Statistical analysis and machine learning-based modelling of kerf width in CO2 laser cutting of PMMA

Abstract

Engineering polymers like PMMA have increasingly replaced traditional materials where feasible, with CO2 laser cutting gaining attention for its high quality and speed in processing these materials. Precise cuts are essential for product accuracy, with kerf width being a key quality attribute for ensuring the final product's quality. This study focuses on the impact of three process variables: stand-off distance, laser power, and cutting speed, on the kerf width in CO2 laser cutting of PMMA. A full-factorial experiment systematically varies process parameters to understand their individual and interaction effects on the cutting process. The kerf width is measured as an indicator of precision to evaluate the quality of the laser cuts. In order to address the non-linear relationships between these process parameters and kerf width, several machine learning models were utilized. Performance comparisons indicated that the Artificial Neural Network (ANN) model provided the highest accuracy, with R² values of 0.98 for the validation dataset and 0.95 for the testing dataset. The optimized ANN model is a robust tool for parameter optimization, determining optimal settings to achieve the desired kerf width and ensure productivity.