INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Machine Learning Based Surface Roughness Prediction for

Parameters of ECM

Sharanya Kalkunte¹, Ritish Hullar¹, S Divyashree¹, Surabhi Satish¹, Gajanan M Naik²

¹Department of Computer Science and Engineering, RV Institute of Technology and Management,

Bengaluru, India

²Department of Mechanical Engineering, RV Institute of Technology and Management, Bengaluru,

India

DOI: https://dx.doi.org/10.51244/IJRSI.2025.1210000360

Received: 02 November 2025; Accepted: 10 November 2025; Published: 24 November 2025

ABSTRACT

Electrochemical Machining (ECM) is a machining technique which is non traditional used for shaping complex

components with superior accuracy and surface finish. However, optimizing surface roughness remains

challenging because of the intricate, non-linear dependency between various process aspects such as electrolyte

concentration, voltage, frequency, duty cycle, temperature, and feed rate. Traditional trial-and-error or analytical

approaches are often time- consuming and inefficient. This study introduces a Machine Learning (ML)-based

predictive modeling approach to estimate and optimize the roughness of the surface in ECM processes using

data obtained by Chen Xuezhen et al.’s tests on the Ti60 titanium alloy.

Keywords— Electrochemical Machining, Surface Roughness, Machine Learning, Process Parameters,

Predictive Modeling

INTRODUCTION

Electrochemical Machining (ECM) is a machining technique which is non traditional used for producing

complex and high-precision components that are difficult to manufacture using conventional methods. It operates

on the principle of anodic dissolution, where the workpiece (anode) is eroded under a controlled electric current

in the presence of an electrolyte, while the tool (cathode) remains unaffected. This enables the machining of

extremely hard materials like titanium alloys, superalloys, and stainless steels without inducing thermal or

mechanical stresses. Industries such as aerospace, biomedical, defense, and automotive rely heavily on ECM

because of its skill to produce intricate geometries with excellent surface finish and dimensional accuracy. These

components were constantly observed and were noticed to be reacting in non-linear ways: for example,

increasing voltage enhances the rate of dissolution but may also lead to over-etching, while electrolyte

concentration and duty cycle affect ion mobility and effective machining time. Traditionally, optimization of

ECM parameters has been achieved by empirical experiments or statistical techniques like Design of

Experiments (DOE) and Response Surface Methodology (RSM). While these methods provide insights, they

often fail to capture the complex, multi-dimensional relationships inherent in ECM and require extensive

experimentation. This has led to growing interest in Machine Learning (ML)-based approaches that can identify

hidden patterns, predict outcomes, and optimize processes efficiently [1]. Machine Learning offers a data-driven

way to model and optimize

ECM by learning from past experimental data and identifying non-linear relationships among parameters. Unlike

traditional modeling, ML does not need explicit equations depicting the process physics. Instead, it predicts

machining outcomes—such as surface roughness or material removal rate—with high accuracy, enabling virtual

experimentation and real-time optimization. In this project, an ML- based predicting model has been developed

to estimate and optimize the surface rougghness of ECM workpieces using experimental data from Chen

Xuezhen et al., involving the machining of Ti60 titanium alloy—a material widely used in aerospace

applications. The dataset includes a range of parameter combinations such as electrolyte concentration, voltage,

Page 4202

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

frequency, duty cycle, temperature, and feed rate, making it ideal for model training and validation. Two

regression models—Linear Regression and Polynomial Regression—are implemented using Python’s scikit-

learn library[2]. The Linear model captures basic additive trends, while the Polynomial model represents non-

linear interactions more effectively. The data undergoes preprocessing steps such as outlier removal,

normalization, and consistency checks, followed by train-test splitting to ensure robust evaluation and prevent

overfitting. Exploratory Data Analysis (EDA) using visual tools like scatter plots and correlation heatmaps

reveals that voltage and feed rate strongly influence surface roughness, while duty cycle and temperature have

moderate effects. These insights align with theoretical expectations and validate the dataset’s reliability. Model

performance is evaluated using metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE),

and R². Results show that the Polynomial Regression model performs significantly better, reducing RMSE by

about 33% and achieving higher R² values. Cross-validation further confirms the model’s robustness and

generalizability [3]. An interactive visualization dashboard built using Matplotlib and Plotly allows users to input

process parameters and instantly visualize predicted surface roughness values with confidence intervals. This

serves as a virtual experimentation tool, helping engineers make informed decisions without conducting costly

physical trials. The integration of Machine Learning in ECM process optimization represents a shift from

traditional trial-based approaches to intelligent, data-driven manufacturing. Predictive modeling not only

minimizes time and cost but also enhances precision and efficiency. The developed framework can be extended

to real-time control systems, adaptive tuning, and hybrid approaches combining physics-based and data-driven

methods.In conclusion, this study demonstrates the potential of ML in transforming ECM into a predictive,

optimized, and efficient process. The developed models enable accurate surface roughness prediction and form

a foundation for intelligent process automation in modern electrochemical manufacturing, aligning with the goals

of Industry 4.0.

LITERATURE REVIEW

Electrochemical Machining (ECM) has been the focus of extensive research due to its ability to machine hard

alloys with high accuracy and minimal tool wear. Traditional research has primarily concentrated on

understanding the electrochemical dissolution mechanisms and optimizing machining parameters using design-

based statistical frameworks. Early studies employed Response Surface Methodology (RSM) and Taguchi

analysis to model relationships between parameters such as voltage, electrolyte concentration, and feed rate;

however, these linear statistical methods often failed to capture the non-linear and interactive nature of ECM

processes. Recent advancements in machine learning have expanded the scope of ECM research. Wu et al. [7]

demonstrated that Support Vector Regression improved prediction accuracy for ECM profile shape compared to

analytical models, especially when machining complex turbine blade geometries. Bahiuddin et al. [1] applied

Random Forests and Gradient Boosting to forecast surface roughness and reported a significant reduction in

prediction error compared to RSM. Similarly, Shang et al. [4] developed an Extreme Learning Machine approach

for ultra-precision milling and emphasized that hybrid data fusion significantly enhances model stability. In the

domain of unconventional machining, machine learning has shown strong generalization capabilities. Qasem et

al. [2] used artificial neural networks for EDM roughness prediction and highlighted the superiority of nonlinear

ML models over conventional polynomial fitting. Batu et al. [3] extended AI-driven roughness prediction to

additive manufacturing, demonstrating that ML models effectively capture microstructural irregularities that

conventional models overlook. Within ECM specifically, Rajesh et al. [9] attempted roughness prediction using

ANN for dry turning, though the paper was later retracted due to data inconsistencies. Nevertheless, their study

highlighted the increasing dependence on AI-driven models for machining applications. Recent comprehensive

reviews by Ko et al. [10] and Yang et al. [14] emphasize a shift toward integrating ML with physics-based

simulations and Digital Twin systems, allowing real-time prediction and closed-loop control of machining

operations. While various ML techniques have been explored for ECM, many studies suffer from small datasets,

lack of cross-validation, or limited parameter ranges. Furthermore, only a few works incorporate visualization

tools or decision-support systems into the predictive framework. Thus, there remains a compelling need for a

reliable, interpretable, and user-friendly predictive model for ECM roughness—motivating the development of

the present study.

Page 4203

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

METHODOLOGY

This study focuses on developing predictive regression models to estimate surface roughness in Electrochemical

Machining (ECM) processes using experimentally obtained data. The research utilizes the dataset provided by

Chen Xuezhen et al., based on controlled machining tests of Ti60 titanium alloy, a material commonly employed

in aerospace blisks due to its strength and temperature resistance. The experiments were conducted with

variations in six key process parameters—electrolyte concentration, applied voltage, pulse frequency, duty cycle,

electrolyte temperature, and tool feed rate—each known to significantly influence the resulting surface finish.

Surface roughness was measured through profilometry at multiple locations on each machined sample to ensure

accuracy and account for spatial variations. The dataset, which encompasses a broad range of parameter

combinations under well-controlled conditions, provides a reliable and comprehensive foundation for

regression-based analysis. The availability of such a validated dataset eliminates the need for further

experimentation while ensuring the consistency and reproducibility required for model development. The

modeling framework involves the implementation of two regression techniques—Linear Regression and

Polynomial Regression—using Python’s scikit-learn library. The Linear Regression model assumes a

straightforward, additive relationship between machining parameters and surface roughness, while the

Polynomial Regression model captures the non-linear dependencies often observed in ECM processes. Prior to

model training, the dataset undergoes essential preprocessing steps including outlier removal, normalization, and

data consistency checks to ensure the integrity and uniformity of input features. The dataset is subsequently

divided into training (70%), validation (15%), and testing (15%) subsets to ensure unbiased evaluation and

prevent overfitting. Model performance is assessed using standard statistical metrics such as Mean Absolute

Error (MAE), Root Mean Square Error (RMSE), and the Coefficient of Determination (R²). Cross-validation

techniques are also employed to verify the robustness and reliability of the models across different data

partitions. This structured methodology ensures that the developed models are both accurate and generalizable,

providing valuable insights into the complex interrelationships governing surface roughness in ECM.

RESULTS AND DISCUSSION

Exploratory Data Analysis (EDA) revealed that surface roughness exhibited a distinctly non-linear dependence

on various process parameters, with voltage and feed rate emerging as the most influential factors. As seen in

Figure 1 (Correlation Heatmap), strong positive correlations were observed between voltage and surface

roughness (r ≈ +0.72) and between feed rate and roughness (r ≈ +0.64), confirming their dominant influence on

machining performance. The scatter plots in Figure 2 further illustrate these trends—higher voltages increased

the material removal rate but also led to localized over-etching, raising surface roughness beyond the desired

threshold. Variations in duty cycle and frequency revealed that increasing the duty cycle initially enhanced the

surface finish up to an optimal point, after which excessive anodic dissolution slightly degraded it. Similarly,

electrolyte concentration displayed a non-linear behavior, where intermediate values produced smoother

surfaces, while higher concentrations caused turbulence and irregular ion movement. Overall, the visualizations

clearly demonstrate the complex interplay among the parameters, validating the need for non-linear modeling

approaches in predicting surface quality in electrochemical machining.

Fig 1. Heatmap

Page 4204

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Feature Importance and Sensitivity Analysis

Understanding how each machining parameter influences surface roughness is essential for interpreting model

predictions and improving process control in Electrochemical Machining (ECM). After training the regression

models, a feature importance and sensitivity analysis was performed to identify which input variables contributed

most significantly to the predicted roughness values. This analysis not only validates the model’s learning

behaviour but also provides practical engineering insights into the ECM process. Since Linear Regression and

Polynomial Regression do not inherently produce feature importance scores like tree-based models, a

combination of statistical coefficient analysis, correlation studies, and controlled sensitivity testing was used.

For each parameter—electrolyte concentration, voltage, frequency, duty cycle, temperature, and feed rate—the

model’s response was observed while systematically varying one parameter at a time and keeping others

constant. This approach helps isolate the direct effect of each factor on the predicted surface roughness. Across

both models, voltage and feed rate emerged as the dominant parameters. Voltage showed the strongest influence,

where increases beyond a certain threshold led to rapid roughness escalation due to intensified anodic

dissolution. Feed rate exhibited a similar pattern—higher feed rates resulted in insufficient time for uniform

material removal, producing rougher surfaces. The polynomial model captured these non-linear behaviours more

clearly, showing that small variations in voltage and feed rate could cause disproportionately large changes in

roughness. The duty cycle and electrolyte temperature demonstrated secondary but still meaningful influence.

Higher duty cycles improved surface finish up to an optimum value by increasing effective machining time, after

which excessive current exposure slightly degraded the surface. Temperature affected ion mobility and

electrolyte conductivity; moderate temperatures contributed to smoother surfaces, while overly high

temperatures increased turbulence and disturbed the dissolution layer. Electrolyte concentration and frequency

showed the least direct impact compared to other parameters, although their effects became more noticeable

when combined with voltage variations. The polynomial model revealed interaction terms—such as voltage ×

concentration—that affected roughness more strongly than either variable alone, highlighting the importance of

capturing parameter interactions in ECM. Overall, the sensitivity analysis confirms that surface roughness in

ECM is governed by a combination of strong primary factors (voltage, feed rate) and supporting secondary

parameters (duty cycle, temperature), along with subtle but relevant interaction effects. These findings closely

align with theoretical machining principles, reinforcing the reliability of the trained models. The insights from

this analysis can guide engineers in prioritizing which parameters to control most tightly during machining and

can assist in designing future optimization strategies for achieving consistent and smooth surface finishes.

Model Testing and Validation

Both of the models were trained using 70% of the available data and validated on the remaining 30%. The Linear

Regression model, which captured only additive parameter effects, achieved reasonable precision but struggled

to account for complex, non-linear interactions among the process parameters. In contrast, the Polynomial

Regression model (with degree = 2) offered a more flexible fit, allowing it to adapt to intricate patterns in the

data. This model demonstrated approximately a 33% reduction in RMSE and a notably higher R² value, reflecting

its stronger predictive capability and its ability to model the non-linear relationships present in the

Electrochemical Machining (ECM) process. The findings of this study confirm that surface roughness of the

produced material can be accurately predicted using machine learning methods. Feed rate and voltage emerged

as the most influential predictors, consistent with the established theoretical understanding of the machining

process. Temperature and duty cycle were identified as secondary factors that affect the rate of electrochemical

dissolution and the conductivity of the electrolyte. Through these insights, the developed model functions as a

virtual surface roughness predictor, enabling engineers to anticipate machining outcomes even before conducting

physical experiments. To enhance usability, a graphical dashboard was created using Matplotlib and Plotly,

allowing users to interactively input ECM parameters and visualize the estimated surface roughness values along

with their respective confidence intervals. This integration of predictive modeling with interactive visualization

provides a practical decision-support tool for process engineers.

Page 4205

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

Fig 2. Line graph

Future Scope

Although the developed machine learning models provide reliable predictions for surface roughness in ECM,

there is considerable room to advance this work both technically and practically. One of the most promising

directions is the expansion of the dataset. The current model is trained on controlled experimental data, but

incorporating real-time machining data from industry would expose the model to wider variations in behaviour,

material inconsistencies, and environmental influences. This would help the system generalize better and

perform accurately under production-level conditions. Future studies can also explore additional process

parameters that were not included in this work but have known influence on ECM dynamics—for example,

electrolyte flow rate, tool geometry, inter-electrode gap variation, and gas bubble formation during machining.

Capturing these aspects would allow the predictive model to represent the electrochemical process more closely

and improve its robustness. On the modelling front, more advanced algorithms can be investigated. Techniques

such as Random Forests, Gradient Boosting, Support Vector Regression, and Deep Learning may offer improved

performance when dealing with highly non-linear relationships. There is also growing interest in hybrid

approaches that combine physics-based simulations with machine learning, allowing models to retain

interpretability while still benefiting from data-driven accuracy. A significant future opportunity lies in real-time

integration. By connecting the prediction model to live sensor data during machining, the ECM system can

evolve into a self-monitoring and self-adjusting process. This would enable automatic parameter tuning

whenever surface quality begins to deviate from the desired range, reducing manual intervention and minimizing

wasted material. Such a system is a natural step toward digital twin platforms, where virtual models continuously

mirror and adjust physical machining processes. Lastly, the interactive dashboard developed in this study can be

expanded into a complete decision-support tool. Features such as automated parameter recommendations, multi-

objective optimization (surface roughness, material removal rate, tool life), and uncertainty estimation would

make the system more practical for industry use. As ECM continues to play a critical role in aerospace,

biomedical, and precision manufacturing, intelligent predictive tools like this have the potential to modernize

machining workflows and significantly reduce cost, time, and trial-based experimentation.

CONCLUSION

The model’s robustness was further verified through five-fold cross-validation, which indicated uniform

performance across all folds with minimal variation (standard deviation of R² ≈ 0.02). Residual analysis showed

that the errors were randomly scattered, thereby satisfying model assumptions and confirming the model’s

generalisation ability. Overall, the working of this machine learning-based approach demonstrates a data-driven

optimization pathway for ECM processes, resulting in fewer experimental trials, reduced material loss, and

enhanced process efficiency.

Page 4206

INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)

ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025

REFERENCES

1. Bahiuddin, I. et al. (2025). Machine learning-based surface roughness prediction in ECM. Journal of

Manufacturing Processes, 68, 1054-1062.

2. Qasem, I. et al. (2025). Machine Learning-Based Prediction of EDM Material Removal Rate and Surface

Roughness. Applied Sciences, 9(8), 274.

3. Batu, T. et al. (2023). Application of Artificial Intelligence for Surface Roughness Prediction of

Additively Manufactured Components. Materials, 16(18), 6266.

4. Shang, S. et al. (2023). Surface Roughness Prediction in Ultra-Precision Milling: An Extreme Learning

Machine Method with Data Fusion. Micromachines, 14(11), 2016.

5. Ge, Z. et al. (2024). Machine Learning-Driven Optimization of Micro-Textured Surfaces in Ultrasonic

Vibration-Assisted Milling. Coatings, 14(12), 1539.

6. Shang, S. et al. (2024). Roughness prediction of end milling surface for digital twin applications. Digital

Twin, 3(4), 100-110.

7. Wu, M. et al. (2022). Profile prediction in ECM using machine learning. Lirias, KULeuven.

8. Antosz, K. et al. (2025). Application of Machine Learning to the Prediction of Surface Roughness in

Milling Operations. Journal of Manufacturing Processes, 75, 123-134.

9. Rajesh, A. S. et al. (2022). [Retracted] Machine Learning Approach: Prediction of Surface Roughness in

Dry Turning of Inconel 625. Journal of Materials Engineering and Performance, 31(9), 6038804.

10. Ko, J. H. et al. (2025). A review of artificial intelligence application for machining surface quality

prediction: from key factors to model development. Journal of Intelligent Manufacturing, 36(1), 1-22.

11. Sharma, V. (2025). Machine learning-based prediction and optimization of surface roughness in FDM.

Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture,

239(5), 1075-1087.

12. Shang, S. et al. (2025). Machine learning-based predictive modeling for surface roughness in precision

machining. Journal of Manufacturing Science and Engineering, 147(4), 041004.

13. Cebi, A. et al. (2024). Implementation of robotic electrochemical machining in aerospace applications.

Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture,

238(1), 123-135.

14. Yang, H. et al. (2024). A review of artificial intelligent methods for machined surface roughness

prediction. Journal of Materials Processing Technology, 300, 117421.

15. Kulkarni, T. et al. (2024). Machine learning-based approach for automated classification of ECM and

cell subtypes. Journal of Manufacturing Science and Engineering, 146(6), 061003.

Page 4207