Kerf Geometry and Surface Roughness Optimization in CO₂ Laser Processing of FFF Plates Utilizing Neural Networks and Genetic Algorithms Approaches

This work deals with the experimental investigation and multi-objective optimization of mean kerf angle (A) and mean surface roughness (Ra) in laser cutting (LC) fused filament fabrication (FFF) 3D-printed (3DP), 4 mm-thick polylactic acid (PLA) plates by considering laser feed (F) and power (P) as the independent control parameters. A CO₂ laser apparatus was employed to conduct machining experiments on 27 rectangular workpieces. An experimental design approach was adopted to establish the runs according to full-combinatorial design with three repetitions, resulting in 27 independent experiments. A customized response surface experiment was formulated to proceed with regression equations to predict the responses and examine the solution domain continuously. After examining the impact of F and P on mean A and mean Ra, two reliable prediction models were generated to model the process. Furthermore, since LC is a highly intricate, non-conventional machining process and its control variables affect the responses in a nonlinear manner, A and Ra were also predicted using an artificial neural network (NN), while its resulting performance was compared to the predictive regression models. Finally, the regression models served as objective functions for optimizing the responses with an intelligent algorithm adopted from the literature.