surface roughness and microhardness prediction: a machine learning approach for electrochemical grinding

This study focuses on the development of predictive machine learning models to estimate key surface integrity parameters in the electrochemical grinding (ecg) of aisi 304 stainless steel. experimental data were collected from a series of 20 controlled tests based on a response surface methodology (rsm) design, varying three primary process parameters: voltage, electrolyte concentration, and grinding wheel speed. using this dataset, gaussian process regression (gpr) models were constructed for four output variables: current density, surface roughness in x- and y-directions (ra_x and ra_y), and surface microhardness (vickers). model performance was evaluated using r² scores, residual analysis, and error distributions across both training and test datasets. the results demonstrate that surface roughness parameters, particularly ra_y (r²_test = 0.970) and ra_x (r²_test = 0.932), were predicted with the highest accuracy and consistency. current density also exhibited strong performance (r²_test = 0.954), though with minor deviations at extreme values. surface microhardness, in contrast, posed greater modeling challenges, achieving the lowest test r² (0.843) and showing systematic underprediction. residual and error analyses confirmed these trends, with minimal bias and variance for ra_x and ra_y, and broader, asymmetric error profiles for hardness. the least roughness was observed under an electrolyte concentration of 140 g/l, an applied voltage of 20 v, and a grinding wheel rotational speed of 2000 rpm. overall, the gpr models proved effective for capturing ecg process behavior and offer potential for process optimization in precision manufacturing this study focuses on the development of predictive machine learning models to estimate key surface integrity parameters in the electrochemical grinding (ecg) of aisi 304 stainless steel. experimental data were collected from a series of 20 controlled tests based on a response surface methodology (rsm) design, varying three primary process parameters: voltage, electrolyte concentration, and grinding wheel speed. using this dataset, gaussian process regression (gpr) models were constructed for four output variables: current density, surface roughness in x- and y-directions (ra_x and ra_y), and surface microhardness (vickers). model performance was evaluated using r² scores, residual analysis, and error distributions across both training and test datasets. the results demonstrate that surface roughness parameters, particularly ra_y (r²_test = 0.970) and ra_x (r²_test = 0.932), were predicted with the highest accuracy and consistency. current density also exhibited strong performance (r²_test = 0.954), though with minor deviations at extreme values. surface microhardness, in contrast, posed greater modeling challenges, achieving the lowest test r² (0.843) and showing systematic underprediction. residual and error analyses confirmed these trends, with minimal bias and variance for ra_x and ra_y, and broader, asymmetric error profiles for hardness. the least roughness was observed under an electrolyte concentration of 140 g/l, an applied voltage of 20 v, and a grinding wheel rotational speed of 2000 rpm. overall, the gpr models proved effective for capturing ecg process behavior and offer potential for process optimization in precision manufacturing.