1757 2494.9 42 0.54 450 643.1

4142 207 1929 2604.2 35 0.43 475 667.2

4340 200 1468 2027.3 38.1 0.48 409 540.3

4340 190 1048 NA NA NA 275 429.8

4340 193 827 1185.9 43.4 0.57 243 301.9

4340 207 1171 1826.8 56 0.82 327 502.2

5160 203 1584 2212.8 39.7 0.51 430 634.2

0030 207 496 724.2 46 0.62 137 196.5

0050A 209 787 936.5 19 0.21 192 211.9

0050A 209 583 734.6 26 0.30 174 200.7

0050A 211 702 NA NA NA 206 258.0

8630 207 1144 1475.8 29 0.34 305 334.6

3.2 Data Preprocessing

3.2.1 Normalization and Standardization

10
 To ensure that all variables are on the same scale and to prevent any single variable
from dominating the modeling process, we applied normalization and standardization
techniques to the dataset. Normalization scaled the variables to a specific range
(typically [0, 1]), while standardization transformed the data to have a mean of 0 and
a standard deviation of 1. These steps enhance the performance of our predictive
model and facilitate meaningful comparisons between variables.

3.2.2 Correlation Analysis

Understanding the relationships between variables is paramount in accurately
predicting the endurance limit of steel. To explore these relationships, we calculated
the correlation matrix. This matrix offers insights into the dependencies and
correlations between variables, particularly their impact on the endurance limit.

Figure 3.1 Correlation Matrix Heatmap

3.2.3 Correlation Analysis: Plots

11
Figure 3.2: σ(L) Vs σ(u)

Figure 3.3: σ(L) Vs σ(f)

Figure 3.3: σ(L) Vs HB

12
Chapter 4
Methodology and model development

4.1 Introduction
In this chapter, the methodology employed in the development of the predictive model for
the endurance limit of steel is outlined. The steps taken to create, train, evaluate, and
optimize the model are described. The goal is to provide a clear understanding of the
techniques and processes used to achieve the research objectives.

4.2 Data Splitting

4.2.1 Training-Testing split
To facilitate model development and evaluation, the dataset was divided into three
subsets: a training set, testing and validation set. The training – testing split was used to
train and evaluate ML models, while the validation set was used for independent
validation on the data which our model has not been before and compare them with
original value.

4.3 Model Selection

4.3.1 Classical Approach
Various Classical ML algorithms present across libraries like scikit-learn,
XGBoost etc. have been compared for the available dataset and the obtained
results are shown in Figure 4.1.

13
 Figure 4.1 : Various regression models along with their scores
We went on with Linear regression as it performed the best among all the other
model available for regression.

4.3.2 Neural Networks

Deep learning techniques, particularly neural networks, have demonstrated
remarkable capabilities in predictive modeling tasks. Given the complexity of the
relationships between the material parameters and the endurance limit of steel, neural
networks were chosen as the primary unconventional modeling approach.

4.3.3 Model Architecture

The neural network architecture was designed to accommodate the input features
(cyclic yield strength, ultimate tensile strength, percentage area reduction, and Brinell
hardness) and predict the endurance limit (σL). The initial model employed two
hidden layers, but additional layers will be explored during the optimization process.

14
4.4 Model Training
4.4.1 Training
Training and evaluation was done using the train-test split data mentioned above for
classical regression and complete data for the neural network.

4.4.2 Loss function

To train the model, a suitable loss function (e.g., Root Mean Squared Error) was
utilized to measure the disparity between the predicted endurance limit and the actual
values. The model was trained using an iterative optimization process, such as
stochastic gradient descent (SGD), to minimize this loss.

15
Chapter 5
Results and Discussion

5.1 Introduction
This chapter presents the results of the research, including the performance regression and
deep learning model in predicting the endurance limit of steel. The findings are analyzed
and discussed in detail, addressing the research objectives and shedding light on the
influence of material parameters on the endurance limit.

5.2 Model Performance

The performance of the predictive model for the endurance limit of steel was rigorously
evaluated using the remaining data from the testing set. The Mean Squared Error (MSE)
and Root Mean Square Error(RMSE)was employed to assess the model's accuracy and its
ability to generalize to unseen data.
The RMSE measures the root of average squared difference between the predicted
endurance limit values and the actual values from the testing set. It quantifies the overall
accuracy of the model's predictions. A lower MSE indicates better predictive
performance.

Figure 5.1: MSE and RMSE score for Neural Network Model

16
Chapter 6
Conclusion

6.1 Results and Discussion

In summary, this study has aimed to predict the endurance limit of materials by
employing both traditional linear regression and advanced deep learning techniques.
The results have proven to be quite promising, underscoring the potential accuracy
and reliability of these predictive models within the field of material science. The
linear regression model achieved a commendable R-squared score of 84.78%,
demonstrating its effectiveness in explaining the variance in the endurance limit based
on monotonic tensile test properties. Additionally, the deep learning model exhibited
its capabilities with a low Root Mean Square Error (RMSE) score of 0.69,
emphasizing its capacity for precise predictions. These outcomes collectively
emphasize the promise of data-driven approaches in improving our comprehension of
material behavior and aiding in the prediction of critical mechanical properties.

Figure 6.1: Predicted Vs Original value using regression

17
Figure 6.2: Predicted Vs original value using Neural Networks

18
References
1. Yuan-ying Qiu, Hai-dong Wang ,and Zhao-xi Wang, Evaluation of the Methods for
Estimating the Fully Reversed Unnotched Fatigue Limits of Steels (2019) .

2. Zachary Lopez, Ali Fatemi, A method of predicting cyclic stress–strain curve from
tensile properties for steels A 556 (2012) 540–550.
3. A. Fatemia, A. Plaseieda , A.K. Khosrovanehb , D. Tanner, Application of bi-linear
log–log S–N model to strain-controlled fatigue data of aluminum alloys and its effect
on life predictions, International Journal of Fatigue 27 (2005) 1040–1050.
4. M.L. Roessle 1 , A. Fatemi , Strain-controlled fatigue properties of steels and some
simple approximations, International Journal of Fatigue 22 (2000) 495–511.

19