OPTIMIZATIONS OF EDM PROCESS PARAMETERS USING

AI/ML

A Mini project report

Submitted in the partial fulfillment of the requirements for the
award of the degree of

BACHELOR OF TECHNOLOGY
in
DEPARTMENT OF MECHANICAL ENGINEERING
By

S.ADITHYA. (B22ME004),

B.RAJESH. (B22ME008),

J.THARUN. (B22ME010),

P.SAIKIRAN. (B22ME005).

Under the guidance of

Dr J.Laxman
Assistant Professor.MED

Dr G.Srinivasa rao
Assistant professor,MED

DEPARTMENT OF MECHANICAL ENGINEERING

KAKATIYA INSTITUTE OF TECHNOLOGY & SCIENCE, WARANGAL
(An Autonomous Institute under Kakatiya University, Warangal)

i
 DEPARTMENT OF MECHANICAL ENGINEERING
KAKATIYA INSTITUTE OF TECHNOLOGY & SCIENCE, WARANGAL
(An Autonomous Institute under Kakatiya University, Warangal)

CERTIFICATE

This is to certify that this is the bonafide record of the Mini project work entitled "
OPTIMIZATIONS OF EDM PROCESS PARAMETERS USING AI/ML" carried out by
S.ADITHYA. (B22ME004), B.RAJESH (B22ME008), J.THARUN (B22ME010), P.SAIKIRAN
(B22ME005),V semester students of B.Tech. MECHANICAL ENGINEERING, Kakatiya
Institute of Technology & Science, Warangal in partial fulfillment for the award of the degree of
Bachelor of Technology in Mechanical Engineering of Kakatiya University, Warangal.

Mini project Guides : Dr. U. Shrinivas Balraj,

Professor of ME.
Dr J.Laxman, (Chairman,MPEC-6ME)
Assistant Professor.MED.

Dr G.Srinivasa rao,
Assistant professor,MED

Dr.P.Srikanth
Professor & HOD of ME

ii
 DECLARATION

We declare that the Mini project report is original and has been carried out in the
Department of Mechanical Engineering, Kakatiya Institute of Technology and Science,
Warangal, Telangana, and to best of our knowledge it has been not submitted elsewhere for
any degree.

S.ADITHYA. (B22ME004),

B.RAJESH. (B22ME008),

J.THARUN. (B22ME010),

P.SAIKIRAN. (B22ME005).

iii
ACKNOWLEDGEMENT

I express our deepest sense of gratitude and indebtedness to my Mini project guides Dr
J.Laxman, Assistant Professor, MED & Dr G.Srinivasa Rao, Assistant professor, MED, KITS,
Warangal for having been a source of consistent inspiration, precious guidance and generous
assistance during project work. We deem it as a privilege to have worked under his able guidance.
Without his close monitoring and valuable suggestions this work wouldn’t have taken this shape. I
feel that this help is not substitute and unforgettable.

I am, thankful to B. Tech Mini project Convener, Dr.U.Shrinivas Balraj, Professor of ME, KITSW,
Mini project members, V.Srikanth , Assistant Professor and P.Anil Kumar, Assistant Professor,
Dept. of MEE, KITSW for timely conduction of Mini projects.

I am, profoundly thankful to Dr.P.Srikanth Professor & HOD of ME for his constant support and
encouragement. I am, express my sincere thanks to Dr. K. Ashoka Reddy, Principal, KITS,
Warangal, for his kind gesture and support.

I am, indebted to the Management of Kakatiya Institute of Technology and Science, Warangal, for
providing the necessary infrastructure and good academic environment in an endeavor to complete the
Mini project and special thanks for providing Department Library of ECE and Digital Library to
access Reputed Journal papers.

I would like to acknowledge the faculty and non-teaching staff of Mechanical Engineering
Department.

S.ADITHYA. (B22ME004),

B.RAJESH. (B22ME008),

J.THARUN. (B22ME010),

P.SAIKIRAN. (B22ME005).

iv
Table of Contents:

S. No. Title Page no:

i Title Page i
ii Certificate ii
iii Declaration iii
iv Acknowledgement iv
v Table of Contents v
vi List of tables vi
vii List of figures vi
1 Introduction 1-2
2 Literature review 3-4
3 Objectives of the Mini project 5-6
4 Idea of proposed methodology 7-11
5 parameter optimization 12
6 Supervised and Unsupervised 13-14
7 Advantages and Disadvantages 15-16
8 Applications of AI/ML 17-18
9 Results and Discussion 19-26
10 Conclusion 27
11 References 28

v
List of Tables

S. No Title Page no.

1. Dataset 19

2. Results of GA 26

List of Figures

S. No Title Page no.

1. Edm process 2

2. Simple regression 8

3. Random Forest 9

4. GA flowchart 10

5. Supervised and 13
unsupervised models

6. AUC Scores 20

7. F1 Scores 21

8. Feature Importance 22

9. IP vs MRR 23

10. IP vs SR 23

11. MRR vs SR 25

12. Result of Model 26

vi
Chapter-1
INTRODUCTION
Electrical Discharge Machining (EDM) could be a non-traditional fabricating prepare broadly
utilized in businesses to machine complex shapes and difficult materials that are challenging to
prepare with routine strategies. EDM is interesting since it doesn't include coordinate contact
between the apparatus and work piece. Instep, it uses controlled electrical releases (sparkles)
between the cathode (apparatus) and the conductive work piece, which comes about in fabric
expulsion through softening and vaporizing. This handle empowers EDM to create complex
geometry, tight resistances, and tall surface wraps up on difficult materials like titanium,
tungsten carbide, and instrument steels, making it irreplaceable in businesses like aviation, car,
and making molds.

Significance of AI and ML in Overseeing Complex,hd Parameter Space:

The addition of Artificial Intelligence (AI) and Machine Learning (ML) has changed the way to
EDM parameter optimization. AI/ML procedures exceed expectations in overseeing high-
dimensional information and recognizing complex, nonlinear connections among parameters,
making them profoundly compelling for EDM prepare enhancement.

Data-Driven Choice Making: ML models can analyze authentic EDM information, extricating
designs and experiences to direct decision-making. This empowers more precise expectations of
the impacts of parameter changes on MRR, TWR, and surface wrap up.

Optimization Algorithms: ML-driven optimization calculations, such as Hereditary Calculations

(GA) and Bayesian Optimization, encourage effective investigation of the parameter space,
lessening the time and assets required for experimentation. These calculations can intellectuals
seek for parameter combinations that abdicate the leading comes about.

Versatile Control and Real-Time Observing: AI-powered frameworks can screen the EDM
prepare in real-time and make energetic alterations to parameters, progressing reaction times and
lessening fabric wastage. Versatile control utilizing ML models permits EDM machines to alter
parameters independently in reaction to real-time handle information, in this manner progressing
consistency and accuracy.

1
Prescient Support: With prescient support models, ML can expect apparatus wear and other
support needs, lessening downtime and expanding apparatus life. This minimizes interferences in
generation, assist progressing effectiveness.

Fig1:Edmprocessdiagram

2
Chapter-2
LITERATURE REVIEW:

2.1 Existing Methods

2.1.1 Statistical Models:

Factual models has been conventional method in EDM optimizing, utilizing exploratory plan
strategies to analyze the impacts of different arguments. Design of Experiments : DoE is
regularly utilized to methodically examine the connections among input factors and yield
reactions. Analysts use factorial plans and reaction surface techniques (RSM) to recognize ideal
parameter settings. For illustration, thinks about have utilized DoE to decide the finest
combinations of release current, beat term, and obligation cycle for maximizing fabric expulsion
rate (MRR) and minimizing surface harshness.

2.1.2 Machine Learning Models:

By assuming a linear connection between characteristics and their probability, LR a

popular classification technique models the likelihood of a binary result (offensive vs. non-
offensive). It can identify subtle patterns in textual data. It can also handle continuous and multi
col-linear data with ease. By estimating weights for every feature, logistic regression enables it
to assess the significance of each feature. To determine if a tweet is likely to be offensive, LR
analyses and interprets literary characteristics like word counts and scores with relation to the
detection of cyber bullying. Despite the effectiveness of logistic regression and effective, it
performs best when the data is well-prepared and balanced.

3
2.1.3 Decision trees & Regression model:

These methods analyze authentic information to anticipate machining results and distinguish
ideal parameter combinations.

Regression Models: Linear regression, polynomial regression, and numerous relapse methods are
commonly utilized to anticipate execution measurements like MRR and surface harshness based
on handle parameters. this models offer assistance set up connections between inputs & yields,
permitting for forecasts in unused cases.

Decision Trees and Random Forests: These gathering learning strategies are compelling for
classification & relapse assignments. Arbitrary woodlands, in specific, are vigorous upon over-
fitting & can handle hd information, making them appropriate for anticipating EDM execution
based on various arguments.

2.1.4 Optimization Algorithms:

Genetic Algorithms (GA): GAs are propelled by common choice & are utilized to discover ideal
arrangements by repetitive selecting, crossing all, and changing parameter combos. They are
viable in exploring complex optimization scenes and have been utilized to increase MRR &
surface wrap up in different thinks about.

Bayesian Optimization: This strategy is especially used for costly to research capacities, making
it perfect for optimizing EDM arguments where every assessment includes noteworthy time
& fetched. Bayesian optimization creates a probabilistic demonstrate of the aim work and
employments it to direct the look for ideal parameters.

4
Chapter-3

OBJECTIVES
Upgraded Understanding of EDM Forms: Give members with a complete usage of Electrical
Discharge Machining , counting its standards, uses, &centrality in present day fabricating.

Investigate the Part of AI and ML in Fabricating:Talk about the affect of manufactured insights
(AI) and machine learning innovations on the optimization of fabricating forms, with a particular
centering on EDM.

Recognize main EDM Parameters for Optimization: Highlight the basic handle parameters
influencing EDM execution, such as Metal Removal rate (MRR), surface roughness, and
apparatus wear, and show the significance of their optimization.

Display Data Collection & Preprocessing Strategies: Exhibit strategies for

searching preprocessing sensor information, guaranteeing high-quality inlet for AI/ML models,
and tending to challenges related to information judgment and unwavering quality.

Present Highlight Choice Techniques: Clarify different include choice methods to distinguish the
foremost powerful parameters affecting machining results, driving to more viable AI/ML
demonstrate improvement.

Illustrate Machine Learning uses in EDM: Show case considers and cases of how directed and
unsupervised learning models can anticipate and read machining execution, displaying their
viability comparing to conventional strategies.

Examine Optimization Calculations: Investigate optimization calculations such as Hereditary

Calculations & Bayesian Optimization, enumerating their application in productively
investigating parameter space and finding ideal EDM sets.

Highlight Real-Time Checking and Versatile Control: Clarify the centrality of real-time
observing frameworks and versatile control components in altering EDM parameters powerfully
amid machining, improving prepare versatility and proficiency.

Present Prescient Support Procedures: Talk about the part of AI/ML in prescient upkeep,
outlining how these techniques can reduce downtime and make strides the life span of EDM gear.

5
Encourage Information Share and Collaboration: Empower members to share experiences,
encounters, and challenge relating to EDM optimization, cultivating collaborating from industry
experts, analysts, and teachers.

By accomplishing these destinations, the class points to prepare members with the information
and abilities essential to use AI and ML innovations successfully in optimizing EDM forms,
eventually contributing to progressions in fabricating hones.

6
Chapter-4
METHODOLOGY
4.1. Information Collection:
Sensor Integration: Prepare the EDM machine with suitable sensors to gather important
information amid the machining handle. Key parameters to incorporate:

Current
Pulse on time
Voltage
Tool lift time
Wear rate
Surface roughness
Metal removal rate
Information Logging: Utilize information securing frameworks to log sensor readings and
machining parameters in real-time. Guarantee that the information is time-stamped for worldly
examination.
Information Differences: Collect information over distinctive materials and machining
conditions to form a comprehensive data set that reflects a wide run of EDM scenarios.

4.2. Information Preprocessing:

Information Cleaning: Expel any clamor or exceptions from the collected information.
Procedures such as sifting or factual investigation can be utilized to upgrade information quality.

Normalization: Normalize the information to guarantee that every parameters were on a

comparative level, which are especially vital for machine learning calculations that depend on
separate measurements

Division: Partition the information into preparing, approval, and test sets to encourage vigorous
show preparing and assessment.

7
4.3. Feature Choice:

Distinguish Basic Parameters: Utilize factual strategies (e.g., relationship investigation) and ML
procedures (e.g., recursive include disposal) to recognize the foremost critical highlights
influencing EDM execution results, such as Metal Removal rate (MRR) & surface
unpleasantness

Dimensionality Diminishment: Consider strategies such as Vital Component Investigation (PCA)

to decrease the dimensionality of the dataset whereas holding fundamental data, subsequently
making strides demonstrate execution and interpretability.

4.4. Model Development:

Select Machine Learning Calculations: Select fitting ML calculations based on the nature of the
information and the optimization destinations. Commonly utilized models incorporate:

Regression models (straight, polynomial): Regression, a statistical approach, dissects the

relationship between dependent and independent variables, enabling predictions through various
regression models.

Fig. 2: Simple regression

8
Decision trees and random forest: Decision Trees and Random Forests are both popular machine
learning algorithms used for classification and regression tasks. They are both based on the concept of
tree-like structures, where decisions are made by splitting data at different nodes.

Fig. 3: Decision tree and Random forest

Artificial Neural Networks (ANN): Demonstrate Preparing: Prepare the chosen models utilizing
the preparing dataset, utilizing cross-validation procedures to anticipate overfitting and guarantee
generalization.

Show Assessment: Assess show execution utilizing the approval dataset, utilizing measurements
such as Cruel Outright Mistake (MAE), Cruel Squared Mistake (MSE), and R-squared values for
relapse models. For classification assignments, measurements like precision, accuracy, and
review ought to be employed.

4.5 Optimization Framework:

Implement optimization calculations to hunt for the leading combo of EDM parameters.
This could incorporate:

9
Hereditary Calculations (GA): Utilize GA to repetively advance parameter sets from on
execution input, successfully exploring the parameter place.

Fig. 8: GA flowchart

Genetic algorithm process: GAs work by reenacting the method of common advancement,
where potential arrangements to an optimization issue are treated as people in a populace. The
calculation iteratively advances this populace through determination, hybrid, and change,
eventually focalizing towards an ideal arrangement. The most steps included in a GA.

Initialization: A arbitrary populace of potential arrangements (people) is produced, where each

person speaks to a special set of EDM parameters.

Assessment: Each person is surveyed employing a wellness work, which measures its execution
based on predefined criteria, such as maximizing the Metal Removal Rate (MRR) and
minimizing surface harshness or instrument wear

Choice: People are chosen based on their wellness scores to create a mating pool. Common
determination strategies incorporate roulette wheel choice and tournament determination

10
Hybrid: Sets of chosen people are combined to create sibling, presenting varieties within the
parmeters. This simulates the hereditary hybrid watched in nature

Change: Arbitrary changes are presented to a few people within the populace to preserve
hereditary differing qualities and anticipate untimely meeting.

Substitution: The unused era of people replaces a few or all of the past era, and the method
emphasizes through the assessment, choice, hybrid, and transformation steps until a halting basis
is met (e.g., a foreordained number of eras or a palatable wellness level).

4.6. Real Time Monitor and Versatile Control:

Observing Framework: Create a real-time information checking framework to track the

machining handle persistently. This framework ought to give criticism on the present machining
conditions and execution measurements.

Versatile Control Execution: Execute versatile control methodologies that permit for energetic
alterations of EDM parameters during the machining prepare based on real-time information. For
illustration, on the off chance that the framework identifies a decrease in surface quality, it can
naturally alter release current or beat term to preserve ideal execution.

Criticism Circle: Set up a criticism circle where real-time execution information is utilized to
refine the ML models persistently. This versatile approach guarantees that the models stay
pertinent and viable in change of conditions.

4.7. Predictive Support:

Support Modeling: Join prescient upkeep procedures based on verifiable execution information
and machine learning methods . Create calculations that foresee when maintenance is needed to
play down downtime and upgrade gear life span

Condition Checking: Utilize information collected amid machining to screen the condition of
devices and apparatus, anticipating wear and tear to plan support actively.

11
Chapter-5
PARAMETER OPTIMIZATION

Optimizing these basic EDM parameters were MRR, SR device carries a few noteworthy
benefits:

Moved forward Machining Proficiency: By fine-tuning parameters, producers can accomplish

higher MRR whereas keeping up the required surface roughness , coming about in speedier
generation cycles.

Improved Item Quality: Optimization makes a difference guarantee that the ultimate item meets
rigid quality benchmarks, which is particularly pivotal in businesses like aviation and restorative
gadgets where accuracy is fundamental.

Fetched Decrease: Successful optimization minimizes squander, diminishes the recurrence of

instrument changes, and brings down vitality utilization, driving to by and large taken a toll
investment funds in fabricating operations.

Expanded Competitiveness: Companies that use optimized EDM forms can react more rapidly to
showcase requests, advertising high-quality items at competitive costs.

Maintainability: Proficient EDM forms contribute to more feasible fabricating by lessening

fabric squander and vitality utilization.

In conclusion, the optimization of key EDM parameters are MRR, surface wrap up, and
apparatus wear were is fundamental for upgrading the in general execution of the machining
prepare. Understanding the connections between these parameters and the different variables
impacting them empowers producers to execute data-driven procedures that progress
productivity, diminish costs, and guarantee high-quality yields. By centering on these basic
zones, companies can accomplish noteworthy progressions in their EDM capabilities, driving
advancement and competitiveness within the fabricating scene.

12
Chapter-6
SUPERVISED AND UNSUPERVISED MODELS

6.1 Supervised Models in EDM Optimization

Administered learning may be a department of machine learning where models are trained on
labeled datasets, allowing them to memorize designs and connections within the information.
Within the setting of Electrical Release Machining (EDM), administered models can be utilized
to foresee key execution measurements based on different input parameters, empowering
optimization of the machining handle. This area investigates the sorts of directed models utilized
in EDM, their applications, and the benefits they offer.

In administered learning, the demonstrate is prepared employing a dataset that comprises of

input-output sets. The input highlights speak to the different EDM handle parameters (e.g.,
release current, beat length, obligation cycle), whereas the yield variable is ordinarily a execution
metric, such as Metal Removal Rate (MRR), surface harshness, or instrument wear.

Fig5:Simple Supervised and Unsupervised models

13
6.2 Unsupervised Models in EDM Optimization:
Unsupervised learning may be a sort of machine learning that bargains with datasets without
labeled data, permitting the model to find inherent structures and designs within the information.
Within the setting of Electrical Release Machining (EDM), unsupervised models can be utilized
to analyze handle information, recognize clusters, and optimize parameters. This area
investigates the application of unsupervised models in EDM, counting their techniques,
preferences, and down to earth applications.

1. Outline of Unsupervised Learning

Unsupervised learning calculations analyze information without earlier labeling, empowering the
recognizable proof of fundamental designs and structures. Common methods incorporate

Clustering: Bunches comparable information focuses based on certain highlights. Prevalent

clustering calculations incorporate:

K-Means Clustering: Segments information into K particular clusters based on highlight

closeness.

Progressive Clustering: Builds a pecking order of clusters by either consolidating or part existing
bunches.

14
Chapter-7
ADVANTAGES AND DISADVANTAGES:
7.1. Advantages:
Improved Productivity: AI/ML calculations can analyze expansive datasets rapidly and precisely,
empowering speedier decision-making and optimization of machining parameters. This leads to
progressed Fabric Expulsion Rates (MRR) and decreased machining times.

Prescient Support: AI and ML can analyze chronicled information to anticipate device wear and
machine disappointments, permitting for proactive support planning. This minimizes downtime
and increments by and large gear proficiency.

Upgraded Quality Control: Machine learning models can anticipate surface wrap up and other
quality measurements based on input parameters, permitting for alterations in genuine time. This
leads to increased quality yields and decreased scrap rates.

Data-Driven Choice Making: AI/ML gives producers with experiences determined from
information investigation, empowering educated decision-making instead of depending
exclusively on involvement or instinct.

Versatile Control Frameworks: AI frameworks can actualize real-time alterations to machining

parameters based on input from sensors, guaranteeing ideal execution all through the machining
handle.

Multi-Objective Optimization: AI/ML can at the same time optimize numerous goals, such as
increasing MRR whereas minimizing surface unpleasantness and instrument wear, driving to
more adjusted machining result

7.2. Disadvantages:
Data Reliance: The viability of AI/ML models depends intensely on the accessibility and quality
of information. Deficiently or poor-quality information can lead to wrong forecasts and
imperfect comes about.

Complexity of Usage: Coordination AI/ML into existing EDM forms can be complex and may
require noteworthy changes to workflows, staff preparing, and speculation in modern
innovations.

Dark Box Nature: Numerous AI/ML models, especially profound learning calculations, work as
"dark boxes," making it troublesome to get it how choices are made. This need of
straightforwardness can be a concern for administrators and engineers.

Starting Costs: Actualizing AI/ML innovations regularly requires critical forthright ventures in
program, equipment, and preparing, which can be a boundary for littler producers.

15
Ability Hole: There may be a deficiency of staff with the essential abilities to create and keep up
AI/ML frameworks, driving to challenges in usage and progressing back.

Over-fitting: AI/ML models can some of the time ended up as well custom fitted to the preparing
information, driving to destitute generalization on modern, inconspicuous information. This will
result in diminished execution in real-world applications. AI frameworks can execute real-time
alterations to machining parameters based on criticism from sensors, guaranteeing ideal
execution all through the machining handle.

Multi-Objective Optimization: AI/ML can at the same time optimize numerous targets, such as
maximizing MRR whereas minimizing surface unpleasantness and device wear, driving to more
adjusted machining result

16
Chapter-8
APPLICATIONS OF AI/ML IN EDM:
8.1 Handle Parameter Optimization:
Objective: To optimize machining parameters such as release current, beat term, and obligation
cycle.

Approach: ML calculations like relapse models and optimization methods (e.g., Hereditary
Calculations, Bayesian Optimization) analyze the connections between input parameters and
execution measurements (like Metal Removal Rate and surface wrap up).

The models can recommend ideal settings to maximize proficiency and minimize absconds.

8.2. Prescient Upkeep:

Objective: To anticipate device wear and machine disappointments some time recently they
happen.

Approach: AI models analyze chronicled information from machines, counting operational

parameters, instrument wear rates, and upkeep records.

Prescient analytic distinguish designs demonstrating potential disappointments, empowering

proactive upkeep planning, diminishing downtime, and expanding instrument life.

8.3. Quality Control and Assessment

Objective: To guarantee high-quality yields and decrease scrap rates.

Approach: ML calculations can be prepared to anticipate surface harshness and other quality
measurements based on input parameters.

Real-time checking frameworks can give input for prompt alterations, making a difference keep
up quality benchmarks all through the machining handle.

8.4. Inconsistency Discovery

Objective: To distinguish anomalous designs in machining information that will demonstrate

issues.

Approach: Unsupervised learning models, such as clustering and inconsistency location

calculations (e.g., DBSCAN), can be utilized to identify exceptions in operational information.

Early distinguishing proof of inconsistencies permits for opportune intercessions, decreasing the
chance of noteworthy disappointments.

8.5. Shrewdly Control Frameworks

Objective: To execute versatile control frameworks that optimize EDM forms in genuine time.

17
Approach: AI frameworks can utilize input from sensors to powerfully alter machining
parameters based on current conditions.

Support learning calculations can be connected to persistently make strides control procedures,
adjusting to changing operational situations.

.6. Data-Driven Choice Making

Objective: To use information for educated decision-making.

Approach:AI/ML models analyze large volumes of machining information to supply significant

bits of knowledge.

Choice back frameworks can offer assistance administrators and engineers optimize prepare
settings and troubleshoot issues based on data-driven proposals.

8.7. Reenactment and Virtual Testing

Objective: To mimic EDM forms for plan and optimization.

Approach: AI-based recreation devices can demonstrate EDM operations, permitting for virtual
testing of diverse parameters without physical trials.

This makes a difference in refining handle techniques and distinguishing ideal setups some time
recently real machining.

18
Chapter-9

RESULTS AND DISCUSSION

Table1 : Data Set.

Dialog: The capacity of AI/ML to analyze complex connections between parameters permits
producers to move absent from conventional trial-and-error strategies. By utilizing data-driven
19
approaches, they can accomplish speedier setup times and more dependable machining results,
driving to noteworthy fetched reserve funds and made strides competitiveness.

AUC Values for Different Models — The chart is comparing the AUC scores of various
models, which is a metric commonly used for evaluating binary classifiers based on their ability
to distinguish between classes.

Among these, Random Forest achieved the highest AUC score, approximately 0.98, indicating
excellent discriminative ability. In contrast, the K-Nearest Neighbors (KNN) model recorded the
lowest AUC, around 0.72, suggesting relatively weaker performance. The other models—Logistic
Regression, SVM, and Gradient Boosting—demonstrated moderate performance with AUC values in
the range of 0.75 to 0.78.

This comparison underscores the effectiveness of ensemble-based methods like Random Forest for
classification tasks, especially when high AUC is a priority.

20
F1 Score Comparison of Classification Models:

The bar chart titled "F1 Scores for Different Models" provides a comparative evaluation of five
classification algorithms based on their F1 scores. The F1 score is the harmonic mean of precision and
recall and is particularly useful when there is an uneven class distribution. It offers a balance between
false positives and false negatives, making it a crucial metric for assessing model performance in
binary classification tasks.

From the chart, it is evident that the Random Forest model achieved the highest F1 score,
approaching 0.97, indicating outstanding performance in terms of both precision and recall. Logistic
Regression also performed well, with an F1 score of approximately 0.83. In contrast, Gradient
Boosting yielded the lowest F1 score, around 0.49, suggesting lower effectiveness in balancing
precision and recall. KNN and SVM performed comparably, both with F1 scores around 0.67.

21
This analysis highlights the superior performance of Random Forest not only in AUC but also in F1
score, reaffirming its robustness as a classification model across multiple evaluation metrics

Feature Importance Using Random Forest:

The bar chart titled "Feature Importance (Random Forest)" displays the relative importance of
four features—TLT, TON, TOF, and IP—as determined by a Random Forest classifier. Feature
importance in this context refers to the contribution of each variable to the model's predictive
performance. Random Forest calculates this by measuring the average decrease in impurity (such as
Gini impurity or entropy) contributed by each feature across all trees in the ensemble.

The importance scores, normalized to sum to 1, are represented on the x-axis, while the corresponding
features are listed on the y-axis.

Based on the chart:

TLT is the most influential feature, with an importance value close to 0.29.

TON and TOF follow closely, each contributing significantly to the model with scores slightly above
0.25.

IP has the lowest importance, contributing approximately 0.17, suggesting it plays a lesser role in the
model's predictions compared to the other features.These insights can guide feature selection and
prioritization in model refinement and further data collection, emphasizing the predictive value of
TLT, TON, and TOF over IP.
22
23
Analysis of Input Current vs Metal Removal Rate and Surface Roughness

Input Current vs Metal Removal Rate (MRR):

The scatter plot titled "IP vs MRR" shows the relationship between Input Current (IP) and Metal
Removal Rate (MRR). The data points are distributed into three clusters corresponding to IP values of
9, 12, and 15 Amperes. Within each IP level, MRR varies, but a general trend can be observed: higher
input currents tend to produce higher maximum MRR values. For instance, the IP = 15 group exhibits
some of the highest MRR values in the data set. This suggests a positive correlation between input
current and metal removal rate, although the variability within each group implies that other process
parameters may also influence MRR.

Input Current vs Surface Roughness (SR):

The second plot, "IP vs SR", displays the relationship between input current and surface roughness.
Similar to the MRR plot, the data points are grouped by IP values. As the input current increases from
9 to 15 A, the surface roughness values also tend to rise. This indicates that higher currents may lead
to rougher surfaces. The trend suggests a trade-off between achieving higher MRR and maintaining
surface quality, as increased IP may improve material removal but degrade surface finish.

The plots demonstrate that increasing input current (IP) generally leads to higher metal removal rates
(MRR) but also results in increased surface roughness (SR). This trade-off is important when
optimizing machining parameters, as higher productivity (MRR) may come at the cost of lower
surface quality (higher SR). Further statistical analysis is recommended to confirm the observed
trends and quantify the strength of these relationships.

24
Pareto Analysis: Surface Roughness vs Material Removal Rate;

The Pareto front plot illustrates the trade-off between Surface Roughness (SR) and Material Removal
Rate (MRR). It can be observed that as SR increases, MRR also tends to increase. This indicates that
higher material removal rates are associated with rougher surface finishes.

This relationship is crucial in machining processes, where optimizing both MRR and SR is often
conflicting. While high MRR improves productivity, it usually comes at the cost of degraded surface
quality. Conversely, achieving a smoother surface generally requires slower machining and, hence,
lower MRR.

Conclusion

The Pareto front represents the set of optimal solutions where no objective (SR or MRR) can be
improved without worsening the other. This visualization supports decision-making by helping to
25
choose appropriate process parameters that balance productivity (MRR) and quality (SR) based on
specific application needs.

This is the Final output of the predictive model and optimal process parameters which are
obtained by applying random forest and integrating genetic algorithm with it.

This result demonstrates the model's capability to estimate outputs based on varying process
conditions, enabling informed decisions for parameter selection depending on the desired
balance between productivity and surface finish.

These values reflect a high material removal rate, though with a corresponding increase in
surface roughness, which aligns with the trade-off seen in the Pareto analysis.

26
10.Conclusion:
This research explored the application of Artificial Intelligence and optimization techniques to
enhance the performance of Electric Discharge Machining (EDM). The core objective was to optimize
two critical output responses—Metal Removal Rate (MRR) and Surface Roughness (Ra)—by
intelligently tuning process parameters such as current, voltage, pulse-on time, and pulse-off time.

A Random Forest regression model was developed to capture the complex, nonlinear relationships
between input variables and output responses. The model demonstrated high predictive accuracy,
confirming its suitability for modeling the EDM process. Random Forest was chosen for its
robustness to overfitting and its ability to handle interactions among variables without requiring
extensive parameter tuning.

To identify optimal parameter combinations, a Genetic Algorithm (GA) was implemented using the
trained Random Forest model as a surrogate for the actual machining process. This approach enabled
efficient optimization without the need for excessive experimental trials. The GA effectively explored
the solution space and identified parameter settings that led to improved MRR while maintaining
acceptable surface roughness levels.

The combination of machine learning and evolutionary optimization proved to be a powerful strategy
for process improvement. The results suggest that this hybrid approach can significantly reduce trial-
and-error in EDM process setup and can lead to better performance outcomes. Furthermore, the
methodology offers flexibility and can be extended to other machining operations or materials with
appropriate adjustments.

In summary, this study demonstrated that integrating data-driven modeling with optimization
algorithms can offer a practical and efficient solution for machining parameter selection. Future work
may involve experimental validation of the optimized conditions, incorporation of additional input
variables, or comparison with alternative modeling and optimization techniques to further enhance
process efficiency and reliability.

27
REFERENCES:

M. B. V. S. V. Prasad, et al. (2020). "Application of Machine Learning Techniques forElectrical

Discharge Machining: A Review." Journal of Manufacturing Processes, 57, 73-85. DOI:
10.1016/j.jmapro.2020.05.028.

K. Jain, et al. (2019). "Optimization of EDM Parameters Using Artificial IntelligenceTechniques:

A Review." Materials Today: Proceedings, 18, 1281-1250. DOI:10.1016/j.matpr.2019.06.263.

S. Kumar, et al. (2021). "Predictive Modeling of EDM Process Parameters Using Machine
Learning." International Journal of Advanced Manufacturing Technology, 118, 1225-1235. DOI:
10.1007/s00170-021-07067-y.

H. S. D. M. A. Abdurrahman, et al. (2021). "Artificial Intelligence and Machine Learning in

Electrical Discharge Machining: A Review." Journal of Cleaner Production, 299, 126827.
DOI:10.1016/j.jclepro.2021.126827.

28