Application of Logistic Regression for Identification of Dross Formation Conditions in CO2 Laser Cutting

Dross formation is an issue of multifold importance in laser cutting. In this study, based on data from two crossed experimental designs with 52 trials, a machine learning model, that is, multiple binary logistic regression, was developed to identify dross formation conditions in CO2 laser cutting. The model inputs were laser power, cutting speed, and oxygen pressure, and the corresponding dependent variable took the value of 1 (in the case of dross formation) or 0 (in the case of no dross formation). Statistical indicators showed a good predictive ability of the developed logistic regression model to discriminate between dross formation and no dross formation classes. It was revealed that the most decisive input regarding dross formation is the cutting speed. Upon the development of the logistic regression model, laser cutting process windows were created, and further analysis and interpretation of the experimental results were performed.