<p>The burnishing process is a simple, low-cost superfinishing technique that enhances surface integrity of mechanical components by reducing mean roughness and increasing hardness, while improving properties such as wear resistance, corrosion resistance, and fatigue strength. This study presents a comparative analysis of three artificial intelligence models: artificial neural networks (ANN), random forests (RF), and support vector regression (SVR) to predict multiple surface integrity responses after ball burnishing. Experimental data from literature were used: for aluminum alloy 6061-T6, models predict fatigue life, maximum residual stress, and diameter change based on speed, force, and feed; for AISI 4340 steel, predictions cover surface roughness, hardness, and roundness error considering speed, feed, force, and number of passes. Models were trained with varying data distributions and input exclusions for sensitivity analysis, evaluated using metrics like coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">$R^2 > 0.97$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mi>R</mi> <mn>2</mn> </msup> <mo>></mo> <mn>0.97</mn> </mrow> </math></EquationSource> </InlineEquation> in optimal cases), Pearson correlation (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">$r > 0.97$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>r</mi> <mo>></mo> <mn>0.97</mn> </mrow> </math></EquationSource> </InlineEquation>), relative percent deviation (RPD <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$< 10~\%$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo><</mo> <mn>10</mn> <mspace width="3.33333pt" /> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation>), and root-mean-square error (RMSE). Results demonstrate high predictive accuracy across models with complete inputs, with ANN excelling in capturing complex nonlinear relationships, RF offering robustness against overfitting, and SVR providing efficient generalization. Key findings highlight influential parameters: force on residual stress, feed on fatigue life and roughness, and number of passes on hardness. This work serves as a reference for selecting AI models to optimize burnishing parameters and enhance manufacturing outcomes.</p>

Artificial intelligence models to predict surface integrity after the burnishing process

期刊论文
发表日期：

2025年12月11日

查看全文

Víctor Alfonso Alcántar-Camarena,
Luis Enrique Raya-González,
Gustavo Capilla-González,
David Alejandro Pérez-Márquez,
Antonio de Jesús Balvantín-García,
Alberto Saldaña-Robles

摘要

The burnishing process is a simple, low-cost superfinishing technique that enhances surface integrity of mechanical components by reducing mean roughness and increasing hardness, while improving properties such as wear resistance, corrosion resistance, and fatigue strength. This study presents a comparative analysis of three artificial intelligence models: artificial neural networks (ANN), random forests (RF), and support vector regression (SVR) to predict multiple surface integrity responses after ball burnishing. Experimental data from literature were used: for aluminum alloy 6061-T6, models predict fatigue life, maximum residual stress, and diameter change based on speed, force, and feed; for AISI 4340 steel, predictions cover surface roughness, hardness, and roundness error considering speed, feed, force, and number of passes. Models were trained with varying data distributions and input exclusions for sensitivity analysis, evaluated using metrics like coefficient of determination ( $R^2 > 0.97$ $R^{2} > 0.97$ in optimal cases), Pearson correlation ( $r > 0.97$ $r > 0.97$ ), relative percent deviation (RPD $< 10~\%$ $< 10 %$ ), and root-mean-square error (RMSE). Results demonstrate high predictive accuracy across models with complete inputs, with ANN excelling in capturing complex nonlinear relationships, RF offering robustness against overfitting, and SVR providing efficient generalization. Key findings highlight influential parameters: force on residual stress, feed on fatigue life and roughness, and number of passes on hardness. This work serves as a reference for selecting AI models to optimize burnishing parameters and enhance manufacturing outcomes.