Kerf characteristics during CO2 laser cutting of polymeric materials: Experimental investigation and machine learning-based prediction

Abstract

This study uses advanced machine learning approaches to predict the kerf open deviation (KOD) when a CO2 laser is used to cut polymeric materials. Four polymeric materials, namely polyethylene (PE), polymethyl methacrylate (PMMA), polypropylene (PP), and polyvinyl chloride (PVC), were cut under the same conditions. The process control factors were the power of the laser beam (80–140 W) and cutting speed (1–6 mm/s), while sheet thickness, standoff distance, and gas pressure were kept constant during experiments. KOD between the upper and lower opens of the kerf was the process response. KOD was predicted using three machine learning models, namely a conventional artificial neural network (ANN), a hybrid neural network–humpback whale optimizer (HWO-ANN), and a hybrid neural network–particle swarm optimizer (PSO-ANN). Experimental data for all polymeric materials were employed to train and test all models. The hybrid neural network–humpback whale optimizer model outperformed other models to predict KOD for all cut materials. The root-mean-square error between predicted and experimental data was 0.349–0.627 µm, 0.085–0.242 µm, and 0.023–0.079 µm for conventional neural network, neural network–particle swarm model, and neural network–humpback whale model, respectively.

Keywords

• Laser cutting
• Polymeric materials
• Kerf open deviation
• Humpback whale optimizer
• Machine learning